Stereo Disparity Hazard Detection for Non-Flat Road Surfaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hazard detection technologies for autonomous vehicles are inadequate due to reliance on inaccurate assumptions, high cost, and inefficiency in detecting small hazards at a distance, particularly when road surfaces are not flat, and require extensive training data.
Innovation Solution
Utilizing stereo disparity between a pair of images to generate a baseline road disparity model, identifying hazards from deviations in the disparity map, and applying blockwise division and clustering algorithms to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR sensors are used to detect hazards, then detection capability is improved, but system cost increases significantly
Solution Approach 1:
The patent uses stereo camera systems to capture image pairs that serve as optical copies of the road scene, replacing expensive LiDAR sensors. The disparity information extracted from these image copies provides depth perception similar to LiDAR but at much lower cost, making the system economically practical for consumer vehicles
Solution Approach 2:
The patent replaces the mechanical LiDAR scanning system with an optical stereo vision system. Instead of using mechanical lasers to measure distance, the system uses passive optical cameras to capture images and computes depth through disparity analysis, eliminating moving mechanical parts and reducing cost
2Device complexity
If parallax-based image processing is used for hazard detection, then system cost is reduced, but detection accuracy deteriorates on non-flat roads
Solution Approach 1:
The patent segments the road surface into multiple local planar regions rather than assuming a single flat plane. By dividing the road into several segments and fitting planes to each, the system can accurately model non-flat road surfaces while maintaining the simplicity and low cost of image-based processing
Solution Approach 2:
The patent changes the parameter representation from a single global plane assumption to multiple local plane parameters. This allows the system to adapt to varying road geometries by adjusting the plane parameters for different regions, improving accuracy on hills and curves while keeping computational requirements manageable
3Reliability
If DNN inference engine is used to detect hazards, then detection robustness is improved, but training data requirements and system cost increase
Solution Approach 1:
The patent enables the system to generate its own training data by using the disparity model to automatically identify hazard candidates in real-world driving scenes. These detected hazards can be labeled and used to train the DNN, creating a self-sustaining system that reduces dependency on manually collected training data and lowers overall system cost
4Measurement precision
If stereo disparity analysis is used to detect hazards, then detection precision is improved, but computational complexity increases
Solution Approach 1:
The patent segments the image processing into distinct stages: disparity map generation, road disparity model creation, and hazard detection. By dividing the computational task into manageable segments and processing only relevant regions, the system achieves high detection precision while controlling computational complexity for real-time operation
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
Provides robust and precise hazard detection by distinguishing road disparities from hazard pixels, enabling effective navigation around obstacles, reducing false positives, and optimizing computational resources.
Implementation Method 1
a left image frame and a right image frame captured by a stereo camera system of the ego-machine
Data Source
AI summary
In various examples, system and methods for stereo disparity based hazard detection for autonomous machine applications are presented. Example embodiments may assist an ego-machine in detecting hazards within its path of travel. The systems and methods may use disparity between a stereo pair of images to generate a baseline path disparity model and further identify hazards from detected disparities that deviate from that path disparity model. A disparity map for the image pair is constructed in which each pixel represents a disparity for a corresponding element of the image captured. Blockwise division may be optionally used to subdivide the disparity map into a plurality of smaller disparity maps, each corresponding to a block of pixels of the disparity map. A V-space disparity map, where a first axis corresponds to disparity values and the second axis corresponds to pixel rows, may be used to simplify estimation of the path disparity model.