DETRAC: A New Benchmark and Protocol for Multi-Object Tracking

In recent years, most effective multi-object tracking (MOT) methods are based on the tracking-by-detection framework. Existing performance evaluations of MOT methods usually separate the target association step from the object detection step by using the same object detection results for comparisons. In this work, we perform a comprehensive quantitative study on the effect of object detection accuracy to the overall MOT performance. This is based on a new large-scale DETection and tRACking (DETRAC) benchmark dataset. The DETRAC benchmark dataset consists of 100 challenging video sequences captured from real-world traffic scenes (over 140 thousand frames and 1.2 million labeled bounding boxes of objects) for both object detection and MOT. We evaluate complete MOT systems constructed from combinations of state-of-the-art target association methods and object detection schemes. Our analysis shows the complex effects of object detection accuracy on MOT performance. Based on these observations, we propose new evaluation tools and metrics for MOT systems that consider both object detection and target association for comprehensive analysis.