Video Significance Object Detection Model Based on Spatiotemporal Progressive Learning

Video salience object detection needs to combine spatial and temporal information to continuously locate the moving-related salience objects in video sequences. The core problem is how to efficiently depict the spatio-temporal characteristics of moving objects. Most of the existing video salience object detection algorithms use optical flow, ConvLSTM and 3D convolution to extract time domain features, which lack the continuous learning ability of time information. Therefore, a robust spatial-temporal progressive learning network (STPLNet) is designed to efficiently locate salient objects in video sequences. In the space domain, a U-shaped structure is used to encode and decode each video frame. In the time domain, by learning the main body part and deformation region features of the moving target between frames in the video sequence, the moving target features are gradually encoded, and the time correlation features and motion tendency of the target can be captured.