Treffer: Grid sampling based hypergraph matching technique for multiple objects tracking in video frames.

Multiple Object Tracking (MOT), which is a vibrant research area in computer vision, has many multi-disciplinary applications. The evolution of Machine Learning (ML) techniques has brought out a wide variety of data analytics schemes especially in video processing for IOT implementations. Recently graph powered ML techniques with its added topological capabilities have gained much importance due to their applicability in computer vision applications. Yet complexity issues remain unresolved as several graph based techniques reduce the computational complexity at the cost of accuracy. Hypergraphs (HG) with its topological and geometric features add values to the existing features which increase the accuracy with reduced complexity and also pave a way to track multiple objects simultaneously. This paper presents one such novel HG matching algorithm (objects described by modified KAZE key-points) and a feature extraction technique to process video frames with multiple objects. For ascertaining independency of HG based KAZE features Chi – square test has been conducted with its value (1.0845), observed to be much less than the tabular value (7.962) to accept the null hypothesis at 5% level of significance. Among the 8 pre-processing schemes KAZE has been found to be an apt model to suit grid sampling-based HG representation and it has been proved by computational experiments that it produces new matches between frames. The evaluation of the proposed technique is done in terms of matching accuracy, score, and processing time. Grid sampling with HG, exhibited an average better tracking performance of 81.51% (of all consecutive pairs of frames with multiple objects) than the recently reported one of 73.88% accuracy (with single object image matching) with same reduced tensor size. Moreover, the matching accuracy between every pair of consecutive video frames is observed to be consistent and lying between 77.78 and 81.51%. Results obtained from this investigation clearly indicate the superiority of the proposed algorithm over recently reported ones in the literature in terms of consistency, accuracy and matching scores and its applicability in quick abnormal event detection in surveillance videos. [ABSTRACT FROM AUTHOR]

Copyright of Multimedia Tools & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)