The Abnormal Events Detection System by Feature Enhancement-Based Deep Learning Network
Chao-Ho Chen, Tsong-Yi Chen, Yonglin Chen, Bing-Hong Liu, Jau‐Ji Jou, Cheng-Kang Wen · 2023
To achieve real-time abnormal events detection that can assist the security personnel to manage the video surveillance system more efficient, this paper presents a feature enhancement-based deep learning network. The main strategy is to improve 2D convolutional architecture and employ the channel attention layer to enhance the important features between the layers, and combine DenseNet to strengthen the correlation between the features. Besides, a threshold-scoring strategy is designed to overcome the bad frame problem for substantially improving the accuracy of detecting abnormal events. Thus, the proposed system can effectively detect abnormal events with a higher accuracy rate than other approaches. Experimental results show that the proposed system can effectively detect abnormal events with a higher accuracy rate for various datasets, compared to other approaches, and can also achieve 33 fps (for real-time video processing).