A Fast Approach for Vehicle-Like Region Proposal Based on 3D Lidar Data

In this paper, an approach for fast vehicle-like region proposal based on LIDAR data is presented. Vehicle detection accuracy highly depends on the performance of the vehicle-like region proposal module. Traditional segmentation algorithms usually tend to produce under-segmentation and over-segmentation errors. The proposed approach adopts the sliding window strategy for fast proposing vehicle-like regions. This approach overcomes object appearance variation caused by the changes of the observed distance and viewing angles. This approach has been tested on real world lidar data and demonstrates satisfactory performance.