A Deep Learning Based System for the Detection of Human Violence in Video Data

Abstract

The number of security cameras positioned within the surrounding area has expanded, increasing the demand for automatic activity recognition systems. In addition to offline assessment and the issuance of an ongoing alarm in the case of aberrant behaviour, automatic activity detection systems can be employed in conjunction with human operators. In the proposed research framework, an ensemble of Mask Region-based Convolutional Neural Networks for key-point detection scheme, and LSTM based Recurrent Neural Network is used to create a deep neural network model (Mask RCNN) for recognizing violent activities (i.e. kicking, punching, etc.) of a single person. First of all, the key-points locations and ground-truth masks of humans in an image are selected using the selected region; the temporal information is extracted. Experimental results show that the ensemble model outperforms individual models. The proposed technique has a reasonable accuracy rate of 77.4 percent, 95.7 percent, and 88.2 percent, respectively, on the Weizmann, KTH, and our custom datasets. As the proposed effort applies to industry and in terms of security, it is beneficial to society.

Behavior Analysis of User Interaction on Online Short Video Platform

“…Formula (12) shows that, the greater the value of δ i , the stronger the influence of user i in OSVP.…”

“…Galdi et al [10] defined the concept of short video, analyzed the dissemination paths and development features of OSVP information, and provided countermeasures for public opinion governance. The human computer interaction (HCI) function of OSVP can effectively boost user participation and improve viewing experience [11][12][13][14]. After analyzing the text contents of online short videos, Yonezawa et al [15] summarized the types of narration and the presentation modes of three interaction motives.…”

Video Violence Detection Using LSTM and Transformer Networks Through Grid Search-Based Hyperparameters Optimization

“…Their study had an accuracy rate of 98.8% for the Hockey dataset and 97.1% for the Crowd dataset. LSTM networks for classification were also proposed by Shoaib and Sayed [10]. ResNet 101 was adopted to extract the features of each frame.…”