Radar Labeling Using Camera Depth Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in object detection using radar images is the difficulty in accurate labeling due to unclear shapes of detection targets, which leads to poor detection performance, as the shape clarity depends on size, pose, and reflection intensity, making traditional labeling methods unreliable.
Innovation Solution
A data processing apparatus and method that includes a target object position determination unit, a depth distance extraction unit, a coordinate transformation unit, and a label transformation unit, utilizing camera images and radar signals to accurately determine and label target objects in radar images by transforming positions into a world coordinate system and generating precise labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional labeling methods are used in radar images, then the labeling process is simple, but the labeling precision deteriorates due to unclear target shapes
Solution Approach 1:
The patent merges camera imaging data with radar measurement data to create a unified labeling system. The camera provides clear visual shape information while the radar provides depth and positional data, combining their strengths to achieve accurate labeling even when radar images alone show unclear target shapes
Solution Approach 2:
The patent introduces a coordinate transformation unit as an intermediary that converts camera image coordinates to radar coordinate systems. This intermediary component enables precise label transfer from the clear camera images to the radar image space, resolving the labeling precision issue without requiring direct interpretation of ambiguous radar shapes
2Loss of information
If camera images alone are used for object detection, then the shape clarity is good, but depth information is insufficient
Solution Approach 1:
The patent adds the depth dimension (z-axis) to the two-dimensional camera image data by integrating radar range measurements. This transforms the planar image coordinates into three-dimensional spatial coordinates, recovering depth information that is lost in traditional 2D image processing while maintaining shape clarity from the camera
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the precision of labeling in radar images, improving the accuracy of object detection by correlating camera and radar data to clarify ambiguous target shapes, thereby improving model performance.
Implementation Method 1
an antenna (radar 2) placed in an x-y plane (a panel 1 in FIG. 21) in a part (A) of FIG. 21 radiates a radio wave and measures a signal reflected from an object (pedestrian)
Data Source
AI summary
A data processing apparatus (100) includes a target object position determination unit (103) that determines, based on a picture image acquired by a first camera, a position of a target object in the picture image, a target object depth distance extraction unit (104) that extracts the depth distance from the first camera to the target object, a coordinate transformation unit (105) that transforms the position of the target object in the picture image into a position of the target object in a world coordinate system by using the depth distance, and a label transformation unit (106) that transforms, by using a position of the first camera in the world coordinate system and imaging information used when an image is generated from a measurement result of a sensor, the position of the target object in the world coordinate system into a label of the target object in the image.