LiDAR Map Annotation via Key Frame Selection and Cumulative Recording
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing annotation methods for LiDAR data are inefficient and lack consistency, leading to repetitive processes and potential loss of annotation results, which affects training efficiency and quality.
Innovation Solution
A method and computing device that selectively choose key frames from a LiDAR map for annotation, displaying partial LiDAR maps and corresponding images, and cumulatively record annotation data to generate an annotated LiDAR map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If annotation is performed per frame for all frames, then complete annotation coverage is achieved, but annotation speed decreases and consistency becomes difficult to maintain
Solution Approach 1:
The patent segments the annotation task by selecting only key frames for annotation instead of annotating all frames. The key frame selection divides the continuous stream of frames into discrete annotation points, reducing the total number of annotation operations while maintaining coverage through cumulative recording of annotation results across all frames.
Solution Approach 2:
The patent performs preliminary key frame selection before the annotation process. By pre-identifying which frames should be annotated based on criteria such as frame differences or importance metrics, the system prepares the annotation workflow in advance, avoiding the need to evaluate and annotate every frame individually.
2Quantity of substance
If annotation is performed on all frames, then comprehensive training data is generated, but repetitive annotating processes decrease efficiency
Solution Approach 1:
The patent uses cumulative recording to copy and preserve annotation results across frames. Instead of re-annotating the same objects in every frame, the system copies annotation data from key frames and accumulates it across the sequence, maintaining comprehensive training data coverage while eliminating repetitive manual annotation of identical objects.
Solution Approach 2:
The patent discards redundant annotation operations by identifying frames where objects have already been annotated. The system recovers previously annotated object information and applies it to subsequent frames, avoiding unnecessary repetitive annotation work while maintaining complete object coverage in the training data.
3Ease of operation
If annotation result is reset on next frame, then each frame is independently annotated, but consistency across frames is lost
Solution Approach 1:
The patent implements continuity of useful action through cumulative recording of annotation results. Annotation operations continue across frames by preserving and accumulating annotation data from key frames, ensuring that annotation consistency is maintained across the entire frame sequence rather than being reset with each new frame.
4Manufacturing precision
If all frames are annotated, then complete object coverage is achieved, but time consumption increases
Solution Approach 1:
The patent applies partial action by annotating only key frames rather than all frames. This selective approach performs annotation on a subset of frames that are sufficient to achieve complete object coverage when combined with cumulative recording, reducing annotation time while maintaining annotation completeness.
Data Source
AI summary
A method for annotation based on a LiDAR map, includes steps of: (a) generating, by a computing device, the LiDAR map and a key frame trajectory using LiDAR point cloud data and a LIDAR SLAM algorithm; (b) annotating, by the computing device, a plurality of key frames included in the key frame trajectory, thereby generating annotation result data; and (c) cumulatively recording, by the computing device, the annotation result data in the LiDAR map.