Ground Truth Generation for Road Participant Camera Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating ground truth for vehicle surroundings, such as manual labeling or using the same sensors for validation, are labor-intensive, costly, and prone to errors, especially in adverse weather conditions.
Innovation Solution
A method and apparatus that utilize first-type cameras to receive and process image information about vehicle surroundings, extracting second image information about other road participants, and generating a ground truth for validation purposes, which can also incorporate image information from other types of sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling method is used to generate ground truth, then precision of ground truth can be ensured, but labor costs increase and time consumption increases
Solution Approach 1:
The patent uses image information from first-type cameras to create a digital copy/reference (ground truth) of the road participant data. Instead of manual labeling, the system automatically generates this reference data by processing images from dedicated first-type cameras, which serve as a reliable copy of the actual scene for validation purposes.
Solution Approach 2:
The system performs self-validation by using first-type cameras to generate ground truth that is then used to validate data from second-type cameras. The same vehicle-mounted system that collects data also generates its own validation reference, eliminating the need for external manual labeling while maintaining precision through the dedicated first-type camera system.
2Productivity
If information collected by lidar sensor is used to generate ground truth, then productivity increases, but measurement precision deteriorates in adverse weather conditions
Solution Approach 1:
The patent creates a visual copy of the scene using first-type cameras that capture image information. This optical copy serves as the ground truth reference, which is more reliable than lidar data in adverse weather conditions since optical cameras maintain precision where lidar sensors fail due to rain, snow, or fog interference.
Solution Approach 2:
The system changes the sensing modality from active illumination (lidar) to passive optical detection (cameras). By using first-type cameras that capture reflected light rather than emitting laser pulses, the system adapts to adverse weather conditions where lidar precision deteriorates, maintaining ground truth quality through parameter change in the sensing approach.
3Device complexity
If the same sensor is used to generate ground truth and for validation, then device complexity decreases, but measurement precision deteriorates due to common errors
Solution Approach 1:
The patent segments the camera system into two distinct functional groups: first-type cameras dedicated to generating ground truth and second-type cameras dedicated to data collection for validation. This segmentation ensures that ground truth and sensed values are generated by different sensor instances, eliminating common cause errors while maintaining manageable system complexity through clear functional separation.
Solution Approach 2:
The first-type cameras serve as an intermediary that creates an independent reference frame (ground truth) separate from the second-type cameras that collect data for validation. This intermediary system acts as a mediator that provides an unbiased reference, allowing validation without the common errors that would exist if the same sensors were used for both purposes.
Data Source
AI summary
A method for generating a ground truth for other road participant(s) is disclosed. The method includes (i) receiving first image information about surroundings of a vehicle from first-type cameras mounted on the vehicle, (ii) extracting second image information about other road participant(s) from the first image information, and (iii) generating a ground truth for the other road participant(s) based at least on the second image information. The ground truth is used to validate a sensed value about the other road participant(s) collected by a second-type camera mounted on the vehicle. In addition, an apparatus for generating a ground truth for other road participant(s), a computer storage medium, and a computer program product vehicle are disclosed.