Depth Maps With Confidence Estimates for Vehicle Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereo vision systems lack the ability to provide reliable depth estimates on a pixel-by-pixel or frame-by-frame basis, leading to potential errors in autonomous vehicle decision-making due to uncertainty in sensor data.
Innovation Solution
The system integrates a vehicle control system with sensor systems that output disparity maps and confidence maps, allowing for the fusion of data from various sensors to enhance the reliability of depth information and improve decision-making in autonomous vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If stereo vision systems provide depth estimates without confidence indicators, then the system complexity is reduced, but the reliability of autonomous vehicle decision-making deteriorates
Solution Approach 1:
The patent segments the depth estimation output into multiple components: depth values, confidence maps, and quality indicators. Each pixel's depth estimate is accompanied by its own confidence score, allowing the system to provide detailed reliability information without requiring a complete system redesign. This segmentation enables selective use of depth data based on confidence levels.
Solution Approach 2:
The patent introduces confidence maps as an intermediary layer between the stereo vision system and the autonomous vehicle control system. These confidence maps serve as mediators that translate raw depth estimation uncertainty into actionable quality metrics, enabling the control system to make informed decisions about which depth data to trust and how to weight different sensor inputs.
2Reliability
If confidence maps are added to depth estimates, then the reliability of depth information is improved, but the loss of information increases due to additional data processing
Solution Approach 1:
The patent implements feedback mechanisms where confidence map information flows back into the sensor fusion process. The confidence levels of depth estimates are used to dynamically adjust the weighting of different sensor inputs, creating a closed-loop system that continuously optimizes the fusion based on the quality of available data. This feedback prevents information loss by ensuring that low-confidence depth data doesn't improperly influence the final perception.
3Manufacturing precision
If stereo vision systems output disparity maps only, then the manufacturing precision is maintained at baseline levels, but the measurement precision of depth values deteriorates due to lack of uncertainty quantification
Solution Approach 1:
The patent adds a new dimension to the depth estimation output by introducing the confidence dimension. Instead of providing only a single depth value per pixel, the system now outputs depth values paired with confidence scores, effectively transforming the output from a 2D depth map to a 3D representation that includes uncertainty information. This additional dimension enables more precise measurement by allowing downstream systems to account for variability in the depth estimates.
Data Source
AI summary
An automated vehicle assistance system is provided for supervised or unsupervised vehicle movement. The system includes a control system and a first sensor system. The first sensor system may receive first image data of a scene and may output a first disparity map and a first confidence map based on the first image data. The control system may output a video stream based on the first disparity map and the first confidence map. The vehicle assistance system also may include a second sensor system that receives second image data of at least a portion of the scene that outputs a second confidence map based on second image data. The video stream may include super-frames, with each super-frame including a 2D image of the scene, a depth map corresponding to the 2D image, and a certainty map corresponding to the depth map.