Multi-Camera Vehicle Distance Estimation Without Visual Odometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles with multiple cameras face errors in distance measurement due to external forces and topographic features, leading to misrecognition of objects and reliance on complex visual odometry processing.
Innovation Solution
A method for vehicles with multiple cameras that performs automated online calibration and vehicle dynamic compensation without relying on visual odometry, using a controller to process images, determine necessary calibration, and estimate distances based on 3D coordinate values and inertial measurement unit data, while storing characteristic relationships and parallax information for accurate object recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual odometry processing is used for automated online calibration and vehicle dynamic compensation, then calibration accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent extracts and removes the visual odometry processing component from the calibration system. Instead of using complex VO processing, the invention directly uses image processing to obtain object characteristics (height, aspect ratio, distance) and establishes characteristic relationships between these parameters, thereby eliminating the computational complexity of VO while maintaining calibration functionality
Solution Approach 2:
The patent substitutes the mechanical/computational visual odometry system with a simplified image processing approach. By using direct image processing to extract object characteristics and establish parametric relationships, the system replaces the complex VO computational mechanism with a more efficient image-based measurement approach
2Reliability
If multiple cameras are used for object recognition, then recognition accuracy is improved, but errors in distance measurement increase due to topographic features and external forces
Solution Approach 1:
The patent implements feedback by establishing characteristic relationships between object parameters (height, aspect ratio, distance) obtained from multiple cameras. The system uses these relationships to verify and correct distance measurements, allowing the multi-camera system to maintain high recognition accuracy while compensating for measurement errors through parametric consistency checks
Solution Approach 2:
The patent makes the object characteristic parameters (height, aspect ratio, distance) serve multiple functions simultaneously. These parameters are used both for object recognition identification and for distance measurement calibration, allowing the same data from multiple cameras to fulfill dual purposes and reduce the impact of individual measurement errors
3Extent of automation
If visual odometry processing is performed, then calibration can be achieved, but the process becomes complex and error-prone
Solution Approach 1:
The patent extracts and removes the visual odometry processing step from the automated calibration pipeline. The system achieves automated calibration directly through image processing of object characteristics and establishing parametric relationships, eliminating the intermediate VO processing stage that adds complexity and potential error sources
Data Source
AI summary
A vehicle includes a first camera to obtain a first image; a second camera to obtain a second image captured in a different field of view; and a controller to obtain a distance between the vehicle and an object by processing images obtained by the multiple cameras. The controller recognizes a first object in a frame of the image, obtains a height of, an aspect ratio of, and a distance from the first object, and assigns each to 3D coordinate values. The controller generates a 3D straight line based on a plurality of 3D coordinate values in each frame, stores a characteristic relationship between the recognized first object and the 3D straight line, obtains a height and aspect ratio of a second object by image processing upon recognizing the second object of the same model as the first object through the cameras, and estimates a distance from the second object based on the characteristic relationship.