Stereo Camera Floor Scoring for Low-Latency Obstacle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional stereodepth techniques used by autonomous mobile devices (AMDs) struggle to accurately distinguish between a floor and obstacles, especially when the stereocamera is close to the floor, leading to incorrect identification of the floor as an obstacle, which can impede movement and reduce the device's speed due to computational intensity and limited resources.
Innovation Solution
The method involves calculating a score for each pixel in the image data from a stereocamera to determine if it is parallel or perpendicular to the floor, using disparity costs under different assumptions, and comparing these scores to a threshold to accurately classify pixels as either floor or obstacle, thereby reducing false positives and computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional stereodepth techniques are used to determine obstacles, then obstacle detection is performed, but the floor is incorrectly identified as an obstacle leading to false positives
Solution Approach 1:
The patent applies local quality by making different parts of the image data undergo different processing. Specifically, pixels within a determined region (likely corresponding to floor area) are excluded from obstacle detection processing, while other pixels continue normal stereodepth analysis. This selective processing based on local image characteristics resolves the contradiction by preventing false positive obstacle detections in floor regions while maintaining accurate obstacle detection elsewhere.
2Reliability
If conventional stereodepth processing is performed on all pixels, then comprehensive obstacle detection is achieved, but computational overhead increases and device speed decreases
Solution Approach 1:
The patent extracts and removes the computationally intensive stereodepth processing from specific regions of the image data that correspond to the floor. By determining a region in the image data and excluding pixels within this region from obstacle detection processing, the system eliminates unnecessary computational overhead while maintaining complete obstacle detection in relevant areas. This extraction principle directly improves device productivity by reducing processing time and enabling faster movement.
3Measurement precision
If stereocamera is placed close to the floor for better navigation, then navigation accuracy improves, but false identification of floor as obstacle increases
Solution Approach 1:
The patent applies local quality by implementing region-specific processing where pixels within a determined region (corresponding to floor area when camera is close to floor) are excluded from obstacle detection. This allows the stereocamera to be positioned close to the floor for improved navigation accuracy while preventing the floor itself from being misidentified as an obstacle through localized processing adjustments.
Data Source
AI summary
An autonomous mobile device (AMD) moving in a physical space determines the presence of obstacles using images acquired by a stereocamera and avoids those obstacles. A floor with few visible features is difficult to characterize. The floor and any obstacles thereon may be difficult to characterize due to noise, perspective effect, and so forth. A score is determined that indicates whether a particular pixel in an image is deemed to be associated with a parallel surface (the floor) or a perpendicular surface (an obstacle). The score is computationally inexpensive to calculate and allows for highly accurate and low latency determinations as to the presence of an obstacle that would impede movement of the AMD. Orientation changes in the AMD, such as movement over a bumpy floor, are well tolerated as are floor features such as flooring transitions, ramps, and so forth which the AMD is able to traverse.