Virtual Sensor Merging Camera Images and 3D-Checkpoints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In highly automated and autonomous motor vehicles, existing object tracking systems face challenges such as ambiguity and systematic measurement errors between sensors, leading to incorrect associations and potential safety hazards like ghost objects or missed recognition of static obstacles.
Innovation Solution
The method involves creating a virtual sensor by merging camera images with 3D-checkpoints, using optical flow to synchronize and project data, and calculating Cartesian velocity vectors without relying on dynamic models, thereby reducing association errors and enhancing robustness in object tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If classic object tracking with predictor-corrector filters is used, then object tracks can be maintained over time, but association errors occur due to ambiguities and systematic measurement errors between sensors
Solution Approach 1:
The patent merges data from multiple sensors (camera, laser scanner, radar) into a unified measurement set before association. By combining measurements at the raw data level rather than processing them separately through individual tracks, the system eliminates ambiguities between sensors and ensures that each measurement is correctly associated with its corresponding object, resolving the association accuracy problem while maintaining tracking reliability
Solution Approach 2:
The patent introduces an intermediary measurement combination step that acts as a mediator between individual sensor measurements and object tracks. This intermediary layer consolidates all measurements into a single coherent dataset, allowing the association algorithm to operate on unified data rather than dealing with sensor-specific ambiguities, thereby improving association accuracy without compromising tracking continuity
2Reliability
If redundant sensors with different measuring principles are installed, then detection reliability improves, but systematic measurement errors between sensors increase
Solution Approach 1:
The patent merges measurements from sensors with different measuring principles (camera, laser scanner, radar) into a unified measurement set before processing. By combining these diverse measurements at the raw data level and then applying a single association algorithm, the system maintains the detection reliability benefits of redundancy while eliminating the measurement consistency problems that arise from sensor-specific processing and systematic errors
3Measurement precision
If Cartesian velocity vectors are estimated through longer observation, then acceleration estimation improves, but prediction errors increase due to large observation times
Solution Approach 1:
The patent replaces the mechanical approach of estimating velocity and acceleration through temporal observation with a direct geometric calculation method. By computing Cartesian velocity vectors directly from the spatial positions of measurements in the unified measurement set, the system obtains accurate velocity and acceleration information without requiring extended observation times, thus eliminating prediction errors caused by time delays while maintaining estimation precision
4Adaptability or versatility
If measurements from different sensors are processed separately, then each sensor's strengths are utilized, but association ambiguities increase
Solution Approach 1:
The patent merges all measurements from different sensors into a single unified measurement set before association processing. This approach maintains the versatility of utilizing each sensor's strengths by incorporating all their measurements, while simultaneously improving association clarity by eliminating the ambiguities that arise when sensors process measurements separately. The unified measurement set allows a single association algorithm to correctly match all measurements with their corresponding objects
Data Source
AI summary
A method, a device and a computer-readable storage medium with instructions for processing sensor data. In a first step, camera images are taken by a camera. Additionally, 3D-checkpoints are detected by at least one 3D-sensor. Optionally, at least one of the camera images can be segmented. The camera images are then merged with the 3D-checkpoints by a data fusion circuit to form data of a virtual sensor. The resulting data are finally output for further processing.