Lidar Point Cloud Filtering for Autonomous Vehicle Obstacle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In autonomous driving applications, lidar systems often emit laser beams that are not parallel to the horizontal plane due to installation restrictions, leading to false obstacle detection and affecting vehicle navigation.
Innovation Solution
A method and apparatus that process information by identifying obstacles from lidar point clouds, determining appearance rates and confidence degrees of grid regions, and filtering out first-time detected obstacles in low-confidence regions to reduce false detection impacts on vehicle control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the lidar is installed at a preset height above the ground to detect low obstacles, then the detection coverage is improved, but the laser beam cannot be parallel to the horizontal plane, causing false detection
Solution Approach 1:
The detection space is divided into multiple grid regions, with each grid having its own confidence degree calculated independently. This segmentation allows the system to treat different spatial zones differently, applying filtering operations selectively to low-confidence grids while preserving detections in high-confidence regions.
Solution Approach 2:
Different confidence degrees are assigned to different grid regions based on local detection statistics. The system calculates appearance rates and confidence degrees locally for each grid, enabling differentiated handling of false detections in different spatial locations rather than applying a uniform threshold globally.
2Adaptability or versatility
If the laser beam is emitted at an angle due to installation restrictions, then the installation flexibility is improved, but false obstacle detection occurs
Solution Approach 1:
The system performs preliminary statistical analysis during the detection process, calculating appearance rates of obstacles in each grid region before making final detection decisions. This preliminary action of accumulating detection data allows the system to identify and filter false detections caused by angled laser beams before they impact vehicle control.
Solution Approach 2:
The system uses feedback from repeated detections to improve reliability. By tracking the appearance rate of obstacles in each grid over time, the system can identify grids with consistently low-confidence detections (likely false detections from angled beams) and apply filtering operations based on this feedback.
3Speed
If all detected obstacles are processed for vehicle control, then the responsiveness is improved, but false detections cause incorrect control actions
Solution Approach 1:
The system extracts and filters out point clouds corresponding to obstacles detected in low-confidence grids before sending control commands to the vehicle. By taking out and removing these potentially false detections, the system prevents incorrect control actions while maintaining rapid response to genuine obstacles in high-confidence regions.
Solution Approach 2:
The system changes the parameter of confidence degree to differentiate between reliable and unreliable detections. By calculating appearance rates and deriving confidence degrees, the system transforms raw detection data into a quality-metric that enables selective processing, allowing fast response to high-confidence obstacles while filtering low-confidence false detections.
Data Source
AI summary
A method and apparatus for processing information are provided. A specific embodiment of the method includes: identifying at least one obstacle from a point cloud collected by a lidar during a traveling process of a vehicle; for an obstacle in the at least one obstacle, determining an appearance rate of the obstacle within a life cycle corresponding to the obstacle; determining a confidence degree of a grid region of at least one grid region based on appearance rates of obstacles in the at least one obstacle; determining a target grid region from the at least one grid region based on the confidence degree, and determining whether an obstacle detected in the target grid region is an obstacle detected for a first time; and if yes, filtering out a point cloud corresponding to the obstacle detected in the target grid region for the first time.


