Search

This paper advances video analytics with a focus on crowd analysis for Hajj and Umrah pilgrimages. In recent years, there has been an increased interest in the advancement of video analytics and visible surveillance to improve the safety and security of pilgrims during their stay in Makkah. It is mainly because Hajj is an entirely special event that involve hundreds of thousands of people being clustered in a small area. This paper proposed a convolutional neural network (CNN) system for performing multitude analysis, in particular for crowd counting. In addition, it also proposes a new algorithm for applications in Hajj and Umrah. We create a new dataset based on the Hajj pilgrimage scenario in order to address this challenge. The proposed algorithm outperforms the state-of-the-art approach with a significant reduction of the mean absolute error (MAE) result: 240.0 (177.5 improvement) and the mean square error (MSE) result: 260.5 (280.1 improvement) when used with the latest dataset (HAJJ-Crowd dataset). We present density map and prediction of traditional approach in our novel HAJJ-crowd dataset for the purpose of evaluation with our proposed method.

Background: This paper focuses on advances in crowd control study with an emphasis on high-density crowds, particularly Hajj crowds. Video analysis and visual surveillance have been of increasing importance in order to enhance the safety and security of pilgrimages in Makkah, Saudi Arabia. Hajj is considered to be a particularly distinctive event, with hundreds of thousands of people gathering in a small space, which does not allow a precise analysis of video footage using advanced video and computer vision algorithms. This paper aims to propose an algorithm based on a Convolutional Neural Networks model specifically for Hajj applications. Additionally, the work introduces a system for counting and then estimating the crowd density. Methods: The model adopts an architecture which detects each person in the crowd, spots head location with a bounding box and does the counting in our own novel dataset (HAJJ-Crowd). Results: Our algorithm outperforms the state-of-the-art method, and attains a remarkable Mean Absolute Error result of 200 (average of 82.0 improvement) and Mean Square Error of 240 (average of 135.54 improvement). Conclusions: In our new HAJJ-Crowd dataset for evaluation and testing, we have a density map and prediction results of some standard methods.

In proportion to the growth in human population, there has been a substantial rise in the number of crowds in public places. The more crowded a place, the more risk of stampedes. Therefore, crowd management is very critical to ensure the safety of the crowds. Crowd monitoring is an effective approach to monitor, control and understand the behavior of the density of the crowd. One of the efficient automated video monitoring techniques to ensure public safety is crowd density estimation. Crowd density analysis is used primarily in public areas that are usually crowded with people such as stadiums, parks, shopping malls and railway stations. In this research, crowd density analysis by machine learning is presented. The main purpose of this model is to determine the best machine learning algorithm with the highest performance for crowd density classification. This model is focusing on machine learning algorithms such as traditional machine learning algorithms and deep learning algorithms. For traditional machine learning algorithms, Histogram Oriented Gradients (HOG) and Local Binary Pattern (LBP) have been used to extract important features from the input crowd images before being fed into Support Vector Machine (SVM) for classification. For deep learning algorithms, custom Convolutional Neural Network (CNN) together with two famous CNN architectures named Residual Network (ResNet) and Visual Geometry Group Network (VGGNet) were implemented as other methods in this paper for comparison. Other than that, the performance evaluation of the algorithm was measured based on the accuracy of the models. The performance of all different models was recorded and compared.

Traffic signs images when captured in real environment are small in size compared to other objects. Thus, making it difficult to be accurately detected, more so identified. Of recent, the convolutional neural network (CNN) has been making tremendous progress in object detection due to its high accuracy and fast execution. YOLO (You Only Look Once) is an object detection method, which uses the CNN core module and suited for detection of traffic sign in real environment. In this paper, we implemented the YOLOv3 framework for identification of Malaysian traffic sign in real environment. With the inclusion of the spatial pyramid pooling (SPP) in our YOLOv3 framework, it enables the identification of far and small traffic signs in real environment due to pooling of multi-sizes grids of SPP. We presented results of this implementation on our data set of 2000 images of Malaysian traffic sign and the results at 0.5% confidence level are in the range of 90% with a mean average precision of 82.5%, which is promising.

The analysis of retinal features, such as blood vessels, optic disc and fovea, plays an important role in the detection of several diseases. This paper presents a method for automated optic disc segmentation from colour fundus images. The proposed method comprises three major stages, namely optic disc localisation, preprocessing and segmentation. Localisation is performed using the fast Fourier transform-based template matching to obtain a seed point located on the optic disc which is then used as an input to the region growing technique for the purpose of segmentation. Three sets of fundus images, namely DRIVE, MESSIDOR and a LOCAL database are used to measure the accuracy of the proposed method. From the experimental results, it is found that the proposed localisation method achieves success rates of 100, 98.91 and 97.56% for these databases, respectively, which are comparable to other known methods. The proposed segmentation method is compared with several known segmentation methods using DRIVE...

Fingertips detection is important for recognition of static gesture in sign language. This paper presents a new method that is based on YCbCr colour space and skeletonization for fingertips detection towards gesture segmentation and recognition. The method begins with the conversion of an image in RGB colour space into YCbCr colour space. Then, the chrominance Cb and Cr are extracted from the YCbCr colour space. A thresholding technique, which is based on a pre-defined range for Cb and Cr components representing the skin colour value is used to extract the hand from the background and to achieve the binary image. A morphological processing is performed on the obtained binary image to remove noise and unwanted image pixels. The candidate fingertips position is then calculated based on skeletonization algorithm and tracing process. The centroid is subsequently found, after which the Euclidean distances between the pixels’ coordinates that belong to the candidate fingertips and the centroid are calculated towards validating their representation of the actual fingertips. Based on the proposed method, the fingertips of twenty-six American Sign Language (ASL) alphabets hand sign samples were detected successfully and the gesture was correctly recognized with an accuracy of 96.3% during the conducted experiments.

This paper describes the acquisition setup and development of a new gait database, MMUGait. This database consists of 82 subjects walking under normal condition and 19 subjects walking with 11 covariate factors, which were captured under two views. This paper also proposes a multiview model-based gait recognition system with joint detection approach that performs well under different walking trajectories and covariate factors, which include self-occluded or external occluded silhouettes. In the proposed system, the process begins by enhancing the human silhouette to remove the artifacts. Next, the width and height of the body are obtained. Subsequently, the joint angular trajectories are determined once the body joints are automatically detected. Lastly, crotch height and step-size of the walking subject are determined. The extracted features are smoothened by Gaussian filter to eliminate the effect of outliers. The extracted features are normalized with linear scaling, which is followed by feature selection prior to the classification process. The classification experiments carried out on MMUGait database were benchmarked against the SOTON Small DB from University of Southampton. Results showed correct classification rate above 90% for all the databases. The proposed approach is found to outperform other approaches on SOTON Small DB in most cases.

This paper presents a method for automated optic disc segmentation from color fundus images. The proposed method comprises three major stages, namely optic disc localization, preprocessing and segmentation. Localization is performed using the fast Fourier transform based template matching to obtain a seed point located on the optic disc which is then used as an input to the region growing technique for the purpose of segmentation. DRIVE database is used to measure the accuracy of the proposed method. From the experimental results, it is found that the proposed localization method achieves success rates of 100%. The proposed segmentation method is compared with several known segmentation methods using the same databse. Based on the results, it is found that the proposed method achieves values of 87.16%, 91.27%, 99.81%, 90.56%, 98.68%, and 89.71% in terms of overlap, sensitivity, specificity, PPV, accuracy, and kappa coefficient respectively, which are higher compared to the results achieved by other known methods.

This paper is about analysing the uniqueness of twins by gait biometric. The motivation arises due to twins, having facial similarity may lay difficulties to a video-based recognition system employing face biometric. Gait, a biometric based on the way a person walk, can perhaps be a useful descriptor. Due to the small size data set, classification via leave-one-out cross validation may not be sufficient to test gait’s viability as a descriptor for twins. Thus, this paper proposes a jackknife-like validation in a matched-pair classification. Comparing between the results of both validation approaches, results of the proposed method have shown to be promising. The results perhaps may point to the uniqueness of each individual twin by gait biometric.

This paper presents a classification analysis of gait biometric on twins and non-twins siblings. The aim of this paper is to investigate the existence or inexistence of similarity in the gait of twins and compare it to the gait of non-twins siblings. The motivation behind this paper is that a video-based surveillance system may not be able to rely on face biometric alone when dealing with twins. The features used are the angular displacement walking trajectories of lower limbs. Also this paper proposes a gait cycle normalization task via Bezier polynomial root-finding and re-sampling to ensure a robust analysis against differences in walking speed. Two established classifiers, the linear discriminant analysis (LDA) and k-nearest neighbor are used to classify the data sets of twins and non-twins siblings. Results may indicate that there is similarity in the gait of twins.

In order to determine the appropriate number of thresholds that a grey-level image should be thresholded by so that the resulting image preserves as much information as possible from the original image using the least possible bits, we propose in this paper, a novel criterion for multilevel image thresholding. The criterion is a weighted sum of within-class variance and the number of edge pixels in the thresholded image. To determine the appropriate number of thresholds, an image has to be thresholded iteratively with increasing number of thresholds by any standard thresholding method and the solution that minimizes the proposed criterion is chosen as the appropriate solution. We also present an efficient technique to compute the number of edge pixels. Experiments on a variety of real-world images show that the proposed criterion gives visually more consistent results compared to the most widely used automatic thresholding criterion proposed by Yen et al. (1995).© (2012) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

This paper presents a simple model-free gait extraction approach for human identification by using Support Vector Machine. The proposed approach consists of three parts: extraction of human gait features from enhanced human silhouette, smoothing process on extracted gait features and classification by Support Vector Machine (SVM). The gait features extracted are height, width, crotch height, step-size of the human silhouette and joint trajectories. To improve the classification performance, two of these extracted gait features are smoothened before the classification process in order to alleviate the effect of outliers. The proposed approach has been applied on SOTON covariate database, which is comprised of eleven subjects walking bidirectional in a controlled indoor environment with thirteen different covariate factors that vary in terms of apparel, walking speed, shoe types and carrying objects. From the experimental results, it can be concluded that the proposed approach is effective in human identification from a distance.

Gait is another potential human biometrics to look into whenever face recognition fails in video-based systems as is the case with siblings that have similar faces. We perform analyses on 10 pairs of siblings where their faces are assumed to have similarities. Our gait features are the angular displacement trajectories of walking individuals. We apply smoothing with the Bezier polynomial in our root-finding algorithm for accurate gait cycle extraction. Then, we apply classification using two different classifiers; the linear discriminant analysis (LDA) and the k-nearest neighbour (kNN). The best average correct classification rate (CCR) is 100% with a city-block distance kNN classifier. Hence, it is suggested that in the case where face recognition fails, gait may be the better alternative for biometric identification.

Gait as a biometric has received great attention nowadays as it can offer human identification at a distance without any contact with the feature capturing device. This is motivated by the increasing number of synchronised closed-circuit television (CCTV) cameras which have been installed in many major towns, in order to monitor and prevent crime. This paper proposes a new approach for gait classification with twelve different covariate factors. The proposed approach is consisted of two parts: extraction of human gait features from enhanced human silhouette and classification of the extracted human gait features using fuzzy k-nearest neighbours (KNN). The joint trajectories together with the height, width and crotch height of the human silhouette are collected and used for gait analysis. To improve the recognition rate, two of these features are smoothened before the classification process in order to alleviate the effect of outliers. Experimental results of a dataset involving nine walking subjects have demonstrated the effectiveness of the proposed approach.

Accurate hand segmentation is a challenging task in computer vision applications. We propose a new method to segment hand based on free-form skin color model. The pixel value of a person's hand is captured and represented in YCbCr color model. The CbCr color space is mapped to a CbCr plane in order to produce a clustered region of skin color. Then, instead of using ellipse to model the skin color, edge detection is performed on the clustered region to construct a free-form skin color model. The hand segmentation result, tested on various complex backgrounds gives promising results.

In this paper, we proposed a new approach for the classification of human gait features with different apparel and various walking speed. The approach consists of two parts: extraction of human gait features from enhanced human silhouette and classification of the extracted human gait features using fuzzy k-nearest neighbours (KNN). The joint angles together with the height, width and crotch height of the human silhouette are collected and used for gait analysis. The training and the testing sets are separable without overlapping. Both sets involve nine different apparel and three walking speed. From the experiment conducted, it can be observed that the proposed system is feasible as satisfactory results have been achieved.

In this paper, a new approach is proposed for extracting human gait features from a walking human based on the silhouette images. The approach consists of six stages: clearing the background noise of image by morphological opening; measuring of the width and height of the human silhouette; dividing the enhanced human silhouette into six body segments based on anatomical knowledge; applying morphological skeleton to obtain the body skeleton; applying Hough transform to obtain the joint angles from the body segment skeletons; and measuring the distance between the bottom of right leg and left leg from the body segment skeletons. The angles of joints, step-size together with the height and width of the human silhouette are collected and used for gait analysis. The experimental results have demonstrated that the proposed system is feasible and achieved satisfactory results.

In this paper, a new approach is proposed for extracting human gait features from a walking human based on the silhouette image. The approach consists of five stages: clearing the background noise of image by morphological opening; measuring the width and height of the human silhouette; dividing the enhanced human silhouette into six body segments based on anatomical knowledge; applying morphological skeleton to obtain the body skeleton; and applying Hough transform to obtain the joint angles from the body segment skeletons. The joint angles together with the height and width of the human silhouette are collected and used for gait analysis. From the experiment conducted, it can be observed that the proposed system is feasible as satisfactory results have been achieved.

Problem of poor camera calibration in Augmented Reality (AR) has always been overcome by employing careful camera calibration steps with the use of specific known object. However, this task is time consuming and mostly performed offline. A Self-Calibrated AR system (SCAR system) of a camera updates the camera intrinsic parameters whenever they change. The system solves the camera parameters based on only three views and requires only the fundamental matrices as the inputs. Point correspondence matching is one of the necessary stages in SCAR system. Correct corner matches ensure good estimation of the fundamental matrix since the outcome of this stage is to be fed into the fundamental matrix estimation stage. This paper proposes a simple and robust algorithm for point correspondence matching in one of the stages of SCAR system. The matching process is efficient and capable of removing outliers from corner detection as well as maintaining a good number of correct matches even in the event of occlusion.

Camera calibration is an essential and important part of an augmented reality (AR) system. The use of a plane-based calibration technique can give a good accuracy, which can be important for AR applications. The calibration technique used in the current ARToolKit requires user intervention, which is prone to error and involves a lengthy calibration time. The camera has to be recalibrated every time the focal length changes which is cumbersome and less suitable for applications where a more automated and easier approach is needed. This paper investigates the use of camera self-calibration for the ARToolKit, which has the advantage of simplicity of implementation. In order to improve its accuracy, a distortion model is also investigated. In this context several interesting results are presented.

Gait as a biometric has received great attention nowadays as it can offer human identification at a distance without any contact with the feature capturing device. This is motivated by the increasing number of synchronised closed-circuit television (CCTV) cameras which have been installed in many major towns, in order to monitor and prevent crime by identifying the criminal or suspect. This paper present a method to improve gait classification results by applying smoothing techniques on the extracted gait features. The proposed approach is consisted of three parts: extraction of human gait features from enhanced human silhouette, smoothing process on extracted gait features and classification by fuzzy k-nearest neighbours (KNN). The extracted gait features are height, width, crotch height, step-size of the human silhouette and joint trajectories. To improve the recognition rate, two of these extracted gait features are smoothened before the classification process in order to alleviate the effect of outliers. The proposed approach has been applied on a dataset of nine subjects walking bidirectionally on an indoor pathway with twelve different covariate factors. From the experimental results, it can be concluded that the proposed approach is effective in gait classification.