Sensor Fusion Tracker for Bounding Box Accuracy in Congested Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor-fusion systems face challenges in accurately sizing and positioning bounding boxes around objects in congested scenes due to unreliable error covariance from vision cameras and the inefficiency of Extended Kalman Filters (EKF) in utilizing low-level track data, leading to inaccurate tracking under high-volume situations.
Innovation Solution
A multiple hypothesis-based data fusion tracker that automatically selects between radar and vision-based pseudo measurements, using predefined error covariance to determine the most accurate measurement type for each situation, balancing accuracy and speed by generating bounding boxes based on radar-only, vision angle, or vision lateral position calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If error covariance from vision camera is used to weight object-tracks, then fusion tracking can be performed, but the reliability deteriorates in certain driving scenarios
Solution Approach 1:
The system dynamically changes the error covariance parameters based on detected driving scenarios. When congestion is detected, the error covariance for vision-based tracks is increased (reducing their weight), while radar-based tracks maintain lower error covariance. This parameter adaptation resolves the contradiction by maintaining fusion tracking capability while adjusting reliability weights according to scenario-specific sensor performance.
2Speed
If EKF is used to reduce data fusion complexity, then processing speed improves, but measurement precision deteriorates due to focusing only on linear sensor data
Solution Approach 1:
The system dynamically selects between EKF and multiple hypothesis fusion based on the number of detected objects. When object count is low, EKF is used for fast processing. When congestion is detected, the system transitions to multiple hypothesis fusion which can handle nonlinear relationships and correlations between sensors, thereby maintaining measurement precision while adapting processing speed to scenario complexity.
Solution Approach 2:
The system changes the fusion algorithm parameter (switching between EKF and multiple hypothesis fusion) based on traffic density. This allows the system to optimize between speed and precision by selecting the appropriate algorithm for the current scenario, rather than being constrained to a single approach.
3Measurement precision
If multiple hypothesis fusion is used to improve tracking accuracy in congested scenes, then measurement precision improves, but device complexity increases
Solution Approach 1:
The system dynamically adapts the fusion algorithm complexity based on traffic conditions. In clear scenarios, simple EKF is used. In congested scenarios, multiple hypothesis fusion is activated. This dynamic adaptation allows the system to improve measurement precision when needed while minimizing device complexity during normal operation.
4Adaptability or versatility
If vision camera data is used for bounding box positioning, then adaptability improves across different scenarios, but reliability deteriorates in congested environments
Solution Approach 1:
The system applies different reliability weights to vision camera data based on the local driving scenario. In congested areas, vision data receives higher error covariance (lower weight). In open scenarios, vision data receives lower error covariance (higher weight). This local quality adjustment maintains adaptability across scenarios while ensuring reliability in congested environments by downweighting vision data where it performs poorly.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This document describes a multiple hypothesis-based data fusion tracker. Each hypothesis aligns to a different pseudo measurement type. The fusion tracker automatically determines, using a predefined error covariance associated with the radar, which pseudo measurement type has a greater chance of being accurate for a current situation. The fusion tracker may rely on either one of two combined radar and vision calculations, or the fusion tracker may ignore the vision-based pseudo measurements and instead, rely on radar pseudo measurements alone. By selecting between three different bounding boxes, a vision angle based box, a vision lateral position based box, or a radar only based box, the fusion tracker can balance accuracy and speed when drawing, repositioning, or resizing bounding boxes, even under congested traffic or other high volume situations.