Detection Algorithm of Laboratory Abnormal Behavior Based on Two-stream Convolutional Neural Network
Lu Gao · 2024
The application of intelligent video surveillance systems can reduce the workload of workers, save human resources, and timely handle abnormal situations, reducing losses. Based on intelligent monitoring systems, research on laboratory abnormal behavior detection algorithms uses dual stream convolutional neural networks to provide security for laboratories. We need to integrate RGB images and optical flow images, and use residual structure dual flow network fusion for comparative analysis. The spatial flow channel obtains appearance information such as texture, contour, and points of interest in a single frame RGB image, while the temporal flow channel obtains motion information in optical flow images. The residual structure effectively overcomes the gradient explosion caused by deepening the network layer, prevents network degradation, uses Dropout to prevent overfitting, transfers learning and fine tunes network parameters to improve the network's generalization performance and feature acquisition ability.