Multi-Sensor Vehicle Object Detection With Reliability-Region Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing environment perception systems in vehicles face reduced detection performance due to unreliable data from sensors under adverse ambient conditions, such as weather or lighting changes, which can impair sensor data fusion and accuracy.
Innovation Solution
A method utilizing at least two independent imaging sensors (e.g., RADAR and LIDAR, or camera) with different characteristics, repeatedly performs object detection, correlates detections, determines a reliability region, and confines sensor fusion to this region, using historical data to adapt to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data fusion is performed using all available sensors, then comprehensive environment perception is achieved, but detection accuracy deteriorates under adverse ambient conditions due to unreliable sensor data
Solution Approach 1:
The patent applies local quality by determining a reliability region for each sensor based on ambient conditions and sensor characteristics. Instead of uniformly treating all sensor data, the system assigns different reliability weights to different spatial regions and sensor types. This allows the fusion algorithm to selectively trust certain sensors in certain regions, thereby maintaining high detection accuracy while adapting to adverse conditions.
Solution Approach 2:
The system dynamically changes parameters by adjusting sensor reliability weights based on ambient conditions (lighting, weather, time of day). The reliability region parameters are continuously updated according to environmental factors and historical data, allowing the fusion algorithm to adapt its behavior and maintain optimal detection accuracy across varying conditions.
2Reliability
If multiple sensors with different characteristics are used, then detection coverage is improved, but system complexity increases
Solution Approach 1:
The patent manages complexity by applying local quality principles to sensor selection and fusion. Different sensor types (camera, LIDAR, RADAR) are assigned to specific reliability regions based on their characteristics and performance under various conditions. This localized approach allows the system to leverage multiple sensors for comprehensive coverage while simplifying the fusion process by region-specific rules rather than complex global optimization.
Solution Approach 2:
The system segments the environment into multiple reliability regions, each with its own optimal sensor configuration. This segmentation allows independent optimization for each region, reducing overall system complexity by breaking down the complex multi-sensor fusion problem into manageable regional sub-problems that can be solved with simpler, region-specific algorithms.
3Measurement precision
If sensor fusion is performed without considering reliability regions, then processing speed is maintained, but detection precision deteriorates due to inclusion of unreliable data
Solution Approach 1:
The patent applies preliminary action by pre-determining reliability regions for each sensor based on ambient conditions and historical performance data. This pre-computation of reliability weights allows the fusion algorithm to quickly select appropriate sensor data without complex real-time reliability assessments, thereby maintaining high processing efficiency while ensuring precise object localization through selective data fusion.
Solution Approach 2:
The system achieves both precision and efficiency by applying local quality to the fusion process. Reliability regions are determined locally for each sensor and environment condition, allowing the algorithm to quickly identify and fuse only the most reliable data sources for each specific situation, thus maintaining high processing speed while ensuring accurate object localization.
Data Source
AI summary
A method for environment perception with at least two independent imaging environment perception sensors, including analyzing images from the environment perception sensors by respective object detection algorithms, performing object detection repeatedly in succession for the respective environment sensors for dynamic object detection, entering the object detections together with position information in one or more object lists, correlating the object detections in the one or more object lists with one another, increasing an accuracy of object localizations by sensor fusion of the correlated object detections, determining a reliability region regarding each object detection by at least one environment perception sensor, and confining the sensor fusion of the object detections to the reliability region, wherein outside the reliability region, object localization takes place on the basis of the object detections by the at least one environment perception sensor to which the determined reliability region does not apply.