A Motion-based Seq-bbox Matching Method for Video Object Detection

Actual video object detection performance should be improved because of the blurring and occlusion defor-unation of objects. Current algorithms, such as FGFA and Seq-NMS, cannot achieve a tradeoff between speed and accuracy in practical applications simultaneously. This study aims to propose a practical video object detection algorithm to improve the accuracy of detection and guarantee real-time detection. In this work, a post-processing method that is based on a single-frame detection algorithm and is called motion-based Seq-Bbox matching is proposed, and an inter-frame motion information is introduced to enhance the detection results. Distance Intersection over Union is utilized to represent the inter-frame motion information, and the idea that the same object between adjacent frames should have similar motion information is proposed. Moreover, the dynamic confidence averaging method is combined to complete the enhancement of the prediction results jointly. Based on YOLOv5, experimental re-sults show that the proposed algorithm achieves 72.7% mean average precision (mAP) and obtains a 5.5% mAP (67.2%-72.7%) improvement, while detection speed reaches 40.4 frames per second. Thus, the proposed algorithm achieves excellent results in terms of balancing speed and accuracy.