3D ROI Object Identification for Occlusion-Aware LIDAR Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object identification systems face reduced accuracy due to occlusion in captured images, particularly when using laser radar and feature-based identification methods.
Innovation Solution
The system sets a three-dimensional Region Of Interest (ROI) using reflection point data from a laser radar, employing modified clustering techniques to integrate overlapping clusters and utilize positional relationships among spaces of interest to identify and exclude non-matching clusters, thereby improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature-based object identification is used in captured images, then object identification can be performed, but accuracy is reduced when occlusion occurs
Solution Approach 1:
The patent transitions from two-dimensional image-based identification to three-dimensional space-based identification by constructing a three-dimensional region of interest (ROI) from laser radar reflection points. This dimensional expansion allows the system to capture occluded objects that are invisible in 2D images, resolving the accuracy reduction caused by occlusion.
Solution Approach 2:
The patent introduces laser radar reflection point data as an intermediary between the captured image and the object identification process. This intermediary provides three-dimensional spatial information that complements the two-dimensional image data, enabling accurate identification even when the object is occluded in the image.
2Productivity
If traditional clustering is applied to reflection points, then object detection can be performed, but overlapping clusters reduce identification accuracy
Solution Approach 1:
The patent merges multiple overlapping clusters by determining their positional relationships and integrating them into a unified three-dimensional region of interest. This merging process resolves the issue of overlapping clusters by combining their spatial information to form a comprehensive ROI that accurately represents the target object.
Solution Approach 2:
The patent resolves cluster overlap by transitioning from two-dimensional cluster analysis to three-dimensional spatial analysis. By constructing a three-dimensional ROI that incorporates depth information from laser radar, the system can distinguish and integrate overlapping clusters that appear merged in 2D projections.
3Measurement precision
If three-dimensional ROI is constructed using laser radar reflection points, then occlusion accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary clustering and spatial relationship determination on laser radar reflection points before constructing the final three-dimensional ROI. This preliminary processing organizes the data in advance, reducing the computational burden during the actual ROI construction and object identification phases.
Solution Approach 2:
The patent segments the complex task of three-dimensional ROI construction into distinct steps: reflection point acquisition, clustering, positional relationship determination, and ROI construction. This segmentation reduces computational complexity by breaking down the overall process into manageable, independent stages that can be processed sequentially.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances object identification accuracy by accurately determining occlusion and integrating relevant clusters, leading to improved feature-based identification even in occluded conditions.
Implementation Method 1
a laser radar 1 to set a three-dimensional region of interest (ROI) 20 based on reflection point data acquired by using the laser radar 1
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides an object identification apparatus, an object identification method, and a control program that improve the accuracy of an object identification. A data conversion processing unit (12) converts a second group including a plurality of reflection point data units in which a reflection point corresponding to each reflection point data unit belongs to a three-dimensional object among a first data unit group into a third group including a plurality of projection point data units by projecting the second group onto a horizontal plane in a world coordinate system. A clustering processing unit (13) clusters the plurality of projection point data units of the third group into a plurality of clusters based on positions of these units on the horizontal plane. A space of interest setting unit (14) sets a space of interest for each cluster by using the plurality of reflection point data units corresponding to the plurality of projection point data units included in each cluster.