Video Surveillance for Violence Detection Using Deep Learning

Introducing efficient automatic violence detection in video surveillance or audiovisual content monitoring systems would greatly facilitate the work of closed-circuit television (CCTV) operators, rating agencies or those in charge of monitoring social network content. In this paper we present a new deep learning architecture, using an adapted version of DenseNet for three dimensions, a multi-head self-attention layer and a bidirectional convolutional long short-term memory (LSTM) module, that allows encoding relevant spatio-temporal features, to determine whether a video is violent or not. Furthermore, an ablation study of the input frames, comparing dense optical flow and adjacent frames subtraction and the influence of the attention layer is carried out, showing that the combination of optical flow and the attention mechanism improves results up to 4.4%. The conducted experiments using four of the most widely used datasets for this problem, matching or exceeding in some cases the results of the state of the art, reducing the number of network parameters needed (4.5 millions), and increasing its efficiency in test accuracy (from 95.6% on the most complex dataset to 100% on the simplest one) and inference time (less than 0.3 s for the longest clips). Finally, to check if the generated model is able to generalize violence, a cross-dataset analysis is performed, which shows the complexity of this approach: using three datasets to train and testing on the remaining one the accuracy drops in the worst case to 70.08% and in the best case to 81.51%, which points to future work oriented towards anomaly detection in new datasets.

show abstract

“…The pseudo-optical flow was obtained by subtracting two adjacent frames [57]. Given a sequence of frames ( f 0 .…”

Section: Training Methodologymentioning

confidence: 99%

ViolenceNet: Dense Multi-Head Self-Attention with Bidirectional Convolutional LSTM for Detecting Violence

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Utilizing features through handcrafted feature descriptors and providing them to a deep classifier is another method to build deep models [18]. [19]. Authors uses pre-trained ResNet-50 architecture to extract necessary features from videos, and send them to ConvLSTM block.…”

Section: Related Workmentioning

confidence: 99%

A Skeleton-based Approach for Campus Violence Detection

Omarov¹,

Narynov²,

Zhumanov³

et al. 2022

Computers, Materials &Amp; Continua

View full text Add to dashboard Cite

In this paper, we propose a skeleton-based method to identify violence and aggressive behavior. The approach does not necessitate highprocessing equipment and it can be quickly implemented. Our approach consists of two phases: feature extraction from image sequences to assess a human posture, followed by activity classification applying a neural network to identify whether the frames include aggressive situations and violence. A video violence dataset of 400 min comprising a single person's activities and 20 h of video data including physical violence and aggressive acts, and 13 classifications for distinguishing aggressor and victim behavior were generated. Finally, the proposed method was trained and tested using the collected dataset. The results indicate the accuracy of 97% was achieved in identifying aggressive conduct in video sequences. Furthermore, the obtained results show that the proposed method can detect aggressive behavior and violence in a short period of time and is accessible for real-world applications.

show abstract

“…The test is done with 10 frames per sequence and 4 sequences per second (non-overlapping), that is, 40 FPS. The time is taken to classify a given video input of length 1 s is 0.923 s. [17] 82.9 ± 0.14% − 81.3 ± 0.21% MoSIFT+KDE+Sparse Coding [9] 94.3 ± 1.68% − 89.05 ± 3.26 Deniz et al [25] 90.1 ± 0% 98.0 ± 0.22% − Gracia et al [26] 82.4 ± 0.4% 97.8 ± 0.4% − Substantial derivative [18] − 96.89 ± 0.21% 85.43 ± 0.21% Bilinski et al [27] 93.4% 99% 96.4% Three streams + LSTM [15] 93.9% − − SELayer-3D CNN (C3D) [28] 99.0% − 98.08% 3D CNN [22] 98.3% 100% 97.0% [29] 96.33% 100% 95.71% [30] 89.0% -92.0%…”

Section: Accuracy Comparisonmentioning

confidence: 99%

CNN-BiLSTM Model for Violence Detection in Smart Surveillance

Halder

Chatterjee

2020

SN COMPUT. SCI.

View full text Add to dashboard Cite

In this paper, a lightweight computational model has been introduced for the better classification of violent and non-violent activities. Nowadays, the occurrences of violent activities get increased in public places. It has various social and economic reasons behind the growth of violent actions. Therefore, the government agencies and public administrators need to check such incidents using smart surveillance. Deep learning-based an efficient violent activity detection model can help the authorities in detecting a violent activity in real-time. The evaluated results can be henceforth sent to store and analyze the captured video to automate the crime monitoring system. Convolutional Neural Network-based Bidirectional LSTM has been used to detect violent activities and also compared with other existing approaches. Our proposed model gives 99.27%, 100% and 98.64% classification accuracies for the widely used standard Hockey Fights, Movies and Violent-Flows video datasets, respectively.

show abstract

Video Surveillance for Violence Detection Using Deep Learning

Cited by 35 publications

References 9 publications

ViolenceNet: Dense Multi-Head Self-Attention with Bidirectional Convolutional LSTM for Detecting Violence

ViolenceNet: Dense Multi-Head Self-Attention with Bidirectional Convolutional LSTM for Detecting Violence

A Skeleton-based Approach for Campus Violence Detection

CNN-BiLSTM Model for Violence Detection in Smart Surveillance

Contact Info

Product

Resources

About