Sensor Data Label Transfer for Occluded Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for labeling objects in sensor data for autonomous vehicles are either time-consuming and costly due to human involvement or require complex algorithms, and often fail to detect objects beyond the vehicle's sensor range or occluded by other objects or weather conditions.
Innovation Solution
A method for automatically generating labels by identifying sensor data from one vehicle and transferring labels from a nearby vehicle with better positioning and orientation, using a conversion from the local coordinate system of the second vehicle to the first, allowing for accurate labeling of objects beyond the initial vehicle's perception range or occluded by other objects or weather conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human operators manually label objects in sensor data, then labeling accuracy can be maintained, but the process becomes time-consuming and costly
Solution Approach 1:
The patent copies labels from a second vehicle's sensor data to the first vehicle's sensor data. Instead of manually creating labels, the system identifies corresponding objects in both datasets and transfers the already-created labels, significantly reducing time and cost while maintaining accuracy through verification processes
Solution Approach 2:
The system enables automatic label generation where the labeling process serves itself by using labels from one vehicle to automatically create labels for another vehicle. The verification mechanism allows the system to self-correct and maintain quality without requiring continuous human intervention
2Reliability
If complex algorithms are used to detect objects beyond sensor range or occluded objects, then detection capability may improve, but the system complexity increases
Solution Approach 1:
The patent uses a coordinate transformation intermediary to bridge between different vehicle perspectives. Instead of complex algorithms to detect occluded objects, the system transforms labels from the second vehicle's coordinate system to the first vehicle's coordinate system, enabling detection of objects that would otherwise be undetectable
Solution Approach 2:
The system adds a spatial dimension to object detection by utilizing data from nearby vehicles. Objects occluded from one vehicle's perspective can be detected from another vehicle's vantage point, effectively using a different spatial dimension to overcome detection limitations
3Measurement precision
If labels are transferred from nearby vehicles with better positioning, then labeling accuracy improves, but coordinate system conversion complexity increases
Solution Approach 1:
The patent implements a universal coordinate transformation framework that can handle transformations between any two vehicle coordinate systems. The transformation module serves multiple functions: coordinate conversion, label adaptation, and quality verification, reducing overall system complexity despite the mathematical transformations required
Data Source
AI summary
Aspects of the disclosure provide for automatically generating labels for sensor data. For instance first sensor data for a first vehicle is identified. The first sensor data is defined in both a global coordinate system and a local coordinate system for the first vehicle. A second vehicle is identified based on a second location of the second vehicle within a threshold distance of the first vehicle within the first timeframe. The second vehicle is associated with second sensor data that is further associated with a label identifying a location of an object, and the location of the object is defined in a local coordinate system of the second vehicle. A conversion from the local coordinate system of the second vehicle to the local coordinate system of the first vehicle may be determined and used to transfer the label from the second sensor data to the first sensor data.