Vehicle LiDAR Contour Separation for Adjacent Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle lidar systems often misrecognize two adjacent objects as a single object due to insufficient separation distance, which can compromise the accuracy of object detection and threaten driver safety in autonomous driving scenarios.
Innovation Solution
The vehicle lidar system employs an object detection method that analyzes contour characteristics to detect the connection point between two objects and separates them by setting a separation reference line, ensuring accurate recognition of distinct objects based on peak points and contour data analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a simple object recognition method is used, then the processing speed is fast, but the accuracy of distinguishing adjacent objects deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the contour of a potentially merged object into multiple segments based on point density thresholds. The contour is split at regions where point density falls below a reference value, creating separate segments that correspond to distinct objects. This allows the system to accurately separate adjacent objects while maintaining efficient processing.
Solution Approach 2:
The patent implements local quality analysis by evaluating point density at different locations along the contour. Regions with low point density are identified as separation candidates, while dense regions are maintained as object boundaries. This localized assessment enables accurate object separation without requiring complex global processing.
2Device complexity
If adjacent objects are treated as a single object, then the processing complexity is reduced, but the object detection accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-identifying potential separation regions in the contour before final object classification. The system first detects low-density regions and marks them as separation candidates, then proceeds with object recognition based on these pre-identified boundaries. This preliminary segmentation simplifies subsequent processing while ensuring accurate object detection.
3Measurement precision
If the separation threshold is set too low, then more objects are separated, but false separations increase
Solution Approach 1:
The patent implements parameter changes by dynamically adjusting the point density threshold based on reference values derived from the overall point cloud characteristics. The system compares local point density against a reference threshold to determine separation candidates, allowing adaptive separation that reduces false positives while maintaining precision for genuine object boundaries.
Data Source
AI summary
An object detection method of a vehicle lidar system according to an embodiment includes, in a contour of an object to be separated formed by connecting peak points among point data of the object to be separated, determining a contour line in which a connecting section of a contour line connecting the peak points includes point data equal to or less than a reference, and setting a separation reference line for separating the contour of the object to be separated into two regions while passing through the contour line including the point data equal to or less than the reference to recognize point data of the two regions as respective objects.