Night-Time Vehicle Distance Estimation Using Camera Geometry and Deep Learning

Vehicle distance estimation has been considered as one of the important topics in traffic video processing. There have been many algorithms proposed to deal with this problem. However, these algorithms all have their own drawbacks such as long processing duration, huge data requirements, extremely computational consumption, and low accuracy response. In this circumstance, in this paper, a novel algorithm based on a combination of deep learning and camera geometry is proposed for vehicle distance estimation. The proposed algorithm demonstrates fast processing duration which is contributed by the deep learning base technique. Moreover, thanks to the simplicity of camera geometry, the proposed algorithm requires minimal data to perform the estimation. By combining deep learning and camera geometry, the proposed algorithm can provide high estimation accuracy while keeping the model simple and fast. The performance of the proposed algorithm is verified through experiments results.