Search

In this paper, a novel indoor localization system with channel state information (CSI) fingerprints is proposed, which learns the spatial and frequency features of CSI in the fifth-generation (5G) cellular network by a Siamese convolution neural network. In particular, considering that the CSI continuously collected by a moving target possesses the implicit spatial association, we locate the target by the successive CSI data gathered within a time interval which can be regarded as an information subspace of the fingerprint database. Therefore, the fingerprint localization can be modeled as a subspace matching problem and solved by the Siamese network-based similarity learning. In the proposed system, we design a structure of CSI fingerprint which includes the information from multiple base stations in spatial and frequency domains. Then, the proposed Siamese architecture extracts the CSI feature and estimates the location of the target by feature similarity comparison. Compared with the existing algorithms, it can increase the positioning accuracy significantly by the feature relevance among the CSI data collected at different positions. The field tests indicate that compared to other CSI fingerprint-based positioning methods, our proposed algorithm can effectively reduce the localization error.

Human activity recognition is a key to a lot of applications such as healthcare and smart home. In this study, we provide a comprehensive survey on recent advances and challenges in human activity recognition (HAR) with deep learning. Although there are many surveys on HAR, they focused mainly on the taxonomy of HAR and reviewed the state-of-the-art HAR systems implemented with conventional machine learning methods. Recently, several works have also been done on reviewing studies that use deep models for HAR, whereas these works cover few deep models and their variants. There is still a need for a comprehensive and in-depth survey on HAR with recently developed deep learning methods.

Indoor localization is important for a variety of applications, such as location-based services, mobile social networks, and emergency response. Fusing spatial information is an effective way to achieve accurate indoor localization with little or with no need for extra hardware. However, the existing indoor localization methods that make use of spatial information are either computationally expensive or sensitive to the completeness of landmarks. In this article, we propose a novel, low-cost, high-accuracy indoor localization method based on a landmark graph. The experimental results show that the proposed method outperforms the state-of-the-art methods.

Indoor localization has been an active research area in the last decade. Various localization systems have been proposed based on different types of signals, including but not limited to WiFi, ultrawideband, inertial measurements, and visual signals. Fingerprinting-based methods are among the most popular methods due to their accuracy and ease of deployment. However, a disadvantage to fingerprinting-based methods is the training phase in which fingerprints have to be collected at known locations and stored for future localization inquiries. Recently a few methods have been proposed to alleviate this problem. In this paper, we focus on the possibility of reducing the burden of the training phase for visual (image-based) indoor localization systems by proposing a system that automatically generates the image-location database. The proposed system will be referred to as automatic visual fingerprinting that can be paired with any indoor image-based localization method. We will also validate the proposed system through extensive experiments.

Pruning is one of the major methods to compress deep neural networks. In this paper, we propose an Ising energy model within an optimization framework for pruning convolutional kernels and hidden units. This model is designed to reduce redundancy between weight kernels and detect inactive kernels/hidden units. Our experiments using ResNets, AlexNet, and SqueezeNet on CIFAR-10 and CIFAR-100 datasets show that the proposed method on average can achieve a pruning rate of more than $50\%$ of the trainable parameters with approximately $, This paper is accepted for presentation at IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2021

Indoor localization is essential for healthcare, security, augmented reality gaming, and many other location-based services. There is currently a wealth of relevant literature on indoor localization. This article focuses on recent advances in indoor localization methods that use spatial context to improve the location estimation. Spatial context in the form of maps and spatial models have been used to improve the localization by constraining location estimates in the navigable parts of indoor environments. Landmarks such as doors and corners, which are also one form of spatial context, have proved useful in assisting indoor localization by correcting the localization error. This survey gives a comprehensive review of state-of-the-art indoor localization methods and localization improvement methods using maps, spatial models, and landmarks.

Dropout is a well-known regularization method by sampling a sub-network from a larger deep neural network and training different sub-networks on different subsets of the data. Inspired by the dropout concept, we propose EDropout as an energy-based framework for pruning neural networks in classification tasks. In this approach, a set of binary pruning state vectors (population) represents a set of corresponding sub-networks from an arbitrary original neural network. An energy loss function assigns a scalar energy loss value to each pruning state. The energy-based model (EBM) stochastically evolves the population to find states with lower energy loss. The best pruning state is then selected and applied to the original network. Similar to dropout, the kept weights are updated using backpropagation in a probabilistic model. The EBM again searches for better pruning states and the cycle continuous. This procedure is a switching between the energy model, which manages the pruning states, and the probabilistic model, which updates the kept weights, in each iteration. The population can dynamically converge to a pruning state. This can be interpreted as dropout leading to pruning the network. From an implementation perspective, unlike most of the pruning methods, EDropout can prune neural networks without manually modifying the network architecture code. We have evaluated the proposed method on different flavors of ResNets, AlexNet, l₁ pruning, ThinNet, ChannelNet, and SqueezeNet on the Kuzushiji, Fashion, CIFAR-10, CIFAR-100, Flowers, and ImageNet data sets, and compared the pruning rate and classification performance of the models. The networks trained with EDropout on average achieved a pruning rate of more than 50% of the trainable parameters with approximately

A typical deep neural network (DNN) has a large number of trainable parameters. Choosing a network with proper capacity is challenging and generally a larger network with excessive capacity is trained. Pruning is an established approach to reducing the number of parameters in a DNN. In this paper, we propose a framework for pruning DNNs based on a population-based global optimization method. This framework can use any pruning objective function. As a case study, we propose a simple but efficient objective function based on the concept of energy-based models. Our experiments on ResNets, AlexNet, and SqueezeNet for the CIFAR-10 and CIFAR-100 datasets show a pruning rate of more than $50\%$ of the trainable parameters with approximately $, This paper is accepted for presentation at IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2021. arXiv admin note: text overlap with arXiv:2006.04270, arXiv:2102.05437

Data representation can facilitate training of deep neural network when limited data is available. We have previously proposed the radial transform sampling method as a data representation technique for training neural networks. In this paper, a probabilistic framework to analyze radial transform is presented. To further elaborate it, performance of training deep neural networks on radial transform generated images for semantic segmentation of kidneys in abdominal computed tomography is evaluated. Our results show that the proposed representation method can achieve higher performance than other similar methods by translating a semantic segmentation problem to a classification problem when limited annotated images are available.

Intelligent Transportation System's main objective is to provide sustainable and optimized means of transportation for citizens of urban centers. Identifying the travel modes used by citizens, in real- time, allows these systems to better adapt transportation infrastructure and services according to user needs and possibilities of interaction. Many works have explored the use of machine learning algorithms and ensemble methods, combined with general and domain specific feature extraction techniques, for this task. More recently, several works evaluated the use of deep learning algorithms while none of them has explored the use of Recurrent Neural Networks (RNNs) in conjunction with domain feature extraction to enable the development of flexible and lightweight travel mode detection solutions based on multiple smartphone sensor readings. In this paper, we propose a deep RNN architecture for building online travel mode detection models using Long-Short Term Memory (LSTM) cells and evaluate its performance using real mobility data collected with a wide variety of sensors. The experiments show that the proposed architecture allows the generation of models with high accuracy and lower memory consumption and computation cost than state-of-the-art supervised machine learning (ML) approaches.

Label imbalance is an unavoidable issue for deep learning in the field of medical imaging. In terms of semantic segmentation, diseased tissues occur infrequently with very limited volumes, which exerts a strong bias to the learning of convolutional neural networks (CNNs). Dice loss functions are often applied for training to alleviate the problem, however they become very unstable when the dataset is extremely imbalanced. In this work, we propose a novel method that first locates the abnormal tissues, and performs segmentation in local regions. We also propose the Ising conditional random fields (CRFs) for post- processing, and apply the Digital Annealer (DA) for optimization. Our experiments demonstrate that the proposed approach is effective in learning with very imbalanced data.

The knowledge of the floor-level location of a user in a multi-storey building is important for many applications, especially for emergency response. Existing floor identification systems suffer from a variety of limitations such as low accuracy, the need for a time-consuming site survey, assumption of user encounters, knowledge of the initial floor, and/or poor applicability. In this paper, we propose a novel, zero-effort, deep learning-based floor identification system, called \textit{ZeeFi}. The proposed system uses the widely-available smartphone sensing to identify on which floor a user is located. By recognizing the ground floor automatically, the proposed system does not require site survey, initial floor knowledge, and other assumptions. To achieve accurate floor identification performance, we have developed a deep learning-based method. Experimental results show that the proposed system outperforms the state-of-the-art systems, and is very promising for large-scale deployment.

Locomotion activity recognition (LAR) is important for a number of applications, such as indoor localization, fitness tracking, and aged care. Existing methods usually use handcrafted features, which requires expert knowledge and is laborious, and the achieved result might still be suboptimal. To relieve the burden of designing and selecting features, we propose a deep learning method for LAR by using data from multiple sensors available on most smart devices. Experimental results show that the proposed method, which learns useful features automatically, outperforms conventional classifiers that require the hand-engineering of features. We also show that the combination of sensor data from four sensors (accelerometer, gyroscope, magnetometer, and barometer) achieves a higher accuracy than other combinations or individual sensors.

In this article, we present a survey of recent advances in passive human behavior recognition in indoor areas using the channel state information (CSI) of commercial WiFi systems. The movement of the human body parts cause changes in the wireless signal reflections, which result in variations in the CSI. By analyzing the data streams of CSIs for different activities and comparing them against stored models, human behavior can be recognized. This is done by extracting features from CSI data streams and using machine learning techniques to build models and classifiers. The techniques from the literature that are presented herein have great performance; however, instead of the machine learning techniques employed in these works, we propose to use deep learning techniques such as long-short term memory (LSTM) recurrent neural networking (RNN) and show the improved performance. We also discuss different challenges such as environment change, frame rate selection, and the multi-user scenario; and finally suggest possible directions for future work.

Dropout is a popular regularization method to reduce over-fitting while training deep neural networks and compress the inference model. In this paper, we propose Ising dropout with node grouping, which represents a deep multilayer perceptron (MLP) neural network as a graph with fixed grouped nodes and uses the Ising energy to drop group of nodes. This method is an extension to our proposed Ising dropout method, which had the limit of solving the Ising energy model for MLPs with limited graph order. The proposed fixed grouping method enables applying drop-out to deep MLPs with any order. Performance of this method is evaluated on handwritten digits (MNIST), Fashion-MNIST, Free Spoken Digit Dataset (FSDD), and Street View House Numbers (SVHN) datasets and compared with the standard dropout and standout methods. Preliminary results show that the proposed approach can keep the classification performance competitive to the original network while eliminating optimization of unnecessary network parameters in each training cycle. This method can compress the inference model significantly while maintaining the classification performance.

In this paper, we propose a distributed machine learning framework for training and inference in machine learning models using distributed data while preserving privacy of the data owner. In the training mode, we deploy an encoder on the end-user device which extracts high level features from input data. The extracted features along with the corresponding annotation are sent to a centralized machine learning server. In the inference mode, the users submit the extracted features from encoder instead of the original data for inference to the server. This approach enables users to contributed in training a machine learning model and use inference services without sharing their original data with the server or a third party. We have studied this approach on MNIST, Fashion, SVHN and CIFAR-10 datasets. The results show high classification accuracy of neural networks, trained with encoded features, and high encryption performance of the encoders.

Deep convolutional neural networks (DCNNs) are demonstrating their strong capability in solving computer vision problems. As for medical imaging semantic segmentation, DCNN models have become one of the most fundamental constituent. However, high-quality semantic segmentation requires pixel-wise prediction with high precision, which usually cannot be achieved only by DCNNs due to their lack of representation of pixel interactions. In this work, we propose a novel method based on conditional random fields (CRFs) and Ising model. It applies the Digital Annealer (DA) as a complement to traditional methods that only employ DCNNs. Our experiment results manifest that the use of DA can enhance the segmentation accuracy on BRATS2012 data set by over 8%. Our work potentially builds a new pathway in this realm.

Overfitting is a major problem in training machine learning models, specifically deep neural networks. This problem may be caused by imbalanced datasets and initialization of the model parameters, which conforms the model too closely to the training data and negatively affects the generalization performance of the model for unseen data. The original dropout is a regularization technique to drop hidden units randomly during training. In this paper, we propose an adaptive technique to wisely drop the visible and hidden units in a deep neural network using Ising energy of the network. The preliminary results show that the proposed approach can keep the classification performance competitive to the original network while eliminating optimization of unnecessary network parameters in each training cycle. The dropout state of units can also be applied to the trained (inference) model. This technique could compress the network in terms of number of parameters up to 41.18% and 55.86% for the classification task on the MNIST and Fashion-MNIST datasets, respectively., This paper is accepted at 44th IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2019

The Received Signal Strength (RSS) based fingerprinting approaches for indoor localization pose a need for updating the fingerprint databases due to dynamic nature of the indoor environment. This process is hectic and time-consuming when the size of the indoor area is large. The semi-supervised approaches reduce this workload and achieve good accuracy around 15 percent of the fingerprinting load but the performance is severely degraded if it is reduced below this level. We propose an indoor localization framework that uses unsupervised manifold alignment. It requires only 1 percent of the fingerprinting load, some crowd sourced readings, and plan coordinates of the indoor area. The 1 percent fingerprinting load is used only in perturbing the local geometries of the plan coordinates. The proposed framework achieves less than 5 m mean localization error, which is considerably better than semi-supervised approaches at very small amount of fingerprinting load. In addition, the few location estimations together with few fingerprints help to estimate the complete radio map of the indoor environment. The estimation of radio map does not demand extra workload rather it employs the already available information from the proposed indoor localization framework. The testing results for radio map estimation show almost 50 percent performance improvement by using this information as compared to using only fingerprints.

The popularity of online social networks (OSNs) makes them attractive platforms to advertise products. Previous work on marketing in OSNs utilized older diffusion models that do not capture the interactions of modern OSNs and hence there is a need to develop a model that accounts for the interactions that occur in current OSNs. In this paper, we introduce a new model for information flow in online social networks that captures the sharing behavior exercised by users when they pass information from one online social network to their social circles in another network. We, then, formulate a problem of maximizing the marketing reach where the diversity of users’ other social networks is taken as a constraint. We also propose a greedy algorithm to solve the aforementioned optimization problem. Numerical results show that the proposed algorithm achieves better results than algorithms that are based on classical degree centrality metric and with comparable running time.

In this paper, we propose a novel technique for sampling sequential images using a cylindrical transform in a cylindrical coordinate system for kidney semantic segmentation in abdominal computed tomography (CT). The images generated from a cylindrical transform augment a limited annotated set of images in three dimensions. This approach enables us to train contemporary classification deep convolutional neural networks (DCNNs) instead of fully convolutional networks (FCNs) for semantic segmentation. Typical semantic segmentation models segment a sequential set of images (e.g. CT or video) by segmenting each image independently. However, the proposed method not only considers the spatial dependency in the x-y plane, but also the spatial sequential dependency along the z-axis. The results show that classification DCNNs, trained on cylindrical transformed images, can achieve a higher segmentation performance value than FCNs using a limited number of annotated images., Comment: This paper is accepted for presentation at IEEE Global Conference on Signal and Information Processing (IEEE GlobalSIP), California, USA, 2018

Efficient training of supervised deep learning models for semantic segmentation requires a massive volume of annotated data. In this paper, we propose an unsupervised semantic segmentation method through the application of a radial transform method in the polar coordinate system to unannotated images. This method generates radial transformed images up to the number of pixels in the input image. Each generated image corresponds to a pixel in the original image, which is a spatial representation of the selected pixel with respect to other pixels, in the polar coordinate system. A dimension reduction method, such as a convolutional autoencoder (CAE), can extract features of these representations for clustering and later, labeling the original image by mapping the labels back to the original image. The advantage of the proposed radial transform technique is that it generates a massive number of training images by sampling pixels in the polar coordinate system from a very limited number of original images in the Cartesian coordinate system. The proposed approach achieved 88.20% accuracy in pixel-level segmentation of left kidney, right kidney, and non-kidney pixels in contrast-enhanced computed tomography (CT) images.

Background: FO is novel and more sensitive in assessing small airways than spirometry. However, FO test interpretation guidelines have not been established, limiting the utility of FO for diagnoses. Multi-layer perceptron(MLP) is a class of artificial neural networks which is capable of categorizing data based on input features using nonlinear operators. Objectives: To construct and evaluate an effective MLP model for detecting and classifying small airway obstruction patterns in FO and spirometry results. Method: The samples used are paired FO-spirometry tests of 200 individuals collected at the PFT Laboratory, UHN. The preliminary experiment involves the first 222 tests of 82 subjects(38F:44M, age=55.4±12.9yrs). Each test is labelled as airflow obstruction(AO) or not according to ATS guidelines, and randomly divided into training, validation, and test datasets at 70:15:15. MLP models with different numbers of inputs, hidden layers, and hidden units are evaluated regarding their performance in classifying AO(positive/negative). Result: The best result is obtained by the MLP model with 44 input features, including 4 biometrics (age, gender, height, weight), 36 FO parameters (R, X, Rin-ex, Xin-ex, each at 9 different frequencies), 4 spirometric parameters (FEV1, FVC, FEF50, FEF75), 1 hidden layers with 140 units and a logistic activation function. By training 200 epochs at 0.001 learning rate, the average validation accuracy is 83% for unseen patterns. Conclusion: MLP is an effective tool for identifying AO patterns in FO and spirometry. The classification accuracy is expected to improve as the training sample size expands and the model parameters are further regulated.

Many popular fingerprinting-based indoor localization methods, such as WiFi-based localization systems, rely on a dataset of fingerprints labelled by known locations (fingerprint-location dataset) to be able to localize a user by comparing the queried fingerprint with the fingerprints in the dataset. Generating and updating such a dataset is a burden and requires significant amount of time, human effort and expertise. In this work, we propose a system to build such a dataset from scratch using crowdsourced data. Consequently we reduce the required time and effort, by leveraging the information shared by the crowd. The proposed system is based on localization of user motion, hence called LocaMotion. LocaMotion takes a supervised learning perspective towards localization of user-sent motion data. In other words, LocaMotion is formalised as a classifier that extracts and assigns features from a user motion data to a sequence of points in the building. Training and testing procedures for LocaMotion will be discussed in details followed by extensive experiments to demonstrate its validity and success to localize motion patterns and generate useful fingerprint-location datasets.

A fast and robust algorithm for motion estimation based on stereo vision is proposed. The algorithm consists of two consecutive stages. First, interest points are found and tracked between four images from two consecutive time steps. A novel robust method for tracking features is proposed for this step to address shortcomings of the traditional methods for indoor hand-held/wearable scenarios. Next, motion of the camera in time is estimated via optimizing the total reprojection error associated with the robust features found using the first stage. A key assumption is that motion between the two time steps is sufficiently small or equivalently, the frames are captured close enough in time. The proposed Visual Ego-motion Estimation (VEE), is robust due to robust feature tracking in the first step, and fast due to efficient reprojection error minimization used to estimate the six-degree-of-freedom (6DoF) motion of the camera. Experiments prove comparable/superior results compared to the state-of-the-art methods in the literature especially for indoor handheld scenarios.

Human activity recognition is important for a large number of applications including indoor localization. Existing methods usually involve manually-designed features, which require expert knowledge and are laborious. Also, previous works use only the accelerometer for activity recognition, which may fail to recognize some complex activities. In this paper, we propose a deep learning-based method for locomotion activity recognition by using the combination of data from multiple smartphone built-in sensors. Eight types of locomotion activities are identified including the new ‘False Motion’ activity introduced for the first time in this work. Experimental results show that the proposed method, which learns useful features automatically, outperforms conventional classifiers that require hand-engineering of features. Also, using data from multiple sensors helps to improve recognition accuracy by about 10% compared to that using accelerometer data only.

Deep learning models have a large number of free parameters that must be estimated by efficient training of the models on a large number of training data samples to increase their generalization performance. In real-world applications, the data available to train these networks is often limited or imbalanced. We propose a sampling method based on the radial transform in a polar coordinate system for image augmentation to facilitate the training of deep learning models from limited source data. This pixel-wise transform provides representations of the original image in the polar coordinate system by generating a new image from each pixel. This technique can generate radial transformed images up to the number of pixels in the original image to increase the diversity of poorly represented image classes. Our experiments show improved generalization performance in training deep convolutional neural networks with radial transformed images., This paper is accepted for presentation at IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2018

Pathfinding in hospitals is challenging for patients, visitors, and even employees. Many people have experienced getting lost due to lack of clear guidance, large footprint of hospitals, and confusing array of hospital wings. In this paper, we propose Halo; An indoor navigation application based on voice-user interaction to help provide directions for users without assistance of a localization system. The main challenge is accurate detection of origin and destination search terms. A custom convolutional neural network (CNN) is proposed to detect origin and destination search terms from transcription of a submitted speech query. The CNN is trained based on a set of queries tailored specifically for hospital and clinic environments. Performance of the proposed model is studied and compared with Levenshtein distance-based word matching., This paper is accepted for presentation at 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications

An indoor tracking and navigation system based on measurements of received signal strength (RSS) in wireless local area network (WLAN) is proposed. In the system, the location determination problem is solved by first applying a proximity constraint to limit the distance between a coarse estimate of the current position and a previous estimate. Then, a Compressive Sensing-based (CS--based) positioning scheme, proposed in our previous work , , is applied to obtain a refined position estimate. The refined estimate is used with a map-adaptive Kalman filter, which assumes a linear motion between intersections on a map that describes the user's path, to obtain a more robust position estimate. Experimental results with the system that is implemented on a PDA with limited resources (HP iPAQ hx2750 PDA) show that the proposed tracking system outperforms the widely used traditional positioning and tracking systems. Meanwhile, the tracking system leads to 12.6 percent reduction in the mean position error compared to the CS-based stationary positioning system when three APs are used. A navigation module that is integrated with the tracking system provides users with instructions to guide them to predefined destinations. Thirty visually impaired subjects from the Canadian National Institute for the Blind (CNIB) were invited to further evaluate the performance of the navigation system. Testing results suggest that the proposed system can be used to guide visually impaired subjects to their desired destinations.

The recent growing interest for indoor Location-Based Services (LBSs) has created a need for more accurate and real-time indoor positioning solutions. The sparse nature of location finding makes the theory of Compressive Sensing (CS) desirable for accurate indoor positioning using Received Signal Strength (RSS) from Wireless Local Area Network (WLAN) Access Points (APs). We propose an accurate RSS-based indoor positioning system using the theory of compressive sensing, which is a method to recover sparse signals from a small number of noisy measurements by solving an `1-minimization problem. Our location estimator consists of a coarse localizer, where the RSS is compared to a number of clusters to detect in which cluster the node is located, followed by a fine localization step, using the theory of compressive sensing, to further refine the location estimation. We have investigated different coarse localization schemes and AP selection approaches to increase the accuracy. We also show that the CS theory can be used to reconstruct the RSS radio map from measurements at only a small number of fingerprints, reducing the number of measurements significantly. We have implemented the proposed system on a WiFi-integrated mobile device and have evaluated the performance. Experimental results indicate that the proposed system leads to substantial improvement on localization accuracy and complexity over the widely used traditional fingerprinting methods.

An indoor positioning and navigation system based on measurements of received signal strength in wireless local area network is proposed. In the system, the location determination problem is solved by applying compressive sensing, which offers recovery of sparse signals from a small number of noisy measurements by solving an l1-minimization problem. The refined estimate is then used with a map-adaptive Kalman filter for real-time tracking. A navigation module integrated with the tracking system guides users to pre-defined destinations with voice instructions. Experimental results with a system that was implemented on a PDA shows that the proposed tracking system is lightweight so that it can be used on a resource constrained platform while outperforming the widely used traditional positioning and tracking systems. A pilot study was carried out with 30 visually impaired subjects from the Canadian National Institute for the Blind. Testing results show that the proposed system can be used to guide visually impaired subjects to their desired destinations with a very high success rate.

In this paper, we address the problem of gesture recognition using the theory of random projection (RP) and by formulating the whole recognition problem as an l1-minimization problem. The gesture recognition system operates primarily on data from a single 3-axis accelerometer and comprises two main stages: a training stage and a testing stage. For training, the system employs dynamic time warping as well as affinity propagation to create exemplars for each gesture while for testing, the system projects all candidate traces and also the unknown trace onto the same lower dimensional subspace for recognition. A dictionary of 18 gestures is defined and a database of over 3700 traces is created from seven subjects on which the system is tested and evaluated. To the best of our knowledge, our dictionary of gestures is the largest in published studies related to acceleration-based gesture recognition. The system achieves almost perfect user-dependent recognition, and mixed-user and user-independent recognition accuracies that are highly competitive with systems based on statistical methods and with the other accelerometer-based gesture recognition systems available in the literature.

The problem of multiple target tracking using a passive direction-finding system is addressed when the number of targets is not known a priori. A new method is proposed that is suitable for systems operating in low signal-to-noise ratio and high clutter. Such conditions cause unpredictable variations of stochastic characteristics of noise and signals (especially for wideband frequency ones) and create ambiguity in the output of direction-finding algorithms. In this paper, we use the predictive description length (PDL) technique, which is an information theoretic approach, and by suitable modeling, we minimize the predictive codelength for statistical data description of position measurements. The PDL-dynamic programming (DP) method is also presented, which employs the DP algorithm to reduce the computational load of the PDL technique. The concept of tracking time-varying number of targets makes PDL-DP a suitable technique for target tracking in practical systems

Crowdsourcing allows for rapid deployment of indoor localization systems. However, compared to the conventional methods, crowdsourcing might collect fewer received signal strength (RSS) values, hence result in greater influence to outliers in RSS values. In this paper, we propose an algorithm to detect such outliers and to substitute them with more suitable RSS values. In particular, we investigate the relationship of RSS values between adjacent locations using a signal propagation model and show that the outliers can be corrected using a signal propagation model. We propose the Signal Propagation-based Outlier Reduction Technic (SPORT) for identifying and adjusting outlier values in both the offline training phase and the online localization phase. Experimental results show that SPORT greatly smoothens the radio map and improves the location accuracy.

We investigate the problem of graph matching to translate topological indoor localization to geographical localization, by modeling the building map and the semantic maps as graphs. A matching algorithm based on hidden Markov models is proposed. The matching algorithm is tested on both simulations and real data and accuracies as high as 94% on real data is achieved, while the matching is shown to be robust to noise, scale variance and partial matching via simulations.

In this paper, we propose an image-based localization system, applicable for a number of indoor scenarios including office buildings, airports, chain stores, etc. In such applications, text/numbers are suitable distinctive landmarks for localization. The proposed system takes advantage of OCR to read the text/numbers and provide a rough estimate using the floor plan. Next, it performs OCR-aided stereo feature matching to refine the estimate by solving a PnP problem. Experiments show that this system achieves a median localization error of less than 50 cm for test positions located as far as 7 meters from a 20cm by 30cm number plate using different test devices in a university building scenario.

Indoor positioning based on the received signal strength (RSS) of the WiFi signal has become the most popular solution for indoor localization. In order to realize the rapid deployment of indoor localization systems, solutions based on crowdsourcing have been proposed. However, compared to conventional methods, lots of different devices are used in crowdsourcing system and less RSS values are collected by each device. Therefore, the crowdsourced RSS values are more erroneous and can result in significant localization errors. In order to eliminate the signal strength variations across diverse devices, the Linear Regression (LR) algorithm is proposed to solve the device diversity problem in crowdsourcing system. After obtaining the uniform RSS values, a graph-based semi-supervised learning (G-SSL) method is used to exploit the correlation between the RSS values at nearby locations to estimate an optimal RSS value at each location. As a result, the negative effect of the erroneous measurements could be mitigated. Since the AP locations need to be known in G-SSL algorithm, the Compressed Sensing (CS) method is applied to precisely estimate the location of the APs. Based on the location of the APs and a simple signal propagation model, the RSS difference between different locations is calculated and used as an additional constraint to improve the performance of G-SSL. Furthermore, to exploit the sparsity of the weights used in the G-SSL, we use the CS method to reconstruct these weights more accurately and make a further improvement on the performance of the G-SSL. Experimental results show improved results in terms of the smoothness of the radio map and the localization accuracy.

In this paper, we propose a homography-aware semi-supervised formulation for the logo-based indoor localization problem using smartphone cameras. Our method labels unmatched feature points detected inside the logo parts of query images with their estimated 3D coordinates. The 3D coordinates are computed using the homography estimated from the matched features. We demonstrate the accuracy improvement and lower localization error variance resulted from our semi-supervised approach via experiments in an indoor scenario.

We propose a weighted KNN Epipolar Geometry-based method for vision-based indoor localization using cellphone cameras. The proposed method is applicable for fine localization whenever a pose-tagged (position + rotation matrix) image database is available rather than just Geo-tagged one. To the best of our knowledge, this is the first that Epipolar geometry has been utilized for fine localization in indoor applications using smartphone images. We compare the performance of our method with two outstanding literature works. It will be also demonstrated that the proposed method can extrapolate the location of queries located outside of the database location set, as well as compensate for the small databases, where database location set is sparse as two additional new features of this method.

The growing interest for location-based services motivates many researchers to study different localization techniques for indoor environments. The main objective of these studies is to find a balance point between the accuracy of the scheme and its deployment/training cost. RSS-based schemes and in particular Graph-based Semi-Supervised Learning (G-SSL) constitute a group of techniques which has low setup cost and good localization accuracy. In this paper, we analyze the G-SLL scheme and show that, despite its high performance, the G-SSL method (in its original format) is not a very accurate model for a localization problem. Based on this observation and to improve the accuracy of localization, we propose an alternative approach which incorporates our knowledge of wireless signal propagation into the label propagation mechanism. Experimental results are then used to evaluate the performance of the proposed scheme compared to the original G-SSL.

There is a recent interest in developing algorithms for the reconstruction of jointly sparse signals, which arises in a large number of applications such as sensor networks. In many of these applications, we encounter extremely large problem sizes for which algorithms with low computational complexity are required. Recently, an algorithm called iterative hard thresholding has been proposed, which is faster than the l 1 -minimization and greedy algorithms for compressed sensing. In this work, we extend the iterative hard thresholding algorithm to jointly sparse signals and will investigate the performance of our proposed algorithm analytically by giving conditions under which the exact reconstruction could happen. We will show that our algorithm is faster than the state of the art algorithms for jointly sparse signals while showing similar performance.

In this paper, we define a new Joint Sparsity Model (JSM) and use Principal Component Analysis followed by Minimum Description Length and Compressive Sensing to reconstruct spatially and temporally correlated signals in a sensor network. The proposed model decomposes each sparse signal into two sparse components. The first component has a common support across all sensed signals. The second component is an innovation part that is specific to each sensor and might have a support that is different from the support of the other innovation signals. We use the fact that the common component generates a common subspace that can be found using the principal component analysis and the minimum description length. We show that with this general model, we can reconstruct the signal with smaller samples that are needed by the direct application of the compressive sensing on each sensor.

We propose a gesture recognition system based primarily on a single 3-axis accelerometer. The system employs dynamic time warping and affinity propagation algorithms for training and utilizes the sparse nature of the gesture sequence by implementing compressive sensing for gesture recognition. A dictionary of 18 gestures is defined and a database of over 3,700 repetitions is created from 7 users. Our dictionary of gestures is the largest in published studies related to acceleration-based gesture recognition, to the best of our knowledge. The proposed system achieves almost perfect user-dependent recognition and a user-independent recognition accuracy that is competitive with the statistical methods that require significantly a large number of training samples and with the other accelerometer-based gesture recognition systems available in literature.

Two techniques based on the minimum description length (MDL) and the predictive stochastic complexity (PSC) are proposed for sinusoidal signal detection. The MDL and PSC criteria are the codelength of the observation and the model. The proposed techniques decompose the observation vector into its components in the signal and noise subspaces. The noise component is encoded for several model orders. The best model is selected by minimizing the codelength.