Stereo Camera Object Recognition Using Disparity Map Surface Angles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems for vehicles face challenges in accurately rejecting non-target objects, such as side walls, due to the reliance on pre-defined models, leading to high calculation costs and potential incorrect tracking when objects differ from the templates, especially when recognizing various side wall shapes on different roads.
Innovation Solution
The system employs block matching processing and clustering techniques to derive disparity values from stereo camera images, generating U-Disparity maps and V-Disparity maps to identify and reject non-target objects by analyzing surface features and angles, thereby improving the accuracy of object recognition and reducing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If pre-defined models are used for object recognition, then recognition speed is improved, but recognition accuracy deteriorates when objects differ from templates
Solution Approach 1:
The patent changes the recognition parameters from template-based identification to geometric feature-based identification. By extracting surface angles and shapes directly from disparity maps and comparing them against geometric criteria rather than pre-defined templates, the system maintains high recognition speed while achieving accurate identification of varied objects including those not in the template database.
2Measurement precision
If multiple models are prepared to recognize various side wall shapes, then recognition accuracy is improved, but calculation cost increases
Solution Approach 1:
The patent extracts only the essential geometric features (surface angles and shapes) from objects in the disparity map, rather than comparing against multiple complete models. This extraction approach reduces calculation cost by focusing only on the critical distinguishing features needed for accurate recognition of side walls and other objects.
Solution Approach 2:
The system changes from model-based recognition requiring multiple stored models to parameter-based recognition using surface angles and shapes. This parameter transformation eliminates the need for storing multiple models while maintaining recognition accuracy for various side wall shapes and configurations.
3Device complexity
If clustering processing is used to group pixels by disparity, then object detection is simplified, but incorrect recognition of non-target objects occurs
Solution Approach 1:
The patent segments the object recognition process into distinct stages: first clustering pixels by disparity to form candidate regions, then separately analyzing the geometric features (surface angles and shapes) of each candidate. This segmentation allows simple clustering to be combined with more sophisticated geometric analysis, improving recognition reliability while keeping the overall system manageable.
Solution Approach 2:
The patent introduces geometric feature analysis as an intermediary step between simple clustering and final object identification. This intermediary analysis of surface angles and shapes acts as a filter that distinguishes target objects from non-target objects like side walls, improving recognition reliability without significantly increasing system complexity.
Data Source
Figure 1~2(b)
Figure 3
Figure 4(a)~4(b)
AI summary
There is provided: an input unit configured to receive a recognized image representing an object; a detecting unit configured to detect at least one surface of the object from a corresponding recognized image corresponding to the recognized image, the corresponding recognized image being on a frequency image indicating frequencies on a plane, the frequencies being of distance values corresponding to pixels of the recognized image; a calculating unit configured to calculate an angle of the at least one surface with respect to a predetermined direction; and a determining unit configured to determine, based on the at least one surface detected by the detecting unit and the angle calculated by the calculating unit, whether or not the recognized image is to be rejected.