Publications

The wide range of potential real-world applications (e.g. smart city, traffic management, crash detection) for the Multi- Camera Vehicle Tracking (MCVT) problem makes it a worthwhile research topic in the computer vision field.

In general, there are two approaches to address the MCVT problem: the global approach, which processes detections to create unified tracks directly, and the more commonly used two-step hierarchical approach, which involves separate stages for intracamera and inter-camera tracking. Typically, the two-step hierarchical MCVT approach can be further divided into four modules: object detection, feature extraction, single camera tracking and multi camera tracking. Each module plays a distinct role in enhancing the overall effectiveness of MCVT solutions. 

To date, there has only been limited research thoroughly examining how these modules individually affect the overall tracking performance. This paper presents an ablation study on the MCVT problem as a case study using the CityFlow V2 dataset. Using a benchmark MCVT framework, various state-of-art algorithms for each module have been implemented back-to-back to assess the impact of these algorithms. The effectiveness of these algorithms is assessed through two key metrics: IDF1 score performance and computational complexity. 

The study provides a comprehensive comparison study to understand the contributions of different algorithms in each module. Among all those modules, automatically generated spatial-temporal constraints maintains the computational efficiency while also contribute a lot on IDF1 score performance which could be the focusing point for future research on real-time real-world application

Multi-Target Multi-Camera Tracking (MTMCT) has broad applications and forms the basis for numerous future city-wide systems (e.g. traffic management, crash detection, etc.). However, the challenge of matching vehicle trajectories across different cameras based solely on feature extraction poses significant difficulties. This article introduces an innovative multi-camera vehicle tracking system that utilizes a self-supervised camera link model. In contrast to related works that rely on manual spatial-temporal annotations, our model automatically extracts crucial multi-camera relationships for vehicle matching.

The camera link is established through a pre-matching process that evaluates feature similarities, pair numbers, and time variance for high-quality tracks. This process calculates the probability of spatial linkage for all camera combinations, selecting the highest scoring pairs to create camera links. 

Our approach significantly improves deployment times by eliminating the need for human annotation, offering substantial improvements in efficiency and cost-effectiveness when it comes to real-world application. This pairing process supports cross camera matching by setting spatial-temporal constraints, reducing the searching space for potential vehicle matches.

According to our experimental results, the proposed method achieves a new state-of-the-art among automatic camera-link based methods in CityFlow V2 benchmarks with 61.07% IDF1 Score.