Vehicle LIDAR Recognition for Road Debris and Noise Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems for vehicles, particularly those using LIDAR, face challenges in accurately identifying fallen objects on a road surface due to noise interference near the road surface, leading to false recognition and inability to detect such objects.
Innovation Solution
The system employs a processor that classifies LIDAR point clouds into noise and non-noise spaces based on terrain map information, separates road surface and fallen object point clouds, and calculates feature quantities to extract fallen object candidates, integrating them with tracking targets when horizontal positions align, thereby reducing false recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the space near the road surface is excluded from recognition target space to avoid noise interference, then false recognition is reduced, but fallen objects on the road surface cannot be detected
Solution Approach 1:
The recognition space is segmented into noise space (near road surface) and non-noise space (above road surface) based on terrain map information. The processor classifies LIDAR point clouds into first point cloud (in noise space) and second point cloud (in non-noise space), allowing different processing strategies for different spatial regions.
Solution Approach 2:
Terrain map information serving as an intermediary is introduced to define the noise space boundary. This intermediary enables the system to distinguish between noise points and potential fallen object points by comparing LIDAR measurement information with pre-stored road surface position data.
2Measurement precision
If the space near the road surface is included as recognition target space to detect fallen objects, then fallen object detection capability is improved, but false recognition increases due to noise interference
Solution Approach 1:
The first point cloud (in noise space) is further segmented into road surface point cloud and fallen object point cloud through separation processing. This segmentation enables the system to distinguish between noise points (road surface) and potential fallen objects by analyzing feature quantities such as distance from LIDAR and position differences between adjacent points.
Solution Approach 2:
The processor calculates feature quantities for each point in the first point cloud and uses these parameter changes to separate road surface points from fallen object points. By analyzing variations in distance and position parameters, the system can identify fallen objects within the noise space while filtering out road surface noise.
3Reliability
If fallen object candidates and tracking target candidates are integrated based on horizontal position consistency, then false recognition of fallen objects is reduced, but the system complexity increases
Solution Approach 1:
The system merges the fallen object candidate extraction process with the tracking target candidate extraction process by integrating candidates whose horizontal positions are consistent. This merging allows the system to use tracking targets as reference information for verifying fallen object detections, reducing false positives through cross-validation.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the accuracy of fallen object recognition by distinguishing between fallen objects and tracking targets, suppressing false recognition and improving detection accuracy.
Implementation Method 1
A sensor such as a LIDAR (Laser Imaging Detection and Ranging) is used for recognition (perception) of an object around a vehicle
Data Source
AI summary
A vehicle includes a LIDAR. An object recognition device: sets a space having a height from an absolute position of a road surface as a noise space; classifies a LIDAR point cloud into a first point cloud included in the noise space and a second point cloud outside the noise space; extracts a fallen object candidate being a candidate for a fallen object on the road surface based on the first point cloud; extracts a tracking target candidate being a candidate for a tracking target based on the second point cloud; determines whether horizontal positions of the fallen object candidate and the tracking target candidate are consistent with each other; integrates the fallen object candidate and the tracking target candidate whose horizontal positions are consistent with each other, to be the tracking target; and recognizes the fallen object candidate not integrated into the tracking target as the fallen object.