Violence detection in compressed video

Online monitoring of public places could prevent the occurrence of violent actions, while the efficiency of utilized methods plays a key role in detecting abnormal events in a short time. Even though some methods have been dedicated to violence detection, all exploit raw video information. Since information is usually transmitted in a compressed format, utilizing available compressed domain data, which obviates the need for further feature extraction, is advantageous. This paper presents a compressed domain method through which residual information obtained by partial decoding is utilized as spatiotemporal features instead of other raw-domain time-consuming approaches. Using residual data enables us to save time in the laborious feature extraction step, which is one of the imperative factors in real-time applications. Furthermore, a new approach for accumulating similar adjacent residuals has been presented, which can considerably drop the number of processed frames for detection. Comparing the proposed methods with the latest state-of-the-art methods reveals their high effectiveness. The entire implementation of this work can be found at https://github.com/hnarges91/violence-detection. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.