Stereo Disparity Filtering for Accurate Road Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection devices struggle with high performance due to the complex shapes of roads and structures, leading to inaccurate obstacle detection.
Innovation Solution
A stereo camera system that generates a first disparity map, estimates the road surface shape, and removes unnecessary disparities based on the estimated shape to generate a second disparity map, enhancing object detection accuracy and performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a stereo camera captures images of the entire road surface to enable object detection, then the detection coverage is improved, but false positives increase due to road structures and complex shapes
Solution Approach 1:
The patent segments the road surface into multiple regions based on distance from the vehicle, applying different processing strategies to each region. The road surface is divided into a first region (closer to vehicle) and a second region (farther from vehicle), allowing selective disparity removal in the first region while preserving object detection capability in the second region.
Solution Approach 2:
The patent applies different quality standards and processing methods to different parts of the road surface. In the first region where road structures cause false positives, aggressive disparity removal is applied. In the second region where objects need to be detected, the original disparity information is preserved with minimal processing.
2Quantity of substance
If disparity information from the entire image is used for object detection, then detection completeness is improved, but detection accuracy deteriorates due to interference from road surface disparities
Solution Approach 1:
The patent extracts and removes disparity information corresponding to the road surface from the disparity map. By identifying pixels that represent road surface disparities and removing them, the system eliminates the harmful interference while preserving disparity information for objects that needs to be detected.
Solution Approach 2:
The patent converts the harmful road surface disparity information into a useful indicator by using it to identify and remove only the problematic disparities. The road surface disparity pattern is exploited to create a mask that selectively removes interfering information while preserving object information.
3Reliability
If the stereo camera system processes all disparity data to ensure comprehensive object detection, then detection thoroughness is improved, but processing complexity increases
Solution Approach 1:
The patent performs preliminary processing to generate a road surface disparity map before the main object detection process. By pre-identifying and removing road surface disparities in advance, the subsequent object detection operates on simplified data with reduced complexity, while still achieving thorough detection.
Solution Approach 2:
The patent introduces an intermediary processing step that creates a processed disparity map as a mediator between the raw stereo images and the final object detection. This intermediate representation separates road surface information from object information, simplifying the final detection process.
Data Source
AI summary
An object detection device is configured to execute a first process, a second process, and an object detection process (third and fourth processes). The first process estimates a shape of a road surface in a real space on the basis of a first disparity map. The first disparity map is generated on the basis of an output of a stereo camera that captures an image including the road surface, and is a map in which a disparity obtained from the output of the stereo camera is associated with two-dimensional coordinates formed by a first direction corresponding to a horizontal direction of the image captured by the stereo camera and a second direction intersecting the first direction. The second process removes from the first disparity map a disparity for which a height from the road surface in the real space corresponds to a predetermined range on the basis of the estimated shape of the road surface to generate a second disparity map. The object detection process (third and fourth processes) detects an object on the basis of the second disparity map.