A generic framework for monitoring local freight traffic movements using computer vision-based techniques

We address the problem of using computer vision based techniques to collect high-level traffic parameters in dynamic, real-world environments. This is a challenging problem that has not been sufficiently studied. Specifically, with a focus on freight vehicles, we propose a comprehensive framework for monitoring freight traffic movements at any well-defined freight traffic generator, using a network of road-side video cameras at strategic locations. The framework employs state-of-the-art computer vision algorithms to detect and recognize license plates. Moreover, we propose a plate merging algorithm to tackle occlusion of freight vehicles. We also propose an algorithm to match the video camera datasets across different locations to derive the operational history for individual freight vehicles. We have implemented the framework and deployed it for monitoring a large retail mall over 4 days in Singapore. The proposed framework has been used to process terabytes of videos and compute freight traffic parameters, with automatic license plate recognition of 91% accuracy under dynamic, uncontrolled video capturing conditions. With a minimal amount of additional manual processing, 98% accuracy can be achieved.