Single Camera Visual Odometry Elevation Map Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle sensors face limitations in accuracy due to latency and inability to account for dynamic effects like tire slip, and they often fail to measure rotational changes about lateral and longitudinal axes, which are essential for precise vehicle motion analysis in six degrees of freedom.
Innovation Solution
A method using a single camera to create an elevation map through structure-from-motion and dense stereo-matching techniques, combined with feature detection and matching, to generate a reliable digital elevation map of the ground surface, enabling robust kerb detection and overcoming limitations of traditional sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional vehicle sensors (accelerometers, gyroscopes, wheel speed sensors) are used to measure vehicle motion, then the system is simple and cost-effective, but the measurement precision and reliability are limited due to latency and inability to account for dynamic effects like tire slip
Solution Approach 1:
The patent replaces mechanical sensors (accelerometers, gyroscopes, wheel speed sensors) with an optical system using a single camera and image processing algorithms. The camera captures images of the environment, and computer vision techniques analyze feature correspondences between successive images to compute vehicle motion in six degrees of freedom, eliminating the need for mechanical sensing components and their associated latency and slip-related errors
Solution Approach 2:
The patent introduces an intermediary processing system that uses image feature detection, matching, and geometric computation to bridge the gap between simple camera input and precise motion measurement. The system uses image processing algorithms as an intermediary to translate visual information into accurate six-degree-of-freedom motion data, achieving high measurement precision without complex mechanical sensors
2Measurement precision
If scaling, rotation and affine distortion invariant feature detectors (SIFT or SURF) are used to detect and match salient features, then the accuracy and repeatability of feature detection are improved, but the computational power required increases significantly
Solution Approach 1:
The patent extracts and uses only the essential features needed for vehicle motion estimation from the full SIFT or SURF feature sets. By focusing on salient features that are most relevant for computing epipolar geometry and fundamental matrices, the system achieves accurate feature matching without the full computational overhead of complete SIFT/SURF processing
Solution Approach 2:
The patent applies partial feature detection and matching - using a subset of features that are most informative for motion estimation rather than processing all detected features. This partial action approach maintains measurement precision while significantly reducing computational power consumption compared to exhaustive feature processing
3Device complexity
If a single camera is used to create elevation maps and detect kerbs, then the device complexity and cost are reduced, but the measurement precision and reliability are typically lower compared to stereo camera systems
Solution Approach 1:
The patent uses temporal dimension by analyzing feature correspondences across multiple frames captured at different time points. By tracking how features move between successive images and using temporal sequences rather than simultaneous stereo pairs, the system extracts depth and elevation information from a single camera, achieving measurement precision comparable to stereo systems with reduced hardware complexity
Solution Approach 2:
The patent exploits the dynamic nature of vehicle motion and the temporal sequence of captured images. By analyzing the dynamic changes in feature positions across multiple frames and using the known or estimated vehicle motion, the system reconstructs three-dimensional elevation information from a single moving camera, maintaining accuracy while simplifying the camera system
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present invention refers to a method for creation of an elevation map based on a single camera, comprising the steps of performing key frame selection to select at least two key frames at two different points of time out of frames provided by the camera, performing feature detection and matching based on the selected at least two key frames at two different points of time, performing coarse outlier rejection for the detected features, performing pose estimation of the camera in respect to the two key frames, performing stereo image rectification to align epipolar lines in parallel to a horizontal axis in a vertical direction, performing stereo matching to create a disparity map out of the rectified stereo image, and generating an elevation map out of the disparity map. The present invention also refers to a method for performing kerb detection based on a single camera, comprising the steps of performing creation of an elevation map according to the above method, and performing kerb detection based on the created elevation map.