3D Information Determination Using Residual Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining three-dimensional information of target objects in intelligent driving scenarios, such as SLAM based on point cloud information, are resource-intensive, lack robustness, and fail in adverse conditions like rain, fog, and high speeds.
Innovation Solution
A computer-implemented method that determines three-dimensional information of a target object by obtaining images from a vehicle camera, calculating residuals based on two-dimensional and reprojection information, and optimizing the information using a factor graph with a least squares problem, without relying on point cloud data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SLAM method based on point cloud information is used to determine three-dimensional information, then the three-dimensional information can be jointly optimized, but the method is resource-consuming and has poor robustness
Solution Approach 1:
The patent extracts and uses only the necessary two-dimensional image information from the camera, discarding the resource-intensive point cloud processing. It reprojects three-dimensional information onto a two-dimensional plane to match the image data, achieving accurate three-dimensional determination without the computational burden of full SLAM methods.
Solution Approach 2:
The patent creates a simplified copy of the three-dimensional information by reprojecting it onto a two-dimensional plane, which can be directly compared with the two-dimensional image data. This copying approach avoids the need for complex point cloud processing while maintaining the essential geometric relationships needed for accurate three-dimensional determination.
2Reliability
If SLAM method based on point cloud information is used to determine three-dimensional information, then the three-dimensional information can be jointly optimized, but the method is likely to fail in rain and fog, night, high speed and other scenarios
Solution Approach 1:
The patent replaces the mechanical point cloud processing system with an optical-based image processing system. By using two-dimensional image information from the camera and reprojecting three-dimensional data onto the image plane, the method becomes more robust to adverse conditions like rain, fog, and night, while maintaining measurement precision through residual optimization.
3Measurement precision
If residual optimization is performed using factor graph and least squares problem, then the three-dimensional information accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent extracts only the essential geometric constraints needed for three-dimensional determination by reprojecting three-dimensional information onto the two-dimensional image plane. This extraction approach simplifies the optimization problem by focusing only on the relevant residual errors between projected and observed features, reducing computational complexity while maintaining accuracy.
4Use of energy by moving object
If two-dimensional information and reprojection information are used instead of point cloud data, then the computational requirements are reduced, but the measurement accuracy may be affected
Solution Approach 1:
The patent creates an accurate two-dimensional copy of the three-dimensional information by reprojecting it onto the image plane. This copying process preserves the essential geometric relationships while using only lightweight two-dimensional image data, achieving both reduced computational requirements and maintained measurement precision through residual optimization.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The embodiments of the present disclosure provide a method and apparatus for determining three-dimensional information of a target object. The method includes: obtaining an image captured by a vehicle camera; and determining, based on a residual, optimized three-dimensional information of the target object in a target image, where the residual is determined based on two-dimensional information and projection information obtained by reprojecting three-dimensional information onto a two-dimensional plane, and the two-dimensional information and the three-dimensional information of the target object are determined in the image. The embodiments of the present disclosure do not rely on three-dimensional point cloud information, but mainly rely on two-dimensional information and three-dimensional information determined from two-dimensional images. Since the two-dimensional information has high measurement accuracy in rain and fog, night, and high speed environments, the method of the present disclosure has good robustness.