Improved Feature Extraction and Similarity Algorithm for Video Object Detection

Video object detection is an important research direction of computer vision. The task video to detect and classify moving objects in a sequence images. Based on the static image detector, most existing methods use unique temporal correlation solve problem missed false caused by occlusion blur. Another model guided optical flow network widely used. Feature aggregation adjacent frames performed estimating field. However, there are many redundant computations for feature frames. To begin with, this paper improved Faster RCNN Pyramid Dynamic Region Aware Convolution. Then S-SELSA module proposed from perspective semantic similarity. similarity obtained modified SSIM algorithm. can aggregate features globally avoid redundancy. Finally, experimental results ImageNet VID DET datasets show that mAP method 83.55%, which higher than methods.