Search

There are several advanced methods to detect objects, humans, etc., which requires large dataset for training. In this paper, a difference of difference (DoD) technique has been proposed to detect normal and abnormal conditions with minimal dataset during surveillance. This dataset has been used for setting a threshold value to detect condition without training. Captured images have been compared with created dataset toget net difference value and this value has been compared with the threshold value to detect normal and abnormal conditions. This technique can be used at no-man zone areas, boundaries and security borders, and also it can be used as a front-liner along with any advanced technique for surveillance application. This method has been simulated on real images using MATLAB and conducted practical experiment using quad rotor, RPi and ROS.

Availability of a large number of application-specific spectral libraries has generated a great deal of interest in semi-blind unmixing of the hyperspectral image in both remote sensing and signal processing community. This study presents a novel, semi-supervised, parameter-free algorithm which employs sparsity measures for library pruning. The overall algorithm includes sparsity criteria based library pruning and sparse inversion method for abundance computation. In the pruning process, each library element is removed from the spectral library and the corresponding sparse abundance matrix is computed. The library elements which lead to higher sparsity are adjudged as image endmembers, based on the assumption that elimination of actual image endmember enhances sparsity level. The authors also present a detailed exploration of standard sparsity measures. They calculate the abundance of the pruned library by maximising Gini index or pq-norm sparsity, which satisfies the desirable sparsity properties and is easier to compute. The abundance calculation task is solved using the adaptive direction method of multipliers. The experimental results on several real and synthetic image datasets demonstrate the computational efficiency and proficiency the authors’ method in the presence of noise and highly coherent spectral library.

Band selection of hyperspectral images is an optimal feature selection method, which aims at reducing the computational burden associated with processing the whole data. The significant and informative bands identified by the band selection process lead to efficient, compact representation of the image data and produce a satisfactory performance in the succeeding applications viz. classification, unmixing, target detection and so on. In this study, the authors present an unsupervised manifold clustering approach for band selection, which accounts for different types of scenarios. Unlike other band selection approaches, the authors’ proposed manifold clustering framework identifies the informative bands by utilising the interrelation between the bands and accounts for the multi-manifold structure prevalent in some real images. The proposed band selection framework identifies the optimal number of clusters by cluster validity index, clusters the bands by manifold clustering and select representative bands from each cluster according to graph weight. Their proposed manifold clustering approach is a generic clustering approach, which produces a satisfactory result even when the data contains non-linearity. The information theoretic performance measures, classification and unmixing performance on real image experiments demonstrate the proficiency of their proposed band selection algorithm.

This paper presents a novel detection based multipath correlation filter (DT-MPT) for moving single object tracking from Unmanned Aerial Vehicles (UAVs). Most object trackers suffer under complicated situations such as the fast motion of objects, camera motion, similar objects, and occlusions, usually found in UAV videos. This paper utilizes a correlation filter based tracking algorithm along with a motion estimation block, a novel detection block, and a new multipath block to cope with those challenges. The detection block uses the concept of the difference image to detect moving objects, while the multipath block considers multiple positions for a few looks ahead frame for accurate localization of an object position. The performance analysis on three challenging benchmark datasets UAV123@10fps, UAV20L and DTB70 show that the proposed tracker DT-MPT outperforms its base tracker as well as most of the other top-performing state of art correlation filter-based trackers while running in almost real-time.

Farmers face a lot of problems, while cultivating agricultural plants. They do not get the desired output in agriculture due to the different types of diseases from which the plants suffer. Therefore, it is important to detect the disease(s) of the plant. In this study, a vision-based plant disease detection method has been developed, where convolutional neural network (CNN) model has been used for classification of the plant disease and vision assistant toolbox of LabVIEW software is utilized for disease detection. This vision-based disease detection through a smart camera will be one of the important modules of an intelligent agricultural robot.

Robotic grasping of arbitrary objects even in completely known environments still remains a challenging problem. Most previously developed algorithms had focused on fingertip grasp, failing to solve the problem even for fully actuated hands/grippers during adaptive/wrapping type of grasps, where each finger makes contact with object at several points. Kinematic closed form solutions are not possible for such an articulated finger which simultaneously reaches several given goal points. This paper, presents a framework for computing best grasp for an underactuated robotic gripper, based on a novel object slicing method. The proposed method quickly finds contacts using an object slicing technique and uses grasp quality measure to find the best grasp from a pool of grasps. To validate the proposed method, implementation has been done on twenty-four household objects and toys using a two finger underactuated robot gripper on simulated environment. Unlike the many other existing approaches, the proposed approach has several advantages: it can handle objects with complex shapes and sizes; it does not require simplifying the objects into primitive geometric shapes; Most importantly, it can be applied on point clouds taken using depth sensors; it takes into account gripper kinematic constraints and generates feasible grasps for both adaptive/enveloping and fingertip types of grasps.

This paper proposes a novel, faster unmixing approach based on a convex geometric approach and slow feature analysis, and extends it to perform efficient library pruning based semi-blind unmixing. The former algorithm performs complete blind unmixing, whereas the subsequent algorithm performs exact library pruning for semi-blind unmixing. Slow feature analysis algorithm impels the pure pixels towards the exterior region. The work identifies the endmembers by detecting the extreme points. On the other hand, the proposed dictionary pruning method augments each library element with the data, extracts the extreme points and calculate a volume of the transformed data. The augmentation of actual image endmember changes the structure of the simplex. The library pruning method proposes an index to capture the change in the volume of the simplex and identify the actual image endmember. We evaluated unmixing performance as well as the runtime on real images, which ratifies the computational edge and proficiency of our proposed method.

This paper presents a new correlation filter based visual object tracking method to improve the accuracy and robustness of trackers. Most of the current correlation filter based tracking methods often suffer in situations such as fast object motion, the presence of similar objects, partial or full occlusion. One of the reasons for that is that object localization is performed by selecting only a single location at each frame (greedy search technique). Instead of choosing a single position, the multipath based tracking method considers multiple locations in each frame to localize object position accurately. In this paper, the multipath based tracking method is applied to improve the performance of the efficient convolution operator with handcrafted features (ECO-HC), which is a top performing tracker in many visual tracking datasets. We have performed comprehensive experiments using our efficient convolution operator with multipath (ECO-MPT) tracker on UAV123@10fps and UAV20L datasets. We have shown that our tracker outperforms most of the state-of-art trackers in all those benchmark datasets.

In this work a memory efficient topological map generation algorithm has been proposed using local descriptors. A topological map is a graphical data structure where each node signifies an area within an environment. These nodes are connected by links which ensure the presence of a physical path between the pair. Experiments have been conducted with feature descriptors using a vocabulary based approach. These approaches take huge memory and time. To deal with these a KD-tree based map generation algorithm has been proposed where each node in the tree stores a descriptor and a table of occurrence. This table stores node ids of the locations, where the corresponding descriptor is present. The map generation algorithm is a two-stage algorithm. In the first stage, the visual similarity based position identification is conducted in order to check for loop-closures. It is followed by a corrective step on validating the decision of loop closure, if any. The table of occurrence keeps track of presence of each descriptor. The least occurring descriptors are pruned at regular intervals, making the algorithm memory-efficient. The approach has been experimented with several benchmark datasets.

In this work, a topological map generation algorithm has been proposed, using global descriptors. A topological map is a graphical data structure where each node signifies a fixed position in space. These nodes are connected by links which ensure the presence of a physical path between the pair. A vision based topological map generation algorithm has been addressed in this work. A sequence of monocular images taken in regular intervals has been used as an input of the mapping algorithm. Descriptors were computed from those images to represent the signature of the corresponding position. A KD-tree has been maintained to store these features in the memory. A correction algorithm has also been developed to rectify false matches based on the nature of the observations. Experiments have been performed on some of the popular benchmark datasets available in the literature.

In today’s era of mass customization, assembly automation systems should be designed with necessary production flexibility to cope with the growing product varieties to adapt to diverse customer requirements, yet the production costs should not be significantly different from those of comparable products made by mass production. In order to cope with this product variety-cost trade-off, robotics offers a flexible automation technology for turning assembly systems into efficient and flexible systems. Despite their great potential for high flexibility, there is a range of issues which must be addressed for its successful implementation. This chapter examines some of these key issues and challenges, reviews the results of previous research and describes our ongoing research on development of a flexible assembly system for mechanical products, using an industrial robot with machine vision guidance and dexterous multi-finger gripper. As part of the research work reported in this chapter, a Sexual Genetic Algorithm (SGA)-based approach for generation of optimal assembly sequence, a knowledge-based system for generating the robot task-level plan, a multi-finger robot gripper for flexible assembly based on a tendon-driven mechanism and an impedance control algorithm, and finally a strategy for implementation of robotic assembly under machine vision guidance have been presented.

Unmixing of hyperspectral image is a widely studied topic. It performs semi-blind source separation to extract endmember spectra and calculate abundance of endmembers. This paper proposes a fast and accurate method for performing complete unmixing of the hyperspectral image-Joint unmixing and estimation of the number of endmembers present in the image relying on the convex properties of abundance of the endmembers. Experiments performed in a large number of synthetic and real hyperspectral images verify the accuracy of the proposed method and advantages over the present state of the art methods.

The digital scans of double sided documents suffer from distortions because the contents on the back side of the document often shows up on the front side in the scans and vice-versa either due to transparency of the paper or due to ink-bleeding. This is show-through effect. In this paper a state-space based approach is proposed for removing this commonly found contamination in the scans of duplex printed documents. Separate state-space representation for signals and parameters are defined and a dual state-parameter estimation approach is employed to alleviate the degradation in the scans. The proposed framework functions with two Kalman filters running simultaneously, Kalman state filter to estimate states and Kalman parameter filter for estimating parameters. The simulation results show the effectiveness of the algorithm in removing the show-through.

Estimation of number of endmembers in a hyperspectral image is the first stage in spectral unmixing of hyperspectral image. Spectral unmixing method is a blind source separation method that tries to estimate the spectra of the endmembers present in the image along with its fractional abundance. Eigenvalues of Covariance matrix of the data gives an estimation number of significant signal sources present in a linear mixture. This paper proposes a nonparametric Eigenvalue based index called GAP Index for estimating the number of endmembers present in a hyperspectral image. Experiments done in a large number of synthetic and real hyperspectral images. The proposed method is a noise robust, computationally efficient, accurate and nonparametric method. Experiments done in a large number of synthetic and real images in different noise levels show the accuracy and noise robustness of the proposed method.

Dictionary pruning step is often employed prior to the sparse unmixing process to improve the performance of library aided unmixing. This paper presents a novel recursive PCA approach for dictionary pruning of linearly mixed hyperspectral data motivated by the low-rank structure of a linearly mixed hyperspectral image. Further, we propose a mutual coherence reduction method for pre-unmixing to enhance the performance of pruning. In the pruning step we, identify the actual image endmembers utilizing the low-rank constraint. We obtain an augmented version of the data by appending each image endmember and compute PCA reconstruction error, which is a convex surrogate of matrix rank. We identify the pruned library elements according to PCA reconstruction error ratio (PRER) and PCA reconstruction error difference (PRED) and employ a recursive formulation for repeated PCA computation. Our proposed formulation identifies the exact endmember set at an affordable computational requirement. Extensive simulated and real image experiments exhibit the efficacy of the proposed algorithm in terms of its accuracy, computational complexity and noise performance.

Anatomy of human hand is very complex in nature. The structure of human hand consists of number of joints, bones, muscles and tendons, which creates a wide range of movements. It is very difficult to design a robotic hand and incorporate all the features of a normal human hand. In this paper, the model of a three finger robotic hand has been proposed. To replace the muscles and tendons of real human hand, it is proposed to use tendon wire and place the actuator at the palm. The advantage of using tendon and placing actuator at remote location is that it actually reduces the size of the hand. Pulling the tendon wire produces flexor motion in the hand finger. Currently torsional spring is considered at the joint for the extension motion of the finger. The purpose of design of such a hand is to grasp different kinds of object shapes. The paper further presents a kinematics model of the three finger hand and a mapping function to map the joint space coordinates to tendon space coordinates. Finally the hand model is simulated to validate the kinematics equations.

Affective computing has become a growing field of research activities due to its wide use of application in human computer interface. Emotion recognition is one of the state-of-the-art techniques in determining current psychological state of human being. Human emotions are very overlapping in nature and thus it needs an efficient feature-extractor and classifier assembly. This paper reports a novel non-invasive technique to classify human emotion through thermal images of face. Hu's moment invariants of different patches have been fused with histogram statistical feature and used as robust features in multiclass support vector machine based classification. It is found that emotions from thermal image can be classified by the proposed method with a satisfactory performance.

This paper presents an iterative Takagi Sugeno Fuzzy Model (TSFM) identification. Interval Type-2 Recursive Fuzzy C-Means (IT2RFCM) clustering algorithm has been used to classify the data space to obtain premise variable parameters and Weighted Recursive Least Square (WRLS) technique has been used to determine consequence parameters of each linear model. IT2RFCM clustering algorithm has been obtained from type-1 Fuzzy C-Means clustering algorithm by introducing fuzziness parameters. The effectiveness of the proposed IT2RFCM algorithm has been validated on Mackey-Glass time series data.

Automated detection of neonatal sleep apnea is essential for constrained environments with high patient to nurse ratio. Existing studies on apnea detection mostly target adults, and use invasive sensors. Few approaches detect apnea using video monitoring, by identifying absence of respiratory motion. They apply frame differencing and thresholding, not suitable for neonates due to their subtle respiratory motion intermixed with other body movements. Proposed method first applies motion magnification. Subsequently, it filters respiration motion using dynamic thresholding. The technique is benchmarked with simulated motion of varying respiration frequencies. When validated with neonatal video data, proposed method achieves both > 90% sensitivity and specificity.

This chapter gives an overview of different ‘soft computing’(also known as ‘computational intelligence’) techniques that attempt to mimic imprecision and understanding of natural phenomena for algorithm development. It gives a detailed account of some of the popular evolutionary computing algorithms such as genetic algorithms (GA), particle swarm optimization (PSO), ant colony optimization (ACO) and artificial immune systems (AIS). The paradigm of fuzzy sets is introduced and two inferencing methods, the Mamdani model and the Takagi–Sugeno–Kang (TSK) model, are discussed. The genesis of brain modelling and its approximation so as to develop neural networks that can learn are also discussed. Two very popular computational intelligence techniques, support vector machines (SVMs) and rough sets, are introduced. The notions of hybridization that have aroused interest in developing new algorithms by using the better features of different techniques are mentioned. Each section contains applications of the respective technique in diverse domains.

In this paper, we use the approach of adaptive critic design (ACD) for control, specifically, the action-dependent heuristic dynamic programming (ADHDP) method. A least squares support vector machine (SVM) regressor has been used for generating the control actions, while an SVM-based tree-type neural network (NN) is used as the critic. After a failure occurs, the critic and action are retrained in tandem using the failure data. Failure data is binary classification data, where the number of failure states are very few as compared to the number of no-failure states. The difficulty of conventional multilayer feedforward NNs in learning this type of classification data has been overcome by using the SVM-based tree-type NN, which due to its feature to add neurons to learn misclassified data, has the capability to learn any binary classification data without a priori choice of the number of neurons or the structure of the network. The capability of the trained controller to handle unforeseen situations is demonstrated.

A technique for building a multilayer perceptron classifier network is presented. Initially, a single perceptron tries to correctly classify as many samples as possible. Misclassified samples are taken care of by adding as bias the output of up to two neurons to the parent neuron. The final classification boundary between the two disjoint half spaces at the output of the parent neuron is determined by a maximum margin classifier type SVM applied jointly to the training set of the parent neuron along with the correcting inputs from its child neuron(s). The growth of a branch in the network ceases when the terminal neuron is able to correctly classify all samples from its training set. No a priori assumptions need to be made regarding the number of neurons in the network or the kernel of the SVM classifier. Examples are presented to illustrate the effectiveness of the technique.