Mobile Activity Recognition Using 1D CNN Pose Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mobile terminals provide limited navigation and positioning services, primarily focusing on direct positioning, routing, and navigation, which restricts user experience.
Innovation Solution
An activity recognition method using one-dimensional convolutional layers to extract feature data from electronic device poses, recognizing user activities such as motion states, actions, and vehicle types, and displaying these as icons on a map interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If mobile terminal provides only basic positioning and navigation services, then the system complexity is low, but the user experience is limited
Solution Approach 1:
The patent applies multi-functionality by enabling the mobile terminal to perform both traditional navigation services and activity recognition functions using the same device sensors. The activity recognition module processes sensor data to identify user activities (walking, running, vehicle transport) and integrates this information with map services, allowing the device to serve multiple purposes without requiring separate specialized equipment
Solution Approach 2:
The patent segments the navigation system into distinct functional modules: a traditional positioning/navigation module and a new activity recognition module. The activity recognition module further segments analysis into multiple dimensions (state of motion, action, vehicle transporting the person), with each dimension processed independently through separate neural network branches before being integrated into a comprehensive activity state description
2Adaptability or versatility
If activity recognition is added to map function, then user experience is enhanced, but processing complexity increases
Solution Approach 1:
The patent divides the activity recognition processing into three independent neural network branches, each responsible for analyzing one dimension of activity (state of motion, action, vehicle transporting). This segmentation allows parallel processing of different feature types without interference, reducing the computational complexity compared to a single monolithic model while maintaining comprehensive activity recognition
Solution Approach 2:
The patent introduces an activity recognition module as an intermediary component between the sensor data input and the map display output. This module processes raw sensor data through multiple neural networks, extracts meaningful activity features, and presents processed activity state information to the map interface, thereby decoupling the complexity of activity analysis from both sensor input and display output systems
Data Source
AI summary
Embodiments of an activity recognition method, a display method, and an electronic device are disclosed. The electronic device can obtain a plurality of poses of the electronic device, and obtain one or more pieces of feature data based on the plurality of poses by using a plurality of one-dimensional convolutional layers. Each of the one or more pieces of feature data indicates a state of motion of a person holding the electronic device, an action of the person, or a vehicle transporting the person. Then, the electronic device displays, as an icon, the types of information indicated by the one or more pieces of feature data. According to the method, feature data for representing an activity state of the person is extracted by using the one-dimensional convolutional layers, and features for representing the activity state of the person may be extracted by using the plurality of one-dimensional convolutional layers.