Multi-Camera Gait Analysis for Occlusion-Resistant Skeleton Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gait analysis methods face limitations due to the need for attachable sensors, which restrict analysis to controlled environments and suffer from accuracy issues due to occlusion phenomena, particularly in obtaining kinematic information of hidden body parts.
Innovation Solution
A system utilizing multiple cameras and identifiers to image walking motions, extract 3D and 2D skeleton information, and classify walking parameters using machine learning algorithms to determine normal or abnormal gait patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If attachable sensors (gyro sensor, acceleration sensor) are used for walking analysis, then measurement capability is improved, but ease of operation deteriorates due to the need to attach sensors for every measurement
Solution Approach 1:
The patent replaces the mechanical sensor attachment system with an optical imaging system using multiple cameras. Instead of using gyro sensors and acceleration sensors that require physical attachment to the body, the system uses cameras to capture walking motions and extracts kinematic information through image processing and skeleton extraction algorithms, thereby eliminating the need for sensor attachment while maintaining measurement capability
Solution Approach 2:
The patent creates a virtual copy of the physical walking motion through multiple camera imaging. By capturing images from multiple directions and reconstructing 3D skeleton information, the system creates a digital representation of the walking motion that can be analyzed without physical sensors, replacing the need for attachable measurement devices
2Device complexity
If a single depth camera is used for walking analysis, then device complexity is reduced, but measurement precision deteriorates due to occlusion phenomenon
Solution Approach 1:
The patent divides the single camera system into multiple camera units arranged in a specific configuration. By using multiple cameras to capture images from different directions (front, side, back), the system segments the observation task across multiple devices, allowing each camera to capture visible body parts without occlusion while collectively providing complete kinematic information
Solution Approach 2:
The patent transitions from a single-view (2D) depth camera to a multi-view (3D spatial) camera system. By arranging cameras in three-dimensional space around the walking path and using multiple imaging directions, the system adds spatial dimensions to the observation, enabling capture of body parts that would be occluded in a single view through geometric positioning
3Measurement precision
If multiple cameras are used to capture walking motions from multiple directions, then measurement precision is improved by overcoming occlusion, but device complexity increases
Solution Approach 1:
The patent designs the camera system with universal functionality where multiple cameras work together as an integrated unit. Each camera performs the same function (capturing walking images) but from different positions, and the system includes unified control and processing mechanisms that manage all cameras simultaneously, allowing the multi-camera system to operate as a cohesive whole rather than separate devices
Solution Approach 2:
The patent combines multiple camera systems into a single integrated gait analysis system. By merging the imaging functions of multiple cameras with unified control mechanisms and centralized image processing, the system integrates multiple components into a coordinated whole that functions as a single analysis platform, reducing the operational complexity despite having multiple physical cameras
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system for analyzing walking behavior according to an embodiment of the present disclosure includes: an identifier configured to identify a walker; a walking information obtainer configured to image a walking motion of the walker in at least one or more directions of a left, a right, and a front on the basis of an identification signal from the identifier, and to extract and provide kinematic information from the taken images; an image combiner configured to combine kinematic information (3D and/or 2D skeleton information) of the taken walking images; a walking parameter extractor/classifier configured to extract at least one or more walking parameter on the basis of the combined kinematic information (3D and/or 2D skeleton information), and then classify the kind of the walking; and a walking result informer configured to determine whether the at least one or more walking parameters come out of a recommendation reference value, or are included in a walking parameter range of abnormal walking, or are classified as a pathological gait, and then provide the result to the walker or an expert.