An Adaptive Threshold Method for Ground Segmentation

Aiming at the problem that the accuracy and real-time of ground segmentation cannot be guaranteed at the same time in the complex scene of lidar point cloud, an adaptive lidar ground segmentation algorithm is proposed. Firstly, the physical characteristics of lidar and the maximum slope of the road scene are used for the rough segmentation. Then, based on the ray characteristics of the lidar, the piecewise linear fitting algorithm with adaptive threshold is designed for the sub-region to screen the cloud of scenic spots. Combined with laboratory experiment platform “YouLong” for three common road scenes are analyzed, the results show that the algorithm in this paper has higher accuracy and recall rate. The single frame recall rate and accuracy stability are both above 0.82. The processing time is short, each frame data available about 15 ms, which meet the real-time requirements of driverless vehicles. The algorithm does not need to preset system threshold according to the actual situation of the road slop, which has the common use of the significance and value.