Towards Real-time Traffic Movement Count and Trajectory Reconstruction Using Virtual Traffic Lanes

In this paper, we discuss our framework and observations for AI City Challenge Track 1: Vehicle Counts by Class at Multiple Intersections. The framework we propose utilizes creating virtual traffic lanes for the movements of interest. Using a Python Graphical User Interface (GUI), the entry polygons for the movements of interest are identified. This leads to labeling the trajectories for the vehicles that have been first detected entering the region of interest via those entry polygons. Those vehicles, forming what we refer to as "virtual traffic lanes" inside the region ofinterest, are then used as identifiers for other vehicles detected further downstream using a nearest neighbors search. The framework we propose can run as an additional layer to any multi-object tracker with minimal additional computation. Our results and evaluation for the challenge track indicate the high potential of our proposed framework, and showcase the momentous value of incorporating domain knowledge in computer-vision applications.