Stereo Image Parallax Correction for Road Obstacle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle to accurately distinguish between road surface paint and objects on the road surface, such as shadows or obstacles, due to vertical offset errors in stereo images, leading to misinterpretation and inaccurate detection.
Innovation Solution
An image processing device that extracts specific regions on the road surface, estimates road surface parameters, and corrects parallax errors using ideal parallax and vertical offset calculations to differentiate between road surface features and obstacles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature point matching is used to detect vertical offset, then parallax can be calculated, but in environments with few features or uniform surfaces, corresponding points cannot be detected
Solution Approach 1:
The patent introduces road surface paint as an intermediary reference object to detect vertical offset. Instead of directly matching features between left and right images, the system uses road surface paint (which has distinct color and pattern characteristics) as a mediator to establish correspondence and calculate vertical offset, thereby solving the problem of insufficient features in uniform environments
Solution Approach 2:
The patent exploits color differences as a key feature to identify road surface paint and establish correspondence between images. By detecting color changes and patterns specific to road surface paint, the system can reliably identify corresponding points even in environments lacking other distinctive features
2Measurement precision
If feature point matching is used to detect vertical offset, then parallax can be calculated, but in environments with many candidate corresponding points, it is difficult to specify the true corresponding point
Solution Approach 1:
The patent applies local quality by focusing on specific regions with distinctive road surface paint characteristics rather than attempting to match all features globally. By identifying local regions with unique color patterns and textures of road surface paint, the system can reliably determine corresponding points even when multiple candidates exist
3Ease of manufacture
If parallax error is not corrected, then processing is simpler, but road surface paint and shadows are misinterpreted as obstacles
Solution Approach 1:
The patent performs preliminary action by calculating and correcting parallax error before conducting obstacle detection. By first establishing accurate vertical offset and parallax values using road surface paint as reference, the system prepares the image data in advance to ensure reliable distinction between road surface paint, shadows, and actual obstacles
Data Source
AI summary
Provided is an image processing device that, by performing highly accurate correction of a parallax error caused by a vertical offset between left and right cameras, makes it possible to accurately distinguish between a control object such as an object dropped on a road surface and road surface paint, and accurately detect a control object having a low step. The image processing device includes a specific region extraction unit that extracts a specific region on a road surface from a plurality of images captured by each of a plurality of in-vehicle cameras; a road surface parameter estimation unit that estimates a road surface structure represented by a road surface parameter, from a road surface portion other than the specific region, the road surface portion being a road surface in the image; a specific region parallax estimation unit that obtains, from the road surface structure, an ideal parallax to be obtained in the specific region; a parallax image generation unit that obtains a parallax from the plurality of images; and a vertical offset calculation unit that obtains a vertical offset amount of the plurality of images by using an actual parallax of the specific region, the ideal parallax of the specific region, and arrangement information in the image of the specific region.