Binocular Camera Long-Distance Obstacle Detection via Disparity Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Binocular stereo cameras struggle to accurately detect long-distance obstacles due to limitations in baseline length and pixel size, making it difficult to generate reliable point cloud data beyond 100 meters, which hinders the adoption of assistant driving technology in mid-low end vehicles.
Innovation Solution
A method and system that utilize a binocular camera to acquire original images, generate disparity maps, partition the target detection range into sub-ranges, calculate integral images, and create heatmaps to identify obstacles, allowing for the detection of long-distance targets by reversing the maximum matching region in the heatmap to the original image, thereby determining the obstacle's position and size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a binocular stereo camera is used for obstacle detection, then the system cost is reduced and non-metal obstacles can be detected, but the detection accuracy for long-distance targets (beyond 100 m) deteriorates
Solution Approach 1:
The patent divides the target detection range into multiple sub-ranges (e.g., 100-200m, 200-300m, etc.) and processes each sub-range separately using disparity threshold segmentation images. This segmentation allows the system to handle long-distance targets by focusing computational resources on specific distance intervals, improving detection accuracy without requiring a completely new sensor system.
Solution Approach 2:
The patent transforms the 2D image data into 3D point cloud data through disparity map generation, and further enhances this by creating heatmap data that combines spatial information with distance information. This dimensional transformation enables the system to represent long-distance obstacles more accurately in three-dimensional space, improving detection precision while using only camera hardware.
2Measurement precision
If the baseline length or pixel size of the binocular camera is increased to improve long-distance detection accuracy, then the detection precision improves, but the device complexity and manufacturing cost increase
Solution Approach 1:
The patent changes the processing parameters by introducing disparity threshold segmentation and integral image calculation. Instead of physically changing the camera hardware parameters (baseline length or pixel size), the system adjusts software processing parameters to enhance long-distance detection accuracy. This approach maintains manufacturing ease while achieving improved measurement precision through algorithmic optimization.
3Measurement precision
If additional sensors (millimeter-wave radar, laser radar) are added to improve detection capability, then the detection accuracy improves, but the system cost and device complexity increase
Solution Approach 1:
The patent makes the binocular camera system multi-functional by enabling it to perform both short-distance and long-distance obstacle detection through a unified processing algorithm. The same camera hardware, when processed with the new disparity threshold segmentation and integral image methods, can accurately detect obstacles across various distances, eliminating the need for separate sensors for different detection ranges and reducing overall system complexity.
Solution Approach 2:
The patent creates a virtual representation of the physical sensor system through computational modeling. By generating disparity maps, point cloud data, and heatmap data from the camera images, the system creates a digital copy that simulates the detection capabilities of more complex physical sensor systems. This computational copying allows the binocular camera to achieve detection accuracy previously requiring additional hardware.
Data Source
AI summary
A method and a system for detecting a long-distance target through a binocular camera, and an intelligent terminal are provided. The method includes the steps of: acquiring original images generated by the binocular camera at a target region, acquiring a disparity map in accordance with the original images, and converting the disparity map into a space information point cloud map; partitioning a predetermined target detection range into a plurality of sub-ranges in accordance with the space information point cloud map, and acquiring a disparity threshold segmentation image for each sub-range; calculating an integral image in accordance with each disparity threshold segmentation image, and acquiring a heatmap of an obstacle in accordance with the integral images; and reversely mapping a maximum matching region in the heatmap to the original image, so as to acquire a target ROI for the obstacle, thereby to acquire a position of the obstacle.