Vehicle Sensor Data Labeling Using Confidence-Based Object Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Obtaining high-quality labelled vehicle sensor data is complex, expensive, and time-consuming, which affects the reliability and accuracy of machine learning systems used in autonomous vehicle control.
Innovation Solution
A computer-implemented method that processes sensor data from distance-ranging sensors on a vehicle with object data from the environment to determine associations, calculates confidence values, and labels sensor data based on attribute data when confidence exceeds a threshold, without requiring expensive cameras or human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human annotation or expensive camera sensors are used to label sensor data, then the quality and accuracy of labelled data is improved, but the cost and time required for data generation increases significantly
Solution Approach 1:
The system uses the vehicle's own distance-ranging sensor data to automatically generate labels for sensor portions, eliminating the need for external human annotation. The processor compares sensor data with object data to self-generate accurate labels efficiently
Solution Approach 2:
The system uses object data (which contains location and attribute information) as a reference copy to create labels for sensor data portions. Instead of manual annotation, the system copies relevant attribute data from objects to label corresponding sensor portions automatically
2Measurement precision
If human annotation or expensive camera sensors are used to label sensor data, then the quality and accuracy of labelled data is improved, but the cost of data generation increases significantly
Solution Approach 1:
The system replaces expensive camera sensors and human annotators with a computationally efficient processing method that uses existing distance-ranging sensor data. The solution uses inexpensive computational resources rather than expensive hardware or human labor
Solution Approach 2:
The system copies attribute data from object data to label sensor portions automatically, eliminating the need for expensive camera sensors or human annotation services while maintaining high label quality
3Productivity
If automated processing without confidence thresholding is used, then the productivity of data generation is improved, but the reliability of labelled data deteriorates
Solution Approach 1:
The system calculates a confidence value for each potential association between sensor data and objects, then uses this feedback to determine whether to apply a label. Only associations exceeding the confidence threshold receive labels, ensuring reliability while maintaining automated processing efficiency
Solution Approach 2:
The system applies labels selectively only when confidence exceeds the threshold, rather than labeling all sensor data portions. This partial action approach ensures high reliability by avoiding uncertain labels while maintaining productive automated processing
Data Source
Figure 1~3
Figure 4
Figure 5~6
AI summary
A computer-implemented method comprising: receiving, at a processor, sensor data generated by at least one distance-ranging sensor on a vehicle within an environment; receiving, at the processor, object data received at the vehicle, the object data originating from a plurality of objects within the environment, wherein the object data comprises location data indicating a respective location of each object and attribute data associated with each respective object; processing the data to determine a possible association between a respective portion of the sensor data and at least one object of the plurality of objects; calculating a confidence value reflecting the reliability of the possible association; if the confidence value exceeds a threshold, labelling the portion of the sensor data with at least a portion of the attribute data of the associated object(s).