Suspicious action recognition in surveillance based on handcrafted and deep learning methods: A survey of the state of the art

Abstract

Suspicious action recognition is a captivating and testing task in the realm of surveillance. An anomaly recognition framework recognizes abnormal happenings uniquely in contrast to existing examples because any anomaly is an example that is not the same as a bunch of standard examples. Security is a fundamental need in each space, whether it is public or private. The utilization of feature extraction techniques, both from hand-crafted and deep learning methods, significantly influences the comprehensive methodology discussed in detail within this paper. This survey paper comprehensively covers multiple areas of advancements in surveillance. Starting with the importance and application of anomaly recognition in surveillance which leads to a comparison of different survey papers is also presented for reference which also includes the areas that are covered in this survey paper. Available datasets in the realm of surveillance are also explored in this survey paper leading to feature extraction methods of both handcrafted and deep learning. This paper also summarizes different methods available for suspicious action recognition in surveillance. The paper delves into the challenges faced when addressing this vital issue, presents valuable findings, and outlines limitations associated with the topic. It provides extensive analysis and ends by outlining potential future trends.