Stereo Driver Assistance Verification for Object Detection Faults
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face a critical hazard of failing to detect objects due to electrical errors, which can lead to hazardous situations, especially in autonomous driving scenarios where the driver is out of the loop.
Innovation Solution
A road profile detector and verification algorithm with higher functional safety levels are used to verify the detection of objects by comparing the road profile with a 3D point cloud, ensuring accurate object detection by identifying malfunctions through deviations from expected distances in the road-object profile.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If object detection software runs on hardware with lower functional safety level (QM or ASIL-A), then computational efficiency and processing speed are improved, but the risk of electrical errors causing false detections increases
Solution Approach 1:
The system divides the detection function into two segments: a primary detection software running on lower-safety hardware for fast processing, and a verification software running on higher-safety hardware for safety-critical validation. This segmentation allows each component to operate on appropriate hardware levels, balancing speed and safety.
Solution Approach 2:
The verification software acts as an intermediary between the object detection software and the braking system. It receives detection results from the primary software and validates them before triggering any safety-critical actions, thereby mediating the risk between computational speed and functional safety.
2Reliability
If verification software runs on hardware with higher functional safety level (ASIL-B to ASIL-D), then the safety against electrical errors is improved, but computational resources and processing power are reduced
Solution Approach 1:
The verification software extracts only the essential safety-critical validation function from the full detection system. By taking out only the necessary verification capability, the system can run on hardware with lower computational power while maintaining ASIL-D safety level, as the full complex detection algorithms are not required on the verification side.
Solution Approach 2:
The verification software is designed as a simpler, less computationally intensive system that performs only safety validation rather than full object detection. This 'cheaper' verification layer replaces the need for high-power hardware, using computational resources efficiently by only validating critical detection results rather than performing complete analysis.
3Extent of automation
If the driver is out of the loop in autonomous driving, then automation level is improved, but the hazard of failing to detect objects increases since the driver cannot react
Solution Approach 1:
The verification software provides feedback validation on object detection results before they are used for safety-critical actions. This feedback mechanism ensures that detection errors are caught and corrected, maintaining high reliability in autonomous driving where the driver cannot immediately react to system failures.
Solution Approach 2:
The system implements beforehand cushioning by validating detection results through the verification software before executing any safety-critical braking actions. This prior validation layer cushions against the risks of autonomous driving by ensuring detection accuracy is verified before any hazardous situation can occur.
Data Source
Figure 1~2
Figure 3~5
AI summary
A driver assistance system for a motor vehicle comprises a stereo imaging apparatus (11) with a plurality of cameras (12) adapted to capture images from a surrounding of the motor vehicle, and a processing device (14) adapted to perform image processing of captured images in order to detect objects in images captured by said imaging apparatus (11). The image processing comprises a stereo algorithm (51) adapted to calculate a 3D point cloud (52) of images captured by said imaging apparatus (11), an object detector (53) adapted to detect objects in images captured by said imaging apparatus (11), and a road profile detector (54) adapted to detect the profile (61) of the road ahead of the motor vehicle and to calculate a road-object profile (68) from a combination of the detected road profile (61) and objects detected by said object detector (53). The image processing further comprises a verification algorithm (55) having a higher functional safety level than said object detector (53), wherein said verification algorithm (55) is adapted to compare the road-object profile (68) with a corresponding part (52) of the 3D point cloud (52), and to determine a malfunction based on said comparison of the road profile (68) with a corresponding part (52) of the 3D point cloud (52).