Skeleton Model Joint Assignment via Sensor Location Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for building skeleton models using depth cameras struggle to accurately distinguish between left and right body parts in clinical monitoring scenarios, particularly when patients are lying on their side, leading to unreliable patient activity monitoring.
Innovation Solution
A device and system that includes a joint identification unit, pose estimation unit, sensor location unit, assignment unit, and skeleton modeling unit, which use depth image data, sensor location data, and machine learning algorithms to accurately assign joints to their respective body locations, even in complex postures, by combining joint location data with sensor location information to disambiguate left and right body parts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If depth camera based skeleton model methods are used, then gaming applications can track body movements, but the methods fail in clinical monitoring scenarios where patients lie in sleeping poses
Solution Approach 1:
The patent applies local quality by using sensor location data from specific body locations to disambiguate corresponding joints in the skeleton model. Instead of treating all joints uniformly, the system locally enhances specific joint assignments by referencing sensor positions on the patient's body, thereby improving reliability in lateral poses where left-right distinction is critical
Solution Approach 2:
The patent introduces sensor location data as an intermediary element that mediates between the depth camera image data and the skeleton model generation. This intermediary provides additional spatial information about sensor positions on the patient's body, which helps resolve ambiguities in joint assignment without requiring changes to the depth camera system itself
2Adaptability or versatility
If the patient is lying on a side of the body, then the monitoring scenario is more realistic for clinical use, but the joints in left and right parts of the body become confused and indistinguishable
Solution Approach 1:
The patent applies preliminary action by obtaining sensor location data before generating the skeleton model. This pre-acquired information about sensor positions on the patient's body is used to guide the joint assignment process, allowing the system to preemptively resolve left-right ambiguities before they affect the overall skeleton model accuracy
Solution Approach 2:
The system enhances local joint identification precision by using sensor location information specific to each body region. Rather than applying a uniform approach to all joints, the method locally improves identification accuracy at joints corresponding to sensor locations, where the ambiguity is most critical
3Productivity
If existing skeleton model methods are used, then the process is simple and fast, but the assignment of joints to body locations is unreliable in complex postures
Solution Approach 1:
The patent merges two data sources: depth camera image data and sensor location data. By combining these complementary information sources, the system maintains the efficiency of automated depth-based skeleton modeling while enhancing reliability through the additional spatial constraints provided by sensor positions
Solution Approach 2:
The system uses sensor location data as feedback to refine and correct joint assignments in the skeleton model. This feedback mechanism allows the system to verify and adjust joint-to-body-location mappings based on the known positions of sensors on the patient's body, improving reliability without significantly increasing processing time
Data Source
AI summary
The present invention relates to a device, system, method and computer program for providing a skeleton model, wherein the device comprises a joint identification unit configured to obtain an image and corresponding image data of the patient comprising depth information and to generate joint location data by localizing one or more joints of the patient in said image, a pose estimation unit configured to generate pose estimation data by estimating a pose of the patient using the joint location data and/or the image data, a sensor location unit configured to obtain body location data, comprising information about a location of a sensor on the patients body, and image location data, comprising information about the location of the sensor in the image, and to generate sensor location data, assigning a sensor location in the image to a body location of the patient, based on the body location data and the image location data, an assignment unit configured to perform an assignment of the one or more joints to one or more body locations of the patient by using the joint location data, the pose estimation data and the sensor location data, and a skeleton modelling unit configured to generate a skeleton model of the patient based on the assignment of the joints to a body location.