Image Processor Action Detection Using Articulation Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle to accurately detect the actions of individuals in need of nursing care, particularly due to the lack of consideration for posture and the inability to differentiate between static and dynamic objects, leading to false detections and reduced accuracy.
Innovation Solution
A system that includes a non-transitory recording medium and image processor capable of obtaining captured images, detecting person and object areas, estimating articulation points, and determining actions based on these points and object areas, utilizing deep neural networks to enhance detection accuracy and handle moving objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If action detection is based on overlapping width between boundary and person area only, then detection process is simple, but false detection occurs when person merely stands in vicinity
Solution Approach 1:
The detection process is segmented into multiple independent analysis stages: (1) detecting person area and object area separately, (2) estimating articulation point positions independently, (3) analyzing posture based on articulation points, and (4) determining action based on combined information. This segmentation allows each stage to focus on specific features, improving overall detection accuracy while maintaining manageable complexity.
Solution Approach 2:
The invention transitions from two-dimensional area overlap analysis to three-dimensional spatial relationship analysis by estimating articulation point positions. This adds the dimension of body posture analysis, enabling differentiation between mere proximity and actual interaction with objects, thereby reducing false detections.
2Ease of manufacture
If action detection is based on fixed object relationships only, then system is simple to implement, but cannot cope with moving objects
Solution Approach 1:
The detection system is designed with universal applicability to both fixed and moving objects. The articulation point estimation and spatial relationship analysis methods work equally well regardless of object motion state, enabling the system to handle diverse scenarios including walking assistance, falling detection, and object interaction without requiring separate detection mechanisms.
Solution Approach 2:
The system incorporates dynamic object tracking capabilities by continuously estimating articulation point positions across multiple frames. This allows the system to adapt to moving objects by analyzing temporal changes in spatial relationships, while maintaining the same fundamental detection approach used for fixed objects.
3Power
If posture information is not considered, then detection computation is reduced, but false detection of actions occurs
Solution Approach 1:
The invention extracts key posture information by focusing specifically on articulation point positions rather than analyzing the entire person area. This selective extraction of critical features provides sufficient posture information for accurate action detection while avoiding the computational burden of processing all pixel data within the person area.
Solution Approach 2:
The system changes the detection parameters from area-based metrics to point-based articulation position metrics. This parameter transformation enables efficient computation through coordinate geometry calculations while capturing essential posture information needed for reliable action differentiation.
Data Source
AI summary
A non-transitory recording medium storing a computer readable image processing program that causes a computer to: obtain a captured image; detect, from the obtained image, a person area representing a person and an object area representing a predetermined object; detect positions of articulation points of the person from the detected person area; and estimate an action of the person based on the detected positions of the articulation points and the detected object area.