Upper Body Motion Tracking Using Three IMUs and HMM Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for tracking biomechanical movements, especially small movements like arm and hand movements, using motion sensors are cumbersome and prone to errors due to the need for multiple sensors and are affected by occlusion and environmental interference.
Innovation Solution
The use of three inertial measurement units (IMUs) attached to the wrists and torso, combined with a hidden Markov Model filter, to track upper body movements without additional hardware, allowing for precise tracking of arm movements while the body is in motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a plethora of motion sensors (at least 7 inertial/magnetic sensors) are attached to multiple joints to track upper body motion, then measurement precision is improved, but device complexity and ease of operation deteriorate due to cumbersome setup and inhibition of natural movements
Solution Approach 1:
The patent segments the sensing system into only three essential IMU units positioned at key body locations (two wrists and one torso), eliminating the need for sensors at every joint. This segmentation maintains sufficient tracking precision by focusing measurements on critical motion points while dramatically reducing the total sensor count from 7+ to just 3 sensors.
Solution Approach 2:
Each IMU unit serves multiple functions: tracking positional movement, orientation, and acceleration simultaneously. The three IMUs collectively provide comprehensive upper body motion tracking data that would otherwise require numerous specialized sensors, achieving multi-functionality with minimal hardware.
2Measurement precision
If conventional motion sensors are used to track small movements like arm and hand movements, then measurement capability is improved, but reliability deteriorates due to accumulation of errors from large amounts of measurement noise
Solution Approach 1:
The patent implements feedback through the hidden Markov model filter that continuously processes IMU data streams, comparing predicted motion states with actual measurements. This feedback mechanism identifies and corrects error accumulations in real-time, maintaining reliability for small movement detection despite the inherent noise in accelerometer and gyroscope measurements.
Solution Approach 2:
The hidden Markov model acts as an intermediary between raw sensor data and final motion tracking results. It mediates the noisy IMU measurements by incorporating motion models and probability distributions to filter out errors and produce reliable position estimates, effectively separating signal from noise.
3Measurement precision
If computer vision or external localization systems are used for tracking biomechanical movements, then measurement precision is improved, but device complexity and ease of operation worsen due to extensive hardware setups and limited coverage areas
Solution Approach 1:
The system uses self-service by leveraging the smartphone's existing IMU sensors and processing capabilities already present in the device. No external cameras, localization infrastructure, or additional hardware setups are required - the smartphone serves itself to perform motion tracking, eliminating complex external hardware dependencies while maintaining tracking accuracy.
Solution Approach 2:
The patent replaces mechanical/optical tracking systems (cameras, external sensors) with an inertial sensing approach using the smartphone's built-in IMU. This substitution eliminates the need for extensive mechanical hardware setups and external infrastructure, using purely inertial measurements combined with computational modeling to achieve movement tracking.
4Ease of operation
If machine learning techniques are applied to sensor data to identify activities, then ease of operation is improved, but measurement precision deteriorates because only coarse semantic meaning is identified without precise location trajectory
Solution Approach 1:
The patent segments the analysis into two distinct components: activity classification (coarse semantic meaning) and position estimation (fine-grained location trajectory). While machine learning handles activity identification, a separate hidden Markov model specifically addresses position estimation, ensuring both coarse and fine-level information are preserved without one compromising the other.
Solution Approach 2:
The patent merges machine learning-based activity recognition with hidden Markov model-based position estimation into a unified framework. Both techniques operate simultaneously on the same IMU data stream, combining their strengths to provide both semantic activity labels and precise location trajectories without interfering with each other's precision.
Data Source
AI summary
Disclosed methods, systems, and storage media may track body movements and movement trajectories using internal measurement units (IMUs), where a first IMU may be attached to a first wrist of a user, a second IMU may be attached to a second wrist of the user, and a third IMU may be attached to a torso of the user. Upper body movements may be derived from sensor data produced by the three IMUs. IMUs are typically not used to detect fine levels of body movements and/or movement trajectory because most IMUs accumulate errors due to large amounts of measurement noise. Embodiments provide arm and torso movement models to which the sensor data is applied in order to derive the body movements and/or movement trajectory. Additionally, estimation errors may be mitigated using a hidden Markov Model (HMM) filter. Other embodiments may be described and/or claimed.