Bounding Point Reliability Estimation for Autonomous Vehicle Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems face challenges in accurately determining the shape of objects, leading to unreliable object detection and tracking, particularly in estimating the position and size of objects, which affects response to object interactions and collision reliability.
Innovation Solution
A method and system that estimate the reliability of bounding points of a track by associating LiDAR contour information with bounding points within a track box, determining scores based on distance and zone analysis, and interpolating scores for corner points, to improve shape estimation and object tracking accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only object detection and tracking techniques focusing on presence are applied, then object presence detection is reliable, but shape estimation accuracy and position reliability deteriorate
Solution Approach 1:
The patent segments the track box into multiple bounding points (corners and side midpoints) and associates each with LiDAR contour points. This segmentation allows independent evaluation of each bounding point's reliability based on its association strength with LiDAR data, resolving the contradiction between simple presence detection and precise shape estimation.
Solution Approach 2:
The patent assigns different reliability scores to different bounding points based on their local association with LiDAR contour points. Not all bounding points are treated equally; corner points and side midpoint points receive different scores based on their geometric role and LiDAR association strength, enabling precise shape estimation while maintaining overall detection reliability.
2Ease of manufacture
If track box dimensions and heading angle are predicted from class information, then processing is simple, but length, width, and heading angle accuracy deteriorate
Solution Approach 1:
The patent introduces LiDAR contour points as an intermediary between the simple class information and the precise track dimensions. The LiDAR data serves as a mediator that bridges the gap between computational simplicity and measurement precision, allowing accurate dimension estimation without complex processing.
3Device complexity
If bounding point reliability is not estimated, then system complexity is low, but response to cut-in and cut-out and collision position accuracy deteriorate
Solution Approach 1:
The patent performs preliminary action by pre-defining multiple bounding points at specific locations (corners and side midpoints) and pre-establishing the association methodology with LiDAR contour points. This preliminary setup enables reliable collision position estimation without adding complex real-time computation, as the framework is prepared in advance.
Data Source
AI summary
A method for estimating reliability of a bounding point of a track includes extracting bounding points from a track box of a target object, determining association between the bounding points and at least one LiDAR contour point based on a distance between the track box and the at least one LiDAR contour point, determining scores of the bounding points based on the association, and estimating reliability of the bounding points based on the scores of the bounding points.