CAMOT: Camera Angle-aware Multi-Object Tracking

Abstract

This paper proposes CAMOT, a simple camera angle estimator for multi-object tracking to tackle two problems: 1) occlusion and 2) inaccurate distance estimation in the depth direction. Under the assumption that multiple objects are located on a flat plane in each video frame, CAMOT estimates the camera angle using object detection. In addition, it gives the depth of each object, enabling pseudo-3D MOT. We evaluated its performance by adding it to various 2D MOT methods on the MOT17 and MOT20 datasets and confirmed its effectiveness. Applying CAMOT to ByteTrack, we obtained 63.8% HOTA, 80.6% MOTA, and 78.5% IDF1 in MOT17, which are state-of-the-art results. Its computational cost is significantly lower than the existing deep-learning-based depth estimators for tracking.

https://gitlab.com/felixlimanta/camot

References in corpus (14)

• SSD: Single Shot MultiBox Detector
• Simple Online and Realtime Tracking
• YOLOX: Exceeding YOLO Series in 2021
• FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking
• MOT16: A Benchmark for Multi-Object Tracking
• HOTA: A Higher Order Metric for Evaluating Multi-Object Tracking
• Tracking without bells and whistles
• MOT20: A benchmark for multi object tracking in crowded scenes
• TransTrack: Multiple Object Tracking with Transformer
• ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth
• Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth
• Accurate Bounding-box Regression with Distance-IoU Loss for Visual Tracking
• Semi-TCL: Semi-Supervised Track Contrastive Representation Learning
• Monocular Depth Estimation using Diffusion Models