Dynamic Obstacle Point Cloud Annotation Using Static Background Mesh
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual annotation of dynamic obstacle point clouds in autonomous driving systems is time-consuming and inefficient, requiring significant manpower and taking 10-20 minutes per frame of point cloud data.
Innovation Solution
A method and apparatus that collect first point cloud data under a static scenario to build a static background mesh model, then use this model to annotate dynamic obstacles in subsequent point cloud data by projecting and clustering points, thereby automating the annotation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation method is used, then annotation accuracy can be ensured, but annotation time and manpower consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-building a static background mesh model from static point cloud data before processing dynamic point cloud frames. This pre-established model serves as a reference framework that enables automatic identification of dynamic obstacles in subsequent frames, eliminating the need for manual annotation in each frame while maintaining accuracy through comparison against the pre-built static structure
Solution Approach 2:
The patent replaces the manual mechanical annotation process with an automated computational system. The system uses point cloud projection onto the static background mesh model, automatic clustering algorithms, and computer vision processing to identify and annotate dynamic obstacles, substituting human manual operations with automated mechanical and computational processes that achieve both speed and accuracy
2Loss of information
If manual annotation method is used, then detailed obstacle information can be captured, but productivity and annotation efficiency decrease
Solution Approach 1:
The patent implements self-service by enabling the system to automatically annotate dynamic obstacles without human intervention. The automated pipeline projects point clouds onto the static background mesh, identifies dynamic regions through clustering, extracts obstacle features, and generates annotations autonomously, allowing the system to serve itself in the annotation task while maintaining comprehensive obstacle information capture
Solution Approach 2:
The patent introduces a static background mesh model as an intermediary between the raw point cloud data and the final annotation output. This intermediary structure facilitates automatic information extraction by serving as a reference framework that enables the system to identify dynamic obstacles and their characteristics without manual intervention, thereby preserving information completeness while dramatically improving productivity
3Productivity
If automated annotation is implemented, then annotation speed increases, but system complexity increases due to mesh modeling and point cloud processing
Solution Approach 1:
The patent applies segmentation by dividing the annotation task into distinct modular stages: static background modeling, point cloud projection onto the mesh, dynamic region identification through clustering, feature extraction, and annotation generation. This segmentation of the complex automated process into manageable modules enables high annotation speed while making the system complexity more tractable through structured organization
Data Source
AI summary
A method comprises: collecting first point cloud data under a static scenario around a target collecting point; building a static background mesh model under the static scenario around the target collecting point according to the first point cloud data; collecting a second point cloud data of a target frame under a dynamic scenario after a dynamic obstacle moves around the target collecting point; annotating the point cloud of dynamic obstacle in the second point cloud data corresponding to the target frame, according to the static background mesh model. Through the technical solution of the present disclosure, it is feasible to automatically annotate the point cloud of the dynamic obstacle, effectively save manpower and annotation time spent in annotating the dynamic obstacle point cloud, and thereby effectively improve the efficiency of annotating the dynamic obstacle.