Multi-Scan Radar-Vision Track Fusion for Uncertainty-Aware Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor fusion systems for autonomous and semi-autonomous driving are computationally complex and often require expensive hardware, with uncertainties in object tracking estimates that may not meet safety regulations, especially when using track-to-track fusion techniques with multiple sensor types like radar and vision.
Innovation Solution
A method for multi-scan sensor fusion that generates radar and vision tracks, maintains hypotheses for associations between them based on multiple data scans, determines probability values for these hypotheses, and outputs matches to control vehicle operations, using Dempster-Shafer theory to quantify uncertainty and improve precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing track-to-track fusion techniques are used to combine radar and vision sensor data, then object tracking accuracy is improved, but computational complexity increases significantly requiring expensive processing hardware
Solution Approach 1:
The patent segments the sensor fusion process into distinct modules: hypothesis generation from multiple scans, mass value calculation for each hypothesis, probability determination, and match selection. This modular approach reduces computational complexity by processing data in discrete, manageable stages rather than attempting simultaneous complex fusion of all sensor data.
Solution Approach 2:
The patent performs preliminary actions by maintaining multiple hypotheses across multiple scans before final match determination. Mass values are calculated in advance for each hypothesis based on accumulated scan data, allowing the system to pre-process and evaluate association possibilities before committing to final matches, thereby reducing real-time computational burden.
2Measurement precision
If existing track-to-track fusion techniques are used to combine radar and vision sensor data, then object tracking accuracy is improved, but expensive processing hardware is required
Solution Approach 1:
The patent employs computationally efficient algorithms that can be implemented on standard vehicle processors rather than requiring expensive specialized hardware. The hypothesis-based approach with mass value calculations uses straightforward mathematical operations that are computationally inexpensive, enabling deployment on cost-effective processing platforms.
3Reliability
If existing track-to-track fusion techniques are used, then sensor data combination is achieved, but uncertainties in tracking estimates cannot be quantified or do not satisfy safety regulations
Solution Approach 1:
The patent implements feedback through the hypothesis maintenance mechanism across multiple scans. Mass values are updated and refined with each new scan, providing continuous feedback on the reliability of each hypothesis. This iterative process allows the system to track and quantify uncertainty in the association estimates, ensuring that only hypotheses meeting confidence thresholds are selected as final matches.
Data Source
AI summary
This document describes techniques, systems, and methods for multi-scan sensor fusion for object tracking. A sensor-fusion system can obtain radar tracks and vision tracks generated for an environment of a vehicle. The sensor-fusion system maintains sets of hypotheses for associations between the vision tracks and the radar tracks based on multiple scans of radar data and vision data. The set of hypotheses include mass values for the associations. The sensor-fusion system determines a probability value for each hypothesis. Based on the probability value, matches between radar tracks and vision tracks are determined. The sensor-fusion system then outputs the matches to a semi-autonomous or autonomous driving system to control operation of the vehicle. In this way, the described techniques, systems, and methods can provide high-precision object tracking with quantifiable uncertainty.