Hierarchical Perception Monitor for Vehicle Point-Cloud Safety
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle perception systems struggle to robustly detect all safety-relevant objects under diverse environmental conditions, leading to potential detection misses and false positives, and lack verifiability and computational efficiency for safety-certified hardware.
Innovation Solution
A hierarchical monitoring system using a model-based height confidence filter and additional filters to validate primary perception system outputs, ensuring accurate detection of safety-relevant objects while maintaining a low false negative and false positive rate, by converting sensor data into a filtered point cloud and occupancy grid.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI-based perception systems are used to detect objects under diverse environmental conditions, then detection coverage is improved, but reliability deteriorates due to detection misses and false positives
Solution Approach 1:
The perception system is divided into a primary AI-based perception system for broad detection coverage and a secondary monitor system for verification. The monitor system segments the detection task by focusing specifically on validating safety-relevant objects, thereby improving overall reliability without sacrificing the primary system's adaptability.
Solution Approach 2:
A monitor system acts as an intermediary between the primary perception system and the vehicle control system. This intermediary validates detection results independently, reducing false positives and detection misses by cross-checking against different detection criteria and data sources.
2Reliability
If comprehensive validation and verification are implemented, then reliability is improved, but device complexity increases
Solution Approach 1:
The validation process is segmented into hierarchical levels: a primary perception system for general detection and a secondary monitor system for specific verification of safety-relevant objects. This segmentation allows comprehensive validation without requiring complete system redesign, thereby managing complexity.
Solution Approach 2:
The monitor system applies partial validation by focusing only on safety-relevant objects rather than validating all detection results equally. This selective approach maintains high reliability for critical detections while avoiding the complexity of exhaustive validation of every detection scenario.
3Measurement precision
If multiple filters and validation layers are added, then detection precision is improved, but computational efficiency deteriorates
Solution Approach 1:
The system applies multiple validation filters selectively only to detections identified as safety-relevant, rather than applying all filters to every detection. This partial application of validation layers maintains high detection precision for critical objects while preserving computational efficiency by avoiding unnecessary processing of non-critical detections.
Solution Approach 2:
Different levels of validation and filtering are applied locally based on the significance of each detection. Safety-relevant objects receive comprehensive multi-layer validation, while other detections receive minimal or no validation, optimizing the balance between precision and computational efficiency.
Data Source
AI summary
Disclosed herein is a device for filtering a point cloud. The device may include processor configured to receive a plurality of sensed points representing distance measurements to points in an area around an entity. The processor may also be configured to determine, for each sensed point of the plurality of sensed points, one or more probabilities that the sensed point is associated with an obstacle in the area around the entity, wherein each probability of the one or more probabilities is based on a corresponding filter of one or more filters. The processor may also be configured to generate a filtered point cloud from the plurality of sensed points based on, for each sensed point, a hierarchical combination of the one or more probabilities.