Contour-Based Vehicle Vision Tracking for Time-to-Collision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing technologies lack the necessary accuracy and robustness for vehicle perception, often requiring combination with other sensors like radar and LiDAR for object identification and tracking, and struggle to determine object positions relative to a vehicle with precision.
Innovation Solution
A sensor processing system that includes a contour extractor and projective contour matcher to determine mapping offsets, image velocity vectors, and object movement by aligning grid points between current and next image maps, enabling accurate object tracking and position determination using image data alone.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing technologies are used for object tracking, then the system is simpler and costs less, but the accuracy and robustness are insufficient
Solution Approach 1:
The patent segments the object boundary into multiple contours extracted from sequential image frames. By dividing the tracking task into contour extraction, matching, and flow calculation components, the system achieves higher precision in object position determination while maintaining computational efficiency
Solution Approach 2:
The patent transitions from 2D spatial image data to 3D space-time representation by incorporating temporal dimension. Contours are tracked across multiple frames to create space-time trajectories, enabling more accurate object position determination and velocity calculation through temporal analysis
2Measurement precision
If multiple sensors (radar, LiDAR) are combined for object identification and tracking, then the accuracy and robustness improve, but the device complexity and cost increase
Solution Approach 1:
The patent makes the optical sensor system multi-functional by extracting not only object position information but also velocity, acceleration, and space-time trajectory data from the same image sequence. This single sensor performs multiple measurement functions that traditionally required multiple different sensors
Solution Approach 2:
The patent creates a computational model (contour cloud) that replicates the spatial-temporal behavior of physical objects. By matching contours across frames and calculating flow vectors, the system generates virtual representations of object motion that provide the same information as physical radar or LiDAR systems would deliver
3Measurement precision
If conventional image processing is used, then the system complexity is lower, but the ability to determine object positions relative to vehicle with precision is insufficient
Solution Approach 1:
The patent implements feedback through iterative contour matching between sequential frames. The system continuously compares current frame contours with previous frame contours, adjusts matching parameters, and refines flow vector calculations to progressively improve position determination precision
Solution Approach 2:
The patent performs preliminary contour extraction and matching operations on sequential image frames before final position calculation. By pre-processing the image data to identify and track contours across frames, the system prepares refined position information that improves subsequent relative position determination accuracy
Data Source
AI summary
Technologies and techniques for vehicle perception. A contour extractor apparatus extracts contours from image data. Mapping offsets are determined between at least some of the grid points of a current image map and respective grid points of a next image map, and image velocity vectors are determined for at least some of mapping offsets. At least some of the grid points of the next image are warped to the respective grid points of the current image map. A determination is made if a contour of the warped grid points of the next image matches a contour of the respective grid points of the current map within a configured parameter. Object movement movement may then be determined from the image data, if the contour of the warped grid points of the next image matches the contour of the respective grid points of the current map within the configured parameter.