Search

Autonomous passenger counting in public transportation represents an integral part of an intelligent transportation system, as it provides vital information to improve the efficiency and resource management of a public transportation network. However, counting passengers in highly crowded scenes is a challenging task due to their random movement, diverse appearance settings and inter-object occlusions. Furthermore, state-of-the-art methods in this domain rely heavily on additional custom cameras or sensors instead of existing onboard surveillance cameras, which consequently limits the feasibility of such systems for large-scale deployment. Hence, this paper puts forward an enhanced appearance descriptor with lateral movement compensation, which addresses the difficulty in counting passengers bidirectionally in cluttered scenes. We first construct a head re-identification dataset, which is used to train an appearance descriptor. This dataset addresses the absence of a person re-identification dataset, which in turn allows for accurate tracking of passengers in cluttered scenes. Then, a novel technique of applying a fedora counting line is introduced to count the number of passengers entering and exiting a bus. This technique compensates the impact of passengers' lateral movement, which crucially increases the accuracy of bidirectional passenger counting using onboard bus surveillance cameras. In addition, a real-time implementation of the proposed method, which includes the integration of DeepStream and fedora counting line, is also presented. Experimental results on a challenging test dataset demonstrate that the proposed method outperforms benchmarked techniques with an average counting accuracy of 93.21% for entering and 96.10% for exiting public buses. Furthermore, the proposed system achieves this accuracy at an average frame rate of 16 frames per second, which represents a practical solution to a real-time application. [ABSTRACT FROM AUTHOR]

The lack of resolution of imaging systems has critically adverse impacts on the recognition and performance of biometric systems, especially in the case of long range biometrics and surveillance such as face recognition at a distance, iris recognition and gait recognition. Super-resolution, as one of the core innovations in computer vision, has been an attractive but challenging solution to address this problem in both general imaging systems and biometric systems. However, a fundamental difference exists between conventional super-resolution motivations and those required for biometrics. The former aims to enhance the visual clarity of the scene while the latter, more significantly, aims to improve the recognition accuracy of classifiers by exploiting specific characteristics of the observed biometric traits. This paper comprehensively surveys the state-of-the-art super-resolution approaches proposed for four major biometric modalities: face (2D+3D), iris, fingerprint and gait. We approach the super-resolution problem in biometrics from several different perspectives, including from the spatial and frequency domains, single and multiple input images, learning-based and reconstruction-based approaches. Especially, we highlight two special categories: feature-domain super-resolution which performs super-resolution directly on the feature space to purposely improve the recognition performance, and deep-learning super-resolution which discusses the most recent advances in deep learning for the super-resolution task. Finally, we discuss the current and open research challenges and provide recommendations into the future for the improved use of super-resolution with biometrics. [ABSTRACT FROM AUTHOR]

Most of the time, when people observe, interact or speak to each other, they focus the attention on the ocular parts of the face. This daily life experience has a strong impact on the analysis of periocular facial regions. These facial regions may be exploited in order to identify individuals for several applications, including access control and services such as telebanking and electronic transactions. In this paper we suggest studying the efficiency of the periocular regions on gender and race prediction. Most researchers propose a local texture description based on LBP (Local Binary Pattern) and HoG (Histogram of Oriented Gradients) for the purpose of predicting gender. On the other hand, Deep learning techniques were proposed to predict the gender. However, this requires a huge labeled periocular data for gender which is not available. Also, the expressivity of gender and race can be decreased on the final representation of the Deep architectures comparing to the earlier stages. To overcome these points and for the aim of predicting gender and race, considering also the high impact of DCNNs (Deep Convolutional Neural Networks) techniques to solve several aspects in biometrics, we suggest a Deep architecture based on visual attention on the periocular part. The visual saliency extraction is based on primary layers' activation by analyzing the feature-maps. We study how the visual attention-based features coupled to Deep Neural Networks can be used to discriminate between gender and race, hence extract a significant feature from periocular regions. Different pretrained architectures such as Alexnet and ResNet-50 were considered to extract visual saliency points or interest points. Several experiments were performed on periocular regions and a comparative study was conducted. The present results not only demonstrate the feasibility but also the robustness of the extracted interest points. • Efficiency of periocular region for gender and race prediction. • Deep architectures based on visual attention on periocular regions. • Visual attention-based features coupled to Deep Neural Networks. • Visual saliency extraction based on primary layers' activation. [ABSTRACT FROM AUTHOR]

This paper presents a novel approach for Human Face Recognition, namely Regularized Bi-partitioned Entropy Component Analysis (RBECA). This conservative approach regularizes the kernel entropy components by deterring the noise and affecting the lower entropy regions area, making the method robust to noise. The kernel feature space, formed by the kernel entropy component analysis (KECA), is divided into two partitions: the High Entropy Space (HES) and the Low Entropy Space (LES). The noise-laden low entropy spectrum is regularized by predicting entropy values obtained from the information-filled High Entropy Spectrum. The corresponding projection vectors are adjusted accordingly. A null space, comprising the negligible information and many dimensions, is eliminated using a Golden Search minimization function at two stages. The method retains the maximum entropy property and high recognition accuracy while using the optimum number of features. This resultant feature vector is classified using the cosine similarity measure. The algorithm is successfully tested on several benchmark databases like AR, FERET, FRAV2D, and LFW, using standard protocols and compared with other competitive methods. The proposed method achieves much better recognition accuracy than other well-known methods like PCA, ICA, KPCA, KECA, LGBP, ERE, etc., in all considered cases. Moreover, we have also proposed a CNN for the comparative analysis. For unbiased or fair performance evaluation, the sensitivity and specificity are also reported. • A noise robust RBECA method is proposed for human face recognition. • The proposed method requires a lower number of features than KECA. • The highest level of discriminatory information is retained. • The algorithm is trained and tested on several benchmark face databases. • A deep learning CNN framework is also implemented for comparative analysis. [ABSTRACT FROM AUTHOR]