High-precision point cloud map creation system and method in complex urban environment

Abstract

The invention discloses a high-precision point cloud map creation system and method in a complex urban environment. The system mainly comprises an odometer module, a loopback detection module, a rear-end optimization module and a map generation module. According to the main idea of the method, the odometer module divides point cloud into ground point cloud and obstacle point cloud through semanticsegmentation, bidirectional correspondence is introduced into point cloud registration, and a robust objective function based on L1 norm is constructed; loopback is effectively detected by using a global feature and local feature two-stage matching method, the global feature is used for rapidly eliminating wrong loopback, and the local feature is used for accurately screening loopback; the rear-end optimization module constructs the robust objective function based on the L1 norm so as to overcome the situation that a loop and a GPS contain external points; and the map generation module processes the ground point cloud and the obstacle point cloud separately, so that the map labeling efficiency and the online positioning robustness are improved.

Description

technical field [0001] The invention belongs to the fields of intelligent transportation and high-precision mapping systems, and in particular relates to a high-precision point cloud map creation system and method in a complex urban environment. Background technique [0002] High-precision mapping is the core module of unmanned driving. By constructing an accurate three-dimensional description of the environment in advance, the uncertainty of the unmanned online perception module can be minimized and the reliability of the entire unmanned system can be improved. On the other hand, high-precision point cloud maps are the basis for high-precision mapping. Feature-level maps such as lane lines / road boundaries / traffic light maps, and topological-level maps such as navigation maps can be further constructed through point cloud maps. [0003] Given the point cloud frame sequence collected by lidar, GPS and IMU information, high-precision mapping methods usually include two steps ...

AI Technical Summary

Problems solved by technology

However, the above two steps are affected by the urban environment and the scanning method of the laser sensor

Pose calculation usually includes odometer, loop detection and back-end optimization. The odometer uses adjacent frame point clouds to calculate the pose. Its core point cloud registration algorithm is easily affected by dynamic obstacles and occlusions. Loop detection often uses GPS to quickly Eliminate false loopbacks, but GPS signals in urban environments are easily affected by tall buildings, tunnels, and tree-lined roads. Backend optimization is based on Gaussian noise assumptions to fuse odometer, loopback detection, and GPS information, but unreliable GPS and falsely detected loopbacks will cause backend Optimization failure; in map generation, due to the complexity of the urban environment and the scanning method of lidar, the generated point cloud map has a high proportion of ground points, which affects the reliability of map labeling and online positioning algorithms

[0004] To sum up, the current mapping system cannot complete the mapping tasks in open and complex urban areas, and does not have the promotion in technology research and development and application

Images