Violence Detection in Videos by Combining 3D Convolutional Neural Networks and Support Vector Machines

Abstract

Video-surveillance has always been a vital tool to enforce safety in both public and private environments. Even though (smart) cameras are nowadays relatively widespread and cheap, such monitoring systems lack effectiveness in most scenarios. In addition, there is no guarantee about a human operator who monitors rare events in live video footages, forcing the use of such systems after unwanted events already took their undisturbed course, as a mere tool for investigations. Having an intelligent software to perform the task would allow to unlock the full potential of video-surveillance systems. To this end, in this paper we propose a solution based on a 3D Convolutional Neural Network that can effectively detect fights, aggressive motions and violence scenes in live video streams. Compared to state-of-the-art techniques, our method showed very promising performance on three challenging benchmark datasets: Hockey Fight, Crowd Violence and Movie Violence.

A dataset for automatic violence detection in videos

“…As highlighted in a previous study [5] , violence detection techniques can fail due to actions and behaviours which are wrongly interpreted as violent, due to fast movements and similarity with violent behaviours. To this end, the non-violent clips were recorded to specifically challenge techniques and prevent false positives, even with datasets unbalanced towards the violent clips, as the one proposed in this paper.…”

“…The pervasiveness of video surveillance cameras and the need of watching footages and making decisions in a very short time [1] boosted the interest of researchers towards techniques for the automatic detection of violence and crimes in videos. In facts, both techniques based on handcrafted features [ 2 , 3 ] and deep learning [ 4 , 5 ] demonstrated their accuracy for automatic violence detection on open datasets such as the Hockey Fight Dataset [6] , the Movie Fight Dataset [6] , and the Crowd Violence Dataset [7] . However, such datasets include few low-res videos, sometimes in too specific environments (e.g.…”

Violence Detection With Two-Stream Neural Network Based on C3D

“…Another popular branch is using C3D (Tran, Bourdev, et al, 2015) for spatiotemporal feature extraction. Researchers use C3D for violence detection, such as Ullah et al (Ullah, Muhammad, 2019), Song et al (Song, Zhang, et al, 2019), Ding et al (Ding, Fan, et al, 2014), Accattoli et al (Accattoli, Sernani, et al, 2020). In addition to the aforementioned work, researchers have also focused on Graph Convolutional Network (GCN) for the detection of abnormal behavior.…”

“…The authors choose the train-test ratios in this way because these ratios are relatively common. For example, Accattoli et al (Accattoli, Sernani, et al, 2020) use a 5-fold cross validation scheme (the train-test ratio is 8:2). Note that all the other parameters in this experiment, such as learning rate and epoch etc., remain unchanged.…”

Violence Recognition Based on Auditory-Visual Fusion of Autoencoder Mapping

“…The unprecedented prosperity of deep learning has prompted scholars to try to use deep networks to identify violence. In Electronics 2021, 10, 2654 2 of 14 terms of visual features, Accattoli et al used a 3D convolutional network to detect violent behaviors [14] and Tripathi et al used a convolutional neural network to extract multi-level video features [15]. Additionally, Deniz et al proposed fast motion detection based on an extreme acceleration mode [16].…”