Contour-Based Vehicle Vision for Time-to-Collision Tracking
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 an affine contour matcher to align optical images by determining contours relative to an optical center, applying scale and translation vectors to align images, and calculating image velocity for object movement analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional image processing technologies are used for vehicle perception, then the system complexity is reduced, but the measurement precision and reliability of object identification and tracking deteriorate
Solution Approach 1:
The patent transforms image processing from conventional intensity-based methods to contour-based parameter extraction. By identifying and tracking contour points across frames and computing their displacement vectors, the system achieves higher measurement precision for object position and velocity without requiring complex multi-sensor fusion architectures.
Solution Approach 2:
The patent replaces complex mechanical sensor systems (radar, LiDAR) with an optical-based contour tracking system. By using image processing to extract contour information and compute object motion parameters, the system achieves comparable or superior measurement precision while reducing hardware complexity and cost.
2Device complexity
If conventional image processing technologies are used for vehicle perception, then the device complexity is reduced, but the reliability of object identification and tracking deteriorates
Solution Approach 1:
The patent implements continuous contour tracking across multiple image frames, maintaining uninterrupted observation of object boundaries. By continuously extracting contour points and computing their displacement, the system achieves reliable object tracking without the need for intermittent sensor fusion, improving both reliability and reducing system complexity.
Solution Approach 2:
The patent employs feedback mechanisms where detected contour points from previous frames are used to guide and verify detections in current frames. This temporal feedback loop enhances tracking reliability by maintaining consistency across frames and providing a reference for verifying object identification, all within a single vision system.
3Measurement precision
If contour matching and affine transformation are applied to align optical images, then the measurement precision of object movement determination is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the image processing task by first extracting contour points from objects, then separately computing displacement vectors for these contour points. This segmentation allows the application of affine transformations and contour matching specifically to the extracted contour data rather than entire images, improving measurement precision while managing computational complexity through focused processing.
Data Source
AI summary
Technologies and techniques for vehicle perception. A first contour of a current image a dna second contour of a next image are determined relative to an optical center. The current image is scaled to the next image relative to the optical center, wherein the scaling includes applying a scale vector to the first contour. A frame offset vector is determined, and the second contour is translated, based on the frame offset vector and the scale vector, to align the translated second contour to a focus of expansion. An image velocity is determined, based on the first contour and the translated second contour, wherein the image velocity is used to determine object movement from the image data.