A comprehensive analysis of real-time video anomaly detection methods for human and vehicular movement

The ubiquitous integration of computer vision technologies has revolutionised various application domains, ranging from agriculture and transportation to healthcare and the military. In this context, real-time anomaly detection in video streams has emerged as a critical facet, particularly concerning the movement of vehicles and humans. Despite the increasing adoption of these technologies in real-life applications, a conspicuous research gap exists, the necessity for robust validation of benchmarked datasets and state-of-the-art methods before deployment in real-time settings. This paper addresses this gap through a holistic approach to extensively review the real-time video anomaly detection literature in Human and Vehicular Movement applications. Specifically, we examine the various architectures used for real-time video anomaly detection and provide a detailed analysis of their performance on benchmarked datasets. Furthermore, we delve into the multifaceted challenges encountered when identifying anomalies across different categories, encompassing issues related to computational resources, noise interference, frame complexity, data sparsity, diversity, ethical considerations and their combinations. We address these challenges by proposing potential future trends in anomaly detection, including the development of more sophisticated datasets, exploration of hybrid architectural models, integration of multi-camera systems, advancement in open-set action recognition algorithms, model development for cutting-edge process automation systems, harnessing the potential of IoT, leveraging edge computing, and incorporating drone technologies into anomaly detection strategies. This paper aims to illuminate the direction of real-time anomaly detection for future research in vehicular and human movement.