Camera Calibration via Motion Vector Depth Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection systems for motor vehicles require complex analysis and are not efficient in detecting elongated environmental features alongside the road, such as walls, kerbs, ditches, vegetation, or standing vehicles, which can bias calibration and increase computational effort.
Innovation Solution
A method using block matching and motion vectors from consecutive images captured by vehicle-side cameras to detect objects based on depth information, where the presence of elongated features is identified by analyzing differences in motion vector lengths without requiring classification or characterization, allowing for fast and reliable detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex analysis of 3D point cloud combined with image-brightness edge information is used to localise and characterise kerbs, then measurement precision of kerb localisation is improved, but device complexity and computational effort increase significantly
Solution Approach 1:
The patent extracts only the essential depth information needed for object detection from the image data, rather than performing complete 3D point cloud analysis. By using motion vectors between consecutive images to calculate depth, the system extracts only the necessary geometric information without full scene reconstruction, thereby reducing computational complexity while maintaining sufficient precision for detecting elongated objects alongside the road.
Solution Approach 2:
Instead of directly analyzing complex 3D point cloud data, the patent creates a simplified representation by computing depth maps from 2D image sequences. The depth information is derived as a copy or projection from motion analysis between frames, allowing object detection to proceed on this simplified depth representation rather than the full complex 3D data, reducing computational burden while preserving essential spatial relationships.
2Loss of information
If full classification and characterisation of detected objects is performed, then object understanding is improved, but processing time and computational effort increase
Solution Approach 1:
The patent applies different processing levels to different objects based on their relevance. For elongated objects alongside the road (kerbs, walls, ditches, vegetation, standing vehicles), only presence detection using depth information is performed. Full classification and characterisation is not applied to all objects, but only to those requiring detailed analysis. This local differentiation of processing quality reduces overall computational effort while maintaining necessary information for safety-critical detections.
3Device complexity
If conventional object detection methods are used without depth information, then device complexity is reduced, but detection reliability and accuracy decrease
Solution Approach 1:
The patent transitions from 2D image analysis to 3D-aware detection by incorporating the depth dimension through motion vector analysis. By calculating depth information from the displacement of features between consecutive images, the system adds the third dimension (depth) to the detection process. This dimensional enhancement significantly improves detection reliability for objects alongside the road, as depth provides critical spatial context that 2D images alone cannot provide, while maintaining relatively simple system architecture.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables efficient and reliable detection of objects alongside the road with low computational effort, improving camera calibration and reducing errors caused by elongated features, thereby enhancing the accuracy of driver assistance systems.
Implementation Method 1
at least two images of an environmental region of the motor vehicle consecutively captured by at least one vehicle-side camera
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for detecting an object (12) alongside a road (10) of a motor vehicle (1) based on at least two images (13) of an environmental region (9) of the motor vehicle (1) consecutively captured by at least one vehicle-side camera (4) for extrinsic calibration of the at least one camera (4), wherein the images (13) at least partially display a texture of a road surface (11) and wherein the following steps are performed: a) determining at least two mutually corresponding blocks (14) based on the at least two images (13); b) determining respective motion vectors for each of the at least two pairs of mutually corresponding blocks (14); c) determining a depth information concerning the at least two images (13) based on the at least two motion vectors; d) detecting the object (12) based on the depth information. The invention also relates to a computing device (3), a driver assistance system (2) as well as a motor vehicle (1).