Object Corner De-Biasing for Accurate Autonomous Vehicle Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for road geometry modeling and object detection in autonomous vehicles are resource-intensive, time-consuming, and prone to bias in object labeling, which can lead to inaccurate object detection and inefficient autonomous driving.
Innovation Solution
A method and apparatus that automatically detect objects in an environment by applying a bias transformation to corner locations in images, using a processor to receive sensor data, determine reference corner locations, and generate de-biased corners to update map databases, thereby reducing labeling bias and improving object detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling methods are used for object detection, then labeling flexibility is maintained, but labeling accuracy deteriorates due to human bias
Solution Approach 1:
The patent introduces an automated image processing system as an intermediary between the object and human labelers. This system pre-processes images to automatically detect and mark object locations, providing a standardized reference that human labelers can follow. This intermediary step eliminates direct human bias in initial detection while maintaining human oversight for quality control, thereby improving both accuracy and consistency.
Solution Approach 2:
The system performs preliminary automated object detection and location marking before human labeling occurs. By pre-identifying object locations using algorithmic image processing, the system establishes a consistent baseline that reduces variability in human labeling. This preliminary action ensures that all labelers start from the same objective reference point, improving reliability while allowing humans to refine labels when necessary.
2Measurement precision
If traditional road geometry modeling methods are used, then measurement accuracy is maintained, but resource consumption increases significantly
Solution Approach 1:
The patent replaces manual mechanical measurement and labeling processes with automated image processing algorithms. Computer vision techniques automatically detect object locations, extract geometric features, and generate map data without human intervention. This substitution dramatically improves processing efficiency while maintaining or enhancing measurement precision through consistent algorithmic application across all images.
Solution Approach 2:
The system transforms the detection process by changing key parameters from manual measurement to automated image analysis. By using digital image processing parameters such as pixel coordinates, edge detection thresholds, and feature matching algorithms, the system achieves higher precision and faster processing speeds compared to traditional manual methods. This parameter transformation enables scalable processing of large datasets.
3Productivity
If automated object detection is implemented, then processing speed is improved, but detection reliability deteriorates due to unknown accuracy
Solution Approach 1:
The patent implements feedback mechanisms where detection results are continuously evaluated and refined. The system compares automated detection outputs against ground truth data, calculates accuracy metrics, and uses this feedback to adjust detection parameters and improve future detections. This closed-loop feedback system builds confidence in automated detection reliability while maintaining high processing speeds through optimized algorithms.
Data Source
AI summary
Described herein are methods of generating learning data to facilitate de-biasing the labeled location of an object of interest within an image. Methods may include: receiving sensor data, where the sensor data is a first image; determining reference corner locations of an object in the first image using image processing; generating observed corner locations of the object in the first image from the determined reference corner locations; generating a bias transformation based, at least in part, on a difference between the reference corner locations and the observed corner locations of the object in the first image; receiving sensor data from another image sensor of a second image; receiving observed corner locations of an object in the second image from a user; and applying the bias transformation to the observed corner locations of the object in the second image to generate de-biased corners for the object in the second image.