3D Sensor Fusion for Autonomous Obstacle Map Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately distinguishing between obstacles and traversable areas, particularly due to difficulties in detecting certain obstacles like chain link fences and distinguishing between reflections and actual obstacles.
Innovation Solution
The system processes greyscale and depth sensor data to create a three-dimensional representation of the environment, which is then segmented and fused to generate an obstacle map. This map is used to determine navigation data, improving the vehicle's ability to avoid obstacles and reduce errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If autonomous vehicles use sensors to detect obstacles, then navigation capability is improved, but measurement precision deteriorates due to difficulty in distinguishing obstacles from reflections and traversable areas
Solution Approach 1:
The patent transforms two-dimensional sensor data (greyscale and depth images) into a three-dimensional point cloud representation of the environment. This dimensional transformation allows the system to better distinguish between obstacles and traversable areas by adding spatial context that is not apparent in 2D images alone.
Solution Approach 2:
The patent segments the three-dimensional point cloud into distinct regions using clustering algorithms. This segmentation separates obstacles from traversable areas and identifies false detections like reflections, enabling more accurate obstacle detection by analyzing the spatial distribution and characteristics of segmented regions.
2Extent of automation
If autonomous vehicles rely on sensor data processing, then automation level is improved, but reliability deteriorates due to false detections and missed obstacles
Solution Approach 1:
The patent implements a feedback mechanism where the system continuously compares sensor data with the generated obstacle map and navigation path. This feedback loop allows the autonomous vehicle to detect errors in obstacle identification, correct false detections, and improve reliability through iterative refinement of its environmental understanding.
Solution Approach 2:
The patent fuses multiple types of sensor data (greyscale images and depth data) to create a composite three-dimensional representation. This fusion of different data modalities compensates for the weaknesses of individual sensors and reduces false detections, improving overall reliability of obstacle identification.
3Measurement precision
If autonomous vehicles process sensor data to create obstacle maps, then navigation accuracy is improved, but computing complexity increases
Solution Approach 1:
The patent divides the complex task of obstacle detection into sequential processing stages: generating three-dimensional representations from sensor data, segmenting the point cloud into regions, identifying obstacles within segments, and creating the final obstacle map. This segmentation of the processing pipeline manages computational complexity by breaking down the overall task into smaller, more manageable operations.
Data Source
AI summary
Systems, methods, and computer-readable media are disclosed for determining obstacle data based on sensor data such as greyscale data and depth data. Based on the obstacle data navigation data may be determined and used by an autonomous vehicle to navigate an environment such as a warehouse or storage facility. The obstacle data may be determined by determining three-dimensional representations of the greyscale data and the depth data and segmented and combining or fusing the three-dimensional representations of the greyscale data and the depth data. The system used to determine the obstacle data may be trained to avoid false obstructions and omitted obstructions.