Vehicle Sensor Fusion Tracking Using Cluster Trace Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for object detection and tracking in vehicles using multiple perception sensors face challenges such as duplicate detection, ghost objects, difficulty in understanding object shape and size, and increased accident risk due to incomplete tracking and missing relevant objects.
Innovation Solution
A method and control arrangement for tracking objects using multiple perception sensors that involve generating clusters for each sensor at different time points, determining a trace using these clusters, calculating a distance measure, and deciding whether additional clusters are indicative of the same object based on a distance threshold, thereby improving tracking accuracy and reducing computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If point cloud-based perception sensors are used to detect objects in the vehicle environment, then a 3D representation of the full environment can be obtained, but duplicate detection and ghost objects may occur due to incomplete aggregation of all points representing a tracked object
Solution Approach 1:
The patent segments the point cloud data into clusters associated with different objects. Each sensor's detections are grouped into clusters, and the system determines whether clusters from different sensors represent the same object by comparing their spatial and temporal characteristics. This segmentation approach prevents duplicate detection by properly associating points from multiple sensors with the same physical object.
Solution Approach 2:
The patent implements a feedback mechanism where the system continuously updates object tracks by comparing new sensor detections with existing tracks. The feedback loop involves calculating distance measures between new clusters and existing object traces, and using this information to either update existing tracks or create new ones, thereby maintaining reliable tracking over time.
2Adaptability or versatility
If multiple perception sensors are mounted on different locations on the vehicle to provide different fields of view, then coverage of the environment is improved, but the complexity of aggregating and processing sensor data increases
Solution Approach 1:
The patent merges sensor data from multiple perception sensors by aggregating detections into unified object tracks. The system combines clusters from different sensors that represent the same object, using spatial and temporal consistency checks. This merging process simplifies the overall system by providing a unified view of the environment despite using multiple diverse sensors.
Solution Approach 2:
The patent creates a universal tracking framework that can handle data from multiple types of perception sensors (LiDAR, camera, RADAR) mounted at different locations. The distance measure calculation and object association logic work universally across different sensor types and mounting configurations, making the system adaptable to various vehicle platforms and sensor arrangements.
3Measurement precision
If all points representing a tracked object are considered in aggregation, then accurate object tracking is achieved, but computational resources and processing time increase
Solution Approach 1:
The patent performs preliminary clustering of sensor data before the association step. By pre-grouping points into clusters based on spatial proximity and temporal coherence, the system reduces the computational complexity of the subsequent association process. This preliminary action allows the system to work with condensed cluster representations rather than individual points, significantly reducing processing time while maintaining tracking accuracy.
4Productivity
If conventional point cloud aggregation methods are used, then object detection can be performed, but difficulties in understanding the shape and size of objects occur due to incomplete aggregation
Solution Approach 1:
The patent replaces traditional geometric processing methods with a physics-inspired approach using traces and distance measures. Instead of relying on complex geometric operations to understand object shape and size, the system uses temporal-spatial traces that naturally encode object geometry through the motion patterns of detected points. This substitution simplifies the process of understanding object properties while maintaining complete information aggregation.
Data Source
Figure 2
Figure 3
Figure 4
AI summary
A method performed by a control arrangement (700) configured to track an object in a scene (110) observed by a plurality of perception sensors comprised in a vehicle, the method comprising: obtaining first sensor data from a first perception sensor (S1), the first sensor data being indicative of the tracked object, the first sensor data comprising a first cluster (120) of sensor data being detected at a first point in time (TA1) and comprising a second cluster (140) of sensor data being detected at a subsequent second point in time (TA2), obtaining second sensor data from a second perception sensor (S2) indicative of a candidate object, the second sensor data comprising a third cluster (130) of sensor data being detected at a third point in time (TB), the third point in time (TB) being between the first point in time (TA1) and second point in time (TA2), determining a trace using the first cluster (120) and the second cluster (140) of the first sensor data, calculating a distance measure between the third cluster (130) of sensor data and the determined trace, determining that the third cluster (130) of sensor data is indicative of the tracked object if the calculated distance measure is below a distance threshold, and tracking the object using the first cluster (120), the second cluster (140) and the third cluster (130) if it is determined that the third cluster (130) of sensor data is indicative of the tracked object, or tracking the object using the first cluster (120) and the second cluster (140) only if it is determined that the third cluster (130) of sensor data is not indicative of the tracked object.