Wearable Camera Activity Recognition via Micro-Activity Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current camera-based wearables face challenges in accurately monitoring and identifying complex human activities involving hands and objects, such as pill-taking, due to variations in sequences, distractions, and individual differences.
Innovation Solution
A wearable device equipped with a camera and processor that uses machine learning models to identify objects and detect micro-activities by focusing on specific regions of attention, allowing for efficient inference of overall activities, with a heuristic understanding engine to evaluate success and handle distractions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If camera-based wearables are used to monitor human activities, then activity monitoring capability is improved, but accuracy in identifying complex hand activities deteriorates due to variations in sequences, distractions, and individual differences
Solution Approach 1:
The patent segments complex hand activities into discrete micro-activities (e.g., grasp, release, hold) that can be individually detected and classified. This segmentation allows the system to break down complex sequences into manageable units, improving identification accuracy despite variations in execution sequences and individual differences.
Solution Approach 2:
The system changes parameters by focusing on specific regions of interest (ROI) in the video feed rather than analyzing the entire frame. This parameter change in the analysis approach allows the system to concentrate computational resources on relevant areas, improving detection accuracy while handling distractions and variations more effectively.
2Productivity
If the system focuses on specific regions of attention to reduce search space, then object identification efficiency is improved, but the complexity of the machine learning model increases
Solution Approach 1:
The system performs preliminary actions by pre-defining regions of interest based on hand position and movement patterns before conducting detailed object identification. This preliminary focusing reduces the search space for the machine learning model, improving identification efficiency while managing model complexity through staged processing.
3Adaptability or versatility
If the system monitors complex sequences of micro-activities, then activity detection capability is improved, but the system's ability to handle distractions and disturbances deteriorates
Solution Approach 1:
The system implements feedback mechanisms where detected micro-activities are continuously evaluated and used to adjust the search focus and model attention. This feedback loop allows the system to maintain reliability under distraction by dynamically adapting to the current scene and distinguishing relevant activities from irrelevant disturbances based on contextual information.
Data Source
AI summary
A system includes a wearable device including a camera. The system further includes at least one processor that can identify objects from video data generated by the camera and monitor how an individual wearing the wearable device manipulates the objects according to predetermined micro-activities of interest to infer an action by the individual.