Fod Tracking Algorithm Based on Detection Box Matching

The intelligent transportation system aims to build a cloud-edge-end environment based on 5G technology to achieve autonomous driving. Achieving unmanned driving requires the application of computer vision techniques to detect and track targets using the road condition information obtained through 5G cameras. Multi-object tracking (MOT) aims to estimate the bounding boxes and identities of objects in videos. Most methods typically obtain the identities of objects by associating detection boxes with scores higher than a certain threshold. However, objects with low detection scores, such as occluded objects, are often discarded, leading to instances of true object missing and fragmented trajectories. Particularly in small object tracking, many small object detectors often yield lower scoring detection boxes. To address this issue, this paper proposes a new data association algorithm based on BYTE's data association algorithm, which no longer uses IoU to associate the detection boxes of small objects, but instead utilizes a new detection box matching algorithm. The proposed detection box algorithm achieves a MOTA score of 70.8, an IDF1 score of 75.3, and an IDS of 186, ranking it at the top compared to existing tracking algorithms.