Data-driven online traffic reconstructions: Interactively optimizing in virtual reality

Abstract

It is important to realize three-dimensional (3D) online traffic reconstructions based on the video data. Existing methods always use structured data to generate 3D traffic reconstructions and do not process unstructured pixel data. In this paper, we propose a 3D online traffic reconstruction method driven by video data and virtual traffic simulations. There are mainly two steps in our method. The first step is data preprocessing. We extract vehicles’ trajectories from video data and transform them into the geodetic coordinate system. The transformation parameters are optimally solved in multiple human–computer interactions. The second step is trajectory reconstructions. We use the traffic simulation method to predict vehicles’ trajectories and use the Kalman filter to optimize vehicles’ trajectories according to the extracted trajectories and the predicted trajectories. We also optimize the traffic simulation parameters iteratively according to the optimized trajectories when needed. The experimental results show that our method can provide realistic and smooth traffic reconstructions. It has nearly the same efficiency as that of existing online multiple object tracking methods.