Road Surface Estimation Using V-Disparity for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle to accurately estimate road surfaces in varying conditions, such as different weather and road gradients, leading to incorrect object detection and potential false activation of safety systems like automatic braking.
Innovation Solution
The system generates a V-Disparity map and employs a multistage estimation process to accurately estimate road surfaces by dividing the map into multiple stages, applying linear approximation to sample points, and using a least-square method to enhance road surface detection, thereby reducing false positives from road surface disparities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all disparity pixels are clustered to detect objects, then object detection coverage is improved, but false detection of road surface elements (white lines, flat areas) as objects increases
Solution Approach 1:
The patent segments the disparity image processing into two distinct stages: first generating a V-disparity map to identify road surface regions, then using this segmentation to exclude road surface pixels from object clustering. This segmentation principle separates road surface detection from object detection, preventing false positives while maintaining comprehensive object detection coverage.
Solution Approach 2:
The V-disparity map serves as an intermediary data structure that mediates between the raw disparity image and the final object detection results. By introducing this intermediate representation that explicitly models road surface geometry, the system can filter out road surface pixels before object clustering, thereby eliminating false detections without losing valid object signals.
2Measurement precision
If statistical techniques like least-square method are used to estimate road surface form, then road surface detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by first generating the V-disparity map and identifying road surface regions before performing object detection. This preliminary road surface estimation using least-square method prepares the data structure in advance, allowing subsequent object detection to proceed more efficiently by working with pre-filtered disparity pixels that have already had road surface components removed.
3Reliability
If clustering processing is applied to all disparity pixels, then comprehensive object detection is achieved, but processing time increases due to unnecessary computation on road surface pixels
Solution Approach 1:
The patent extracts and removes road surface pixels from the disparity image before performing object clustering. By taking out the road surface components that would otherwise consume computational resources during clustering, the system achieves comprehensive object detection while significantly reducing processing time by eliminating unnecessary computations on road surface areas.
Data Source
Figure 1A~1C
Figure 2
Figure 3
AI summary
According to an embodiment, an image processing device (400, 30) includes a dividing unit (411, 403), an estimating unit (413), and a selecting unit (412). The dividing unit (411, 403) is configured to divide, in a frequency image associating positions in a vertical direction of a distance image with regard to an object and a distance value at each of the positions and representing a frequency distribution of the distance values, a predetermined division in a direction of the distance value into one or more divisions in accordance with multiple dividing patterns having a different number of divisions. The estimating unit (413) is configured to estimate a form of a road surface in each of the divisions at each of the dividing patterns based on each frequency point of the frequency image. The selecting unit (412) is configured to select at least a final estimated road surface in the predetermined division based on an estimated road surface estimated by the estimating unit (413) at each of the dividing patterns and in each of the divisions.