Vehicle Sensor Fusion Using Reference Sectors for Pedestrian Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in accurately recognizing pedestrians due to the low positional accuracy of sensors like RaDAR and camera, making it difficult to fuse information effectively, which is crucial for reliable pedestrian recognition and safety.
Innovation Solution
A sensor information fusion method and apparatus that integrates LiDAR, RaDAR, and camera sensors to accurately fuse pedestrian position, classification, and speed information by setting a reference sector based on a first sensor track, selecting target tracks within this sector, and comparing angles and distances to generate fusion tracks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor information fusion is performed using traditional methods, then the system can process multiple sensor inputs, but the positional accuracy and reliability of pedestrian recognition remain low due to inconsistent track data from different sensors
Solution Approach 1:
The patent transforms sensor track data by converting position coordinates into angular parameters relative to a reference sector. This parameter transformation allows inconsistent positional data from different sensors to be compared and fused based on angular relationships, resolving the track consistency issue while maintaining position accuracy.
Solution Approach 2:
The reference sector serves as an intermediary framework that mediates between tracks from different sensors. By defining a reference sector based on one sensor's track and comparing other sensor tracks against this reference, the system creates a common coordinate system that reconciles inconsistencies between sensors.
2Adaptability or versatility
If the system uses multiple sensors (LiDAR, RaDAR, camera) to improve pedestrian detection coverage, then the detection capability increases, but the difficulty of fusing information from sensors with different accuracy characteristics increases
Solution Approach 1:
The patent segments the information fusion process into distinct stages: selecting a reference sensor track, defining a reference sector, comparing target tracks against the reference, and fusing only compatible tracks. This segmentation simplifies the overall fusion complexity by breaking down the multi-sensor integration into manageable, sequential steps.
Solution Approach 2:
The reference sector methodology serves as a universal framework that can accommodate multiple sensor types (LiDAR, RaDAR, camera) with different accuracy characteristics. The same angular comparison approach works regardless of sensor modality, providing a multi-functional solution for heterogeneous sensor fusion.
3Reliability
If the system expands the autonomous driving control area to include pedestrian protection, then the safety requirement increases, but the demand for accurate pedestrian recognition and classification increases
Solution Approach 1:
The system performs preliminary classification of detected targets as pedestrians or non-pedestrians before final fusion. By using the reference sector approach to pre-identify potential pedestrian tracks and their angular relationships, the system prepares classification-ready data structures that improve both safety and classification accuracy.
Data Source
AI summary
An embodiment sensor information fusion method includes acquiring a first sensor track from a first sensor mounted in a vehicle and setting a reference sector based on the first sensor track, selecting at least one target track included in the reference sector, and comparatively analyzing an angle between the at least one target track and the first sensor track and generating a fusion track through fusion with the first sensor track based on an analysis result value.