Video Object Detection based on Non-local Prior of Spatiotemporal Context

摘要

The appearances of objects in video sequence affected by complex background, motion blur and partial occlusion, which make the object detection in video sequence a hard work. Due to these problems, traditional image object detection methods cannot perform well in video sequence image. A fast and effective video object detection method is necessary to improve the detection efficiency. In this paper, we propose a non-local prior based spatiotemporal attention model based for video object detection. Unlike existing attention models, the proposed model can make full use of the spatiotemporal contextual information extracted from video sequence images. We apply our models in common object detection framework and evaluate it on Overhead Contact System (OCS) driving recorder dataset and OTB50 dataset. The proposed model achieves a greater increase in mAP value which proves our model can gains good performance in various complex video sequences.