Vehicle Sensor Data Filtering via Predicted Road Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern vehicles face challenges in processing large volumes of sensor data from multiple sensors, such as LIDAR and RADAR, where it is difficult to determine which detections are important and which can be discarded, especially at longer distances where road surface geometry estimation becomes unreliable due to low signal return and grazing angles.
Innovation Solution
The method involves obtaining an estimated road surface geometry close to the vehicle and predicting it beyond the sensor field horizon using a clothoid model, maximum vehicle lateral acceleration, or other factors, to prioritize and filter sensor detections, allowing only important detections within the predicted road surface geometry to be reported, thereby reducing the computational burden and improving decision-making for vehicle control systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all sensor detections are reported to the processing system, then complete obstacle detection coverage is achieved, but the computational burden becomes prohibitively large and system performance degrades
Solution Approach 1:
The sensor field of view is divided into multiple regions of interest (ROIs) based on road surface geometry predictions. Each ROI corresponds to a specific area where obstacles are most relevant (e.g., road surface, sidewalks, curbs). By segmenting the field of view and only processing detections within these ROIs, the computational burden is significantly reduced while maintaining reliable obstacle detection coverage for critical areas.
Solution Approach 2:
Different regions of the sensor field of view are assigned different levels of processing priority and detail based on their importance. Regions corresponding to road surfaces and areas where obstacles are most likely to impact vehicle safety receive higher processing priority and more detailed analysis, while less critical regions receive reduced processing. This local quality approach ensures computational resources are focused where they are most needed.
2Loss of information
If sensor detections at long distances are included in processing, then complete scene awareness is achieved, but the amount of data becomes prohibitively large due to grazing angles and reduced signal return
Solution Approach 1:
Road surface geometry is predicted in advance using clothoid models and vehicle trajectory information before sensor data is processed. These predictions define regions of interest that are prepared beforehand, allowing the system to quickly filter sensor detections against pre-computed ROIs. This preliminary action reduces the need to process all long-distance detections by establishing expected road geometry boundaries ahead of time.
Solution Approach 2:
The system processes sensor detections selectively rather than comprehensively. By using predicted road surface geometry to define ROIs, the system processes only the partial set of detections that fall within these regions, discarding or reducing processing of detections outside the ROIs. This partial action approach maintains adequate scene awareness for safety-critical areas while significantly reducing overall data volume.
3Measurement precision
If road surface geometry estimation is performed using sensor data, then accurate road mapping is achieved, but the method fails at longer distances where grazing angles cause no ground return to be detected
Solution Approach 1:
A clothoid model serving as an intermediary is introduced to predict road surface geometry beyond the sensor's direct detection range. The model uses vehicle trajectory, speed, and previously detected road geometry as inputs to generate predictions about road curvature and surface characteristics at distances where direct sensor measurement is unreliable. This intermediary model bridges the gap between measured and unmeasured regions, extending effective detection range while maintaining reasonable accuracy.
Solution Approach 2:
Instead of trying to directly measure road geometry at long distances where sensor signals are weak, the approach inverts the problem by predicting road geometry from vehicle dynamics and shorter-range measurements. The system uses vehicle acceleration, speed, and steering information to infer road curvature, then uses these predictions to guide sensor processing. This inversion allows the system to work around the physical limitations of sensor grazing angles.
Data Source
Figure 1
Figure 2A~3
Figure 4A~5
AI summary
The present disclosure relates to a method for reporting sensor data from a vehicle sensor (110) to a sensor data processing system (120). The method comprises obtaining (S1) an estimated road surface geometry (121) in a first region (118) that is positioned in a vicinity of the vehicle sensor (110), obtaining (S2) a predicted road surface geometry (122) in a second region (119) that is positioned beyond the first region (118), and obtaining (S3) one or more sensor detections (160, 170) in the second region (119). The method further comprises determining (S4) a first sub-set of sensor detections (160) out of the one or more sensor detections (160, 170) in the second region (119) by selecting sensor detections comprised within the predicted road surface geometry (122), and reporting (S5) the first sub-set of sensor detections (160) as sensor data.