Object Detection and Tracking Algorithm Based on Yolov7 and Deepsort

To address the issues of distant ship misdetection and occlusion-caused miss detections in port images, as well as the low real-time detection accuracy of video information, this paper proposes an inland vessel supervision algorithm based on the fusion of image and video information, aimed at improving supervision efficiency and accuracy. First, an improved YOLOv7 ship detection algorithm is proposed, which significantly improves detection accuracy by introducing the DecIoU and Push-IoU loss functions. Next, an improved DeepSort-based ship multi-object tracking algorithm is developed, replacing Faster R-CNN with the improved YOLOv7, and combining HS-ResNet with BYTE and OCR technologies to enhance tracking performance in complex scenarios. Experimental results show that this algorithm performs excellently on the SeaShips and SMD datasets, validating the effectiveness of the improvements.