Recurrent Neural Network Action Recognition Without Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human body action recognition technologies rely heavily on static image labeling and sensor data, limiting their applicability and universality, as models trained for specific actions struggle to generalize to other actions and scenarios, making them less economical and versatile for human body action monitoring.
Innovation Solution
A method utilizing a trained recurrent neural network model to recognize organism actions by processing body feature information from successive image frames, eliminating the need for sensor data by deriving feature information from image analysis, and enabling adaptability across different scenes and actions through bidirectional recurrent neural networks and convolutional neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor data (acceleration sensor, gyroscope) is used for human body action detection, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent replaces mechanical sensor systems (acceleration sensors, gyroscopes) with an optical-based computer vision system using image processing and neural networks to detect and recognize human body actions, thereby reducing device complexity while maintaining measurement precision
Solution Approach 2:
The patent introduces an intermediary processing system consisting of feature extraction modules and neural network models that transform raw image data into actionable insights, replacing the need for direct sensor attachment and complex sensor data processing
2Manufacturing precision
If models are trained for specific actions based on static images, then manufacturing precision is improved, but adaptability deteriorates
Solution Approach 1:
The patent transitions from static image-based models to dynamic video sequence processing using recurrent neural networks, enabling the system to adapt to different actions and scenarios while maintaining high recognition precision through temporal feature analysis
Solution Approach 2:
The patent develops a universal action recognition framework that can handle multiple types of actions and scenarios through a single trained model, eliminating the need for separate models for each specific action and significantly improving adaptability
Data Source
AI summary
The present disclosure relates to a method and an apparatus for recognizing an organism action, a server and a storage medium. The method includes: obtaining body feature information of an organism corresponding to time-successive multiple frames of images; creating a feature information sequence, wherein the feature information sequence includes body feature information respectively corresponding to multiple frames of images which are arranged according to a time sequence of the multiple frames of images; and inputting the feature information sequence into a trained recurrent neural network model, and determining an action of the organism corresponding to the feature information sequence according to an output of the recurrent neural network model.