LiDAR Voxel Mapping for Real-Time Moving-Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lidar sensor-based map generation techniques are slow and difficult to update in real time, and existing systems struggle to detect objects of various shapes effectively.
Innovation Solution
A method involving voxel division of lidar data into cells, calculating point occupation frequency changes, and generating dynamic maps using background cells with a predetermined threshold, combined with a pre-trained neural network for object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional map generation techniques are used, then map accuracy can be maintained, but the processing time increases and real-time updates become difficult
Solution Approach 1:
The patent divides the three-dimensional space into multiple voxels, and further divides each voxel into multiple cells. This segmentation allows the system to process only relevant portions of the point cloud data rather than all points, significantly reducing processing time while maintaining map accuracy through selective point occupation frequency calculation.
Solution Approach 2:
The patent calculates point occupation frequency changes only for cells that contain points or are adjacent to cells containing points, rather than processing all cells in the voxel space. This partial action approach reduces computational load and processing time while maintaining sufficient map accuracy for dynamic environments.
2Measurement precision
If deep-learning techniques are used for object detection, then human detection accuracy improves, but the system cannot effectively detect objects of various shapes
Solution Approach 1:
The patent creates a universal detection framework that works for objects of any shape by representing objects as collections of foreground cells in three-dimensional space. The object determination unit can identify various object types (humans, animals, vehicles, etc.) using the same cell-based representation and clustering approach, making the system versatile across different object shapes and sizes.
Solution Approach 2:
The patent changes the detection parameter from learned human attributes to point occupation frequency changes and cell occupation patterns. By monitoring changes in cell occupation over time and using clustering based on spatial relationships between cells, the system can detect objects of various shapes without relying on shape-specific deep learning models.
Data Source
AI summary
A method for generating a dynamic map, which is performed by a dynamic map generation apparatus, is provided. The method includes arranging point data acquired from a lidar sensor in a plurality of cells included in a voxel, calculating a point occupation frequency change of each of the plurality of cells according to an attribute and a status of each of the plurality of cells, and generating the dynamic map by using, among the plurality of cells, background cells in each of which a value accumulating the point occupation frequency change is equal to or greater than a predetermined threshold.