Bounding Box Size Estimation Using Kalman Filter Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of object size estimation using range sensor detections, such as LiDAR and radar sensors, deteriorates when the number of detections decreases over time, leading to a shrinking bounding box, which affects the precision of object tracking in automotive advanced safety and autonomous driving systems.
Innovation Solution
A computer-implemented method that determines pseudo measurements of object dimensions using range sensor detections, incorporating a confidence measure based on measurement uncertainty, and applies a Kalman filter to stabilize and improve the accuracy of size estimation by weighting pseudo measurements according to object distance and orientation, thereby preventing fluctuating bounding boxes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If range sensor detections are used to estimate object size, then object tracking is enabled, but measurement precision deteriorates when the number of detections decreases over time
Solution Approach 1:
The patent implements a feedback mechanism where the Kalman filter continuously updates object size estimates based on incoming range sensor detections and previous estimates. The filter uses measurement uncertainty to weight current detections against historical data, maintaining stable size estimates even when detection density varies over time. This feedback loop prevents the bounding box from shrinking when detections decrease.
Solution Approach 2:
The patent changes the parameter used for size estimation from direct bounding box dimensions to Kalman filter-based estimated dimensions that incorporate measurement uncertainty. By transforming the estimation approach to use probabilistic parameters rather than direct measurements, the system maintains precision despite varying detection counts.
2Device complexity
If bounding box size is directly derived from range sensor detections, then object size estimation is simple, but the bounding box fluctuates and shrinks over time
Solution Approach 1:
The patent introduces the Kalman filter as an intermediary between raw range sensor detections and final size estimates. This intermediary component processes the detections through probabilistic modeling, smoothing out fluctuations and preventing direct translation of detection variations into bounding box changes. The filter acts as a mediator that transforms noisy detection data into stable estimates.
Solution Approach 2:
The patent replaces the direct mechanical/geometric relationship between detections and bounding box with a statistical/probabilistic model. Instead of directly computing box dimensions from detection coordinates, the system uses statistical estimation that accounts for measurement uncertainty, substituting geometric determinism with statistical probability.
3Productivity
If all range sensor detections are treated equally, then processing is simple, but estimation accuracy decreases when detection quality varies
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different detections based on their individual measurement uncertainty. Rather than treating all detections equally, the system assigns different weights to detections based on their quality, with higher-quality detections (lower uncertainty) having greater influence on the final estimate. This allows the system to maintain accuracy while processing variable-quality data.
Solution Approach 2:
The patent changes the processing approach from uniform treatment of detections to differentiated treatment based on measurement uncertainty parameters. By introducing uncertainty as a weighting parameter, the system can selectively trust different detections to different degrees, improving accuracy without significantly increasing processing complexity.
Data Source
AI summary
A computer-implemented method for estimating an object size using range sensor detections detected by a range sensor mounted on a host vehicle. The method includes, for an object detected by the range sensor detections, over time, determining, based on the range sensor detections, pseudo measurements of dimensions of a bounding box enclosing the detected object. The method includes determining estimates for the dimensions of the bounding box, based on respective previous pseudo measurements and current pseudo measurements. The estimates are determined by taking into account a confidence measure for the range sensor detections. The method includes determining the object size based on the determined estimates.