Obstacle Detection Using MstSegmentation Clustering on 3D Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current obstacle detection methods for driverless vehicles are time-consuming and labor-intensive, requiring sample labeling and model training, and can result in missed detections of unfamiliar obstacles, reducing detection accuracy.
Innovation Solution
The method involves obtaining a 3D point cloud from a driverless vehicle, determining sets of vertices and edges, and clustering them using the MstSegmentation algorithm to generate a smallest generated tree for obstacle detection, eliminating the need for sample labeling and model training, and allowing detection of any type of obstacle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sample labeling and model training are performed to detect obstacles, then detection accuracy can be improved for known obstacles, but the workload increases and time is consumed
Solution Approach 1:
The system performs self-service by automatically generating detection models without requiring external sample labeling and training. The MstSegmentation algorithm processes 3D point clouds autonomously to identify obstacles, eliminating the need for manual annotation and model training time while maintaining detection accuracy for various obstacle types
Solution Approach 2:
The invention changes the fundamental parameters of obstacle detection by transitioning from a supervised learning approach (requiring labeled samples and training) to an unsupervised clustering approach (MstSegmentation algorithm). This parameter change allows the system to detect obstacles directly from raw 3D point cloud data without time-consuming sample preparation and model training processes
2Measurement precision
If sample labeling and model training are performed, then detection accuracy for trained obstacles improves, but the system cannot detect unfamiliar obstacles, reducing overall detection accuracy
Solution Approach 1:
The MstSegmentation algorithm provides universal obstacle detection capability by clustering 3D point clouds based on spatial relationships and geometric characteristics rather than trained categories. This universal approach enables the system to detect any type of obstacle (pedestrians, vehicles, bicycles, or unknown objects) without requiring specific training data, thereby improving both detection accuracy and versatility simultaneously
3Productivity
If traditional obstacle detection methods are used, then processing can be performed, but the workload is time-consuming and laborious
Solution Approach 1:
The system eliminates manual workload by implementing self-service processing where the MstSegmentation algorithm automatically processes 3D point cloud data to generate obstacle detection results. This self-service mechanism reduces both the time and complexity of obstacle detection work, improving processing efficiency while maintaining detection quality without requiring manual intervention or complex training procedures
Data Source
AI summary
The present disclosure provides an obstacle detecting method and apparatus, a device and a storage medium, wherein the method comprises: obtaining a 3D point cloud collected by a driverless vehicle during travel; determining a set of vertexes and a set of edges respectively according to the obtained 3D point cloud; clustering vertexes in the set of vertexes according to the set of edges to obtain a smallest generated tree as an obstacle detection result. The solutions of the present disclosure can be applied to reduce workload and improve the accuracy of detection results.