Fight detection with spatial and channel wise attention-based ConvLSTM model

• 全文链接
• 请求原文

作   者：

Kunal Chaturvedi;  Chhavi Dhiman;  Dinesh Kumar Vishwakarma; 

作者机构：

New Delhi;  Delhi;  India;  Departmentof Information and Technology;  India||Department of Electronics andCommunication Engineering;  India;  Biometric Research Laboratory;  DelhiTechnological University; 

关键词：

deep learning;  LSTM;  fight detection;  attention;  convolutional neural networks;  violence; 

期刊名称：

Expert systems: The international journal of knowledge engineering

i s s n：

0266-4720

年卷期：

2024 年 41 卷 1 期

页   码：

e13474.1-e13474.11

页   码：

摘   要：

An automated detection of aggressive and violent behaviour in videos has immensepotential. It enables efficient online content filtering by restricting access to extremecontent and also, when integrated with security systems, helps to monitor violence insurveillance videos. In this work, a convolutional neural network is combined withthe proposed Spatial and Channel wise Attention-based ConvLSTM encoder(SCan-ConvLSTM). The proposed architecture performs an efficient spatiotemporalfusion of the features extracted from the video sequences containing fight scenes. Inorder to focus selectively on regions of utmost importance, this blended attentionmechanism adjusts the weights of outputs in different locations and across differentchannels. This recurrent attention mechanism enhances the sequential refinement ofactivation maps and boosts the model performance. Finally, the experimental resultshave been presented that show the proposed architecture achieves superior resultson the benchmark datasets (RWF-2000, Violent-flow, Hockey-fights, and Movies).