Target Detection Using 3D Point Cloud Supporting Planes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional target detection methods generate unreasonable redundant region proposals, leading to waste of computing resources and time during feature extraction and other operations.
Innovation Solution
The method involves acquiring a scene image and a three-dimensional point cloud, identifying a supporting plane, generating region proposals based on the image and the supporting plane, and performing target detection on these proposals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional region proposal generation methods (sliding window, selective search, deep learning) are used, then region proposals can be generated, but unreasonable redundant region proposals are generated causing waste of computing resources and time
Solution Approach 1:
The patent applies preliminary action by identifying the supporting plane before generating region proposals. The method first acquires a three-dimensional point cloud of the scene, identifies the supporting plane from the point cloud, and then uses this plane information to guide region proposal generation. This preliminary identification of the supporting plane allows the system to pre-filter unreasonable regions (such as generating proposals above the supporting plane when objects cannot exist there), thereby reducing redundant region proposals before the computationally expensive feature extraction stage.
2Productivity
If conventional region proposal generation methods are used, then region proposals can be generated, but the number of redundant region proposals increases leading to waste of computing resources
Solution Approach 1:
The patent performs preliminary action by identifying the supporting plane before region proposal generation. This early identification allows the system to establish constraints based on the supporting plane's position and orientation, which are then used to filter out unreasonable region proposals before they are generated or before feature extraction is performed on them, thereby reducing computing resource consumption.
Solution Approach 2:
The patent extracts and utilizes the supporting plane information from the three-dimensional point cloud as a separate, independent element. By extracting the supporting plane's geometric properties (position, orientation, normal vector) and using these as filtering criteria, the system can remove unreasonable region proposals from consideration, thereby reducing the number of proposals that require subsequent computational processing.
Data Source
AI summary
The present disclosure provides a target detection method. The method includes: acquiring a scene image of a scene; acquiring a three-dimensional point cloud corresponding to the scene; identifying a supporting plane in the scene based on the three-dimensional point cloud; generating a plurality of region proposals based on the scene image and the supporting plane; and performing a target detection on the plurality of region proposals to determine a target object to be detected in the scene image. In addition, the present disclosure also provides a target detection device, a terminal device, and a medium.