Transport Mode Estimation Using Dual-Branch Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies require six sensors installed in every smartphone to predict a user's mode of transport, making the method inefficient and costly.
Innovation Solution
A machine learning technology that uses a learning model with a first network composed of a first and second branch, and a second network to estimate a user's mode of transport from their movement trajectory and derived movement information, without the need for multiple sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If six sensors are installed in every smartphone to predict mode of transport, then prediction accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts the mode of transport prediction function from the smartphone hardware and relocates it to a cloud-based server. Instead of requiring six sensors in every smartphone, the system uses a single smartphone to collect movement trajectory data and transmits it to the server, which performs the complex prediction using a learned model. This extraction eliminates the need for multiple sensors in mobile devices while maintaining prediction accuracy.
Solution Approach 2:
The patent introduces a server as an intermediary between the smartphone and the prediction function. The server receives movement trajectory data from the smartphone, processes it through the learned model, and outputs the mode of transport prediction. This intermediary handles the computational complexity and sensor requirements, allowing smartphones to remain simple while still achieving accurate predictions.
2Measurement precision
If six sensors are installed in every smartphone to predict mode of transport, then prediction accuracy is improved, but manufacturing cost increases
Solution Approach 1:
The patent extracts the expensive sensor components from the smartphone and relocates the prediction functionality to a server. This eliminates the need for manufacturers to equip every smartphone with six sensors, significantly reducing component costs and assembly complexity while maintaining prediction accuracy through the server-based learned model.
Solution Approach 2:
The patent creates a digital copy of the prediction function in the form of a learned model stored on the server. Instead of physically installing sensors in every device, the system uses this digital model to perform predictions remotely, eliminating hardware costs while maintaining functional equivalence.
3Measurement precision
If movement trajectory data is collected and processed to predict mode of transport, then prediction accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent extracts the complex data processing and prediction functions from the smartphone and relocates them to a server. The smartphone only performs simple data collection and transmission, while the server handles the complex processing of movement trajectory data through the learned model, significantly reducing the processing burden on mobile devices.
Solution Approach 2:
The patent introduces a server as an intermediary that handles the complex data processing tasks. The server receives raw movement trajectory data from the smartphone, processes it through the learned model, and generates predictions. This intermediary absorbs the data processing complexity, allowing smartphones to remain simple while achieving accurate predictions.
Data Source
AI summary
An information processing apparatus acquires a movement trajectory of a user, derives, from the movement trajectory, movement information indicating features relating to movement, and estimates, using a learning model, a mode of transport of the user from the movement trajectory and the movement information, wherein the learning model includes a first network, which is composed of a first branch and a second branch, and a second network that follows the first network, the first branch generates feature amounts of the movement trajectory from the movement trajectory, the second branch generates feature amounts of the movement information from the movement information, and the second network is configured to generate combined feature amounts by combining the feature amounts of the movement trajectory and the feature amounts of the movement information, and output data indicating the mode of transport of the user from the combined feature amounts.