LIDAR Error Detection via Optical Flow Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LIDAR systems in robots, such as autonomous vehicles, can produce erroneous data due to sensor malfunctions or malicious spoofing, leading to potential accidents by incorrectly detecting obstacles.
Innovation Solution
A system that combines LIDAR point-cloud data with optical-flow data from camera images to identify erroneous data by measuring spatial correspondence using a 2D grid and a 1D hash table, allowing the processor to determine if the data satisfies predetermined criteria for validity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LIDAR sensors are used to detect obstacles in autonomous vehicles, then the robot can perceive the environment, but erroneous data may be produced due to sensor malfunction or spoofing
Solution Approach 1:
The patent combines LIDAR point-cloud data with optical-flow data from camera images to create a fused representation of the environment. By merging data from these two different sensing modalities, the system can cross-validate obstacle detections and identify inconsistencies that indicate spoofed or erroneous LIDAR data, thereby improving reliability while filtering harmful false detections
Solution Approach 2:
The system implements a feedback mechanism where the processor continuously compares LIDAR-derived optical flow with camera-derived optical flow. When discrepancies exceed predetermined thresholds, the system flags the LIDAR data as erroneous and adjusts its trust level accordingly, creating a closed-loop verification system that reduces the impact of harmful factors
2Measurement precision
If LIDAR point-cloud data is processed to detect obstacles, then obstacle detection capability is improved, but computational complexity increases due to data fusion and verification processes
Solution Approach 1:
The patent segments the complex verification process into distinct computational stages: generating LIDAR point-cloud data, computing LIDAR-derived optical flow, capturing camera images, computing camera-derived optical flow, fusing the data in a 2D grid, and performing consistency checks. This segmentation allows the processor to handle complex measurements through modular operations, managing computational complexity while maintaining high measurement precision
Solution Approach 2:
The system transforms the 3D LIDAR point-cloud data into a 2D grid representation for fusion with optical-flow data. This dimensionality reduction simplifies the computational burden by projecting three-dimensional spatial information onto a two-dimensional plane, enabling efficient data fusion and verification while preserving the essential spatial relationships needed for accurate obstacle detection
3Reliability
If multiple data sources are fused to verify LIDAR data, then detection reliability is improved, but processing time increases
Solution Approach 1:
The system implements partial verification by comparing LIDAR and camera data only at critical decision points or when consistency thresholds are breached, rather than performing exhaustive continuous comparison. This partial action approach maintains high detection reliability for critical obstacle identification while reducing overall processing time by avoiding redundant full-verification cycles in low-risk situations
Data Source
AI summary
Techniques for detecting erroneous LIDAR data are disclosed herein. One embodiment receives LIDAR point-cloud data pertaining to a robot's environment; receives image data and generates segmented optical-flow data therefrom; fuses, in a 2D grid, a plurality of objects including LIDAR points and optical-flow pixels; executes a hash function that generates, for the plurality of objects, a 1D hash table and an associated index; performs one or more queries using the 1D hash table and the associated index to measure the extent of spatial correspondence between the LIDAR points and the optical-flow pixels; identifies the LIDAR point-cloud data as erroneous, when the extent of spatial correspondence fails to satisfy one or more predetermined criteria; and identifies the LIDAR point-cloud data as valid and controls operation of the robot based, at least in part, on the LIDAR point-cloud data, when the extent of spatial correspondence satisfies the one or more predetermined criteria.