3D Point Cloud Integration for Human Tracking Stability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human-tracking technologies in smart retail environments face instability and resource inefficiency due to reliance on 2D detection and ReID methods, which struggle with occlusion, crisscrossing, and cumulative errors, especially when sample sizes are small, leading to inaccurate tracking over time.
Innovation Solution
The implementation of a human-tracking method using point clouds from binocular cameras, where human point clouds are integrated into a total point cloud for robust 3D tracking, reducing dependence on 2D tracking and ReID, and utilizing clustering algorithms for accurate location information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If 2D detection and ReID methods are used for human tracking, then the system can operate with simpler hardware, but tracking stability and accuracy deteriorate due to occlusion, crisscrossing, and cumulative errors
Solution Approach 1:
The patent transitions from 2D detection and tracking to 3D point cloud-based tracking. By capturing spatial information in three dimensions and representing humans as point clouds with depth information, the system overcomes the limitations of 2D methods where occlusion and crisscrossing cause tracking failures. The 3D spatial relationships provide robust tracking even when objects overlap in 2D view.
2Use of energy by moving object
If 2D tracking with ReID is used, then resource consumption is lower, but tracking accuracy deteriorates over time due to cumulative errors
Solution Approach 1:
The system adopts 3D point cloud representation instead of 2D bounding boxes and ReID features. The point clouds contain rich spatial distribution information that enables more accurate tracking without relying on cumulative ReID matching. This dimensional enhancement provides direct geometric constraints for tracking, reducing cumulative errors.
Solution Approach 2:
The patent replaces the computational ReID matching mechanism with a geometric point cloud matching approach. Instead of comparing appearance features frame by frame, the system uses spatial point cloud data and clustering algorithms to track objects, substituting a feature-matching system with a geometry-based system that is more accurate and less prone to cumulative errors.
3Reliability
If point cloud integration is performed for 3D tracking, then tracking accuracy and stability improve, but device complexity and computational resources increase
Solution Approach 1:
The patent segments the complex tracking task into distinct modules: point cloud acquisition from multiple cameras, point cloud integration/merging, clustering algorithm application, and tracking result generation. This segmentation allows each module to be optimized independently and simplifies the overall system architecture despite the increased capability.
Solution Approach 2:
The system introduces point cloud data as an intermediary representation between image capture and tracking output. Instead of directly processing images or features, the patent uses point clouds as a mediating data structure that preserves 3D spatial information while being amenable to clustering and integration operations, thereby managing complexity through a useful intermediate representation.
4Reliability
If multiple cameras are used for point cloud acquisition, then tracking robustness improves, but device complexity and data processing load increase
Solution Approach 1:
The patent merges point clouds from multiple cameras into a unified 3D representation. By integrating the point cloud data in a common coordinate system and applying clustering algorithms, the system combines information from multiple sources to improve tracking robustness while managing the data volume through efficient merging operations rather than processing each camera's data separately.
Data Source
AI summary
The disclosure provides human-tracking methods, systems, and storage media. The method includes: acquiring a plurality of human point clouds of a current frame from a plurality of cameras; generating a total point cloud of the current frame by integrating the plurality of human point clouds of the current frame; acquiring a plurality of human point clouds of a next frame from the plurality of cameras; acquiring a total point cloud of the next frame by integrating the plurality of human point clouds of the next frame; and performing human tracking based on the total point cloud of the current frame and the total point cloud of the next frame.