ML Indoor Localization Using RSSI Fingerprints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization systems, such as GPS, are ineffective in indoor environments due to satellite obstructions, and existing indoor localization systems are complex and require expert setup.
Innovation Solution
A machine learning method using received signal indicator data from wireless devices to estimate location, involving feature vector generation, model training, and location estimation, with optional clustering and ensemble algorithms, to provide accurate indoor positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS satellite geolocation systems are used, then location determination is achieved in outdoor environments, but the system becomes ineffective in indoor environments due to satellite obstructions
Solution Approach 1:
The patent introduces wireless access points and mobile hotspots as intermediary devices to enable location determination in indoor environments. These intermediaries relay location information when direct satellite signals are blocked, allowing the system to function across both outdoor and indoor settings without requiring a complete system change.
2Reliability
If traditional indoor localization systems such as trilateration are implemented, then indoor location determination is achieved, but the system becomes complex and requires expert setup and reference measurements
Solution Approach 1:
The system automatically discovers available wireless access points and mobile hotspots using standard wireless communication protocols, eliminating the need for manual reference measurement collection by experts. The machine learning model is trained automatically on collected signal data, and the system self-configures without requiring specialized setup procedures, making it as easy to deploy as any wireless network.
3Measurement precision
If machine learning models are trained on collected signal data, then accurate location estimation is achieved, but the training process requires significant computational resources and time
Solution Approach 1:
The system performs model training in advance during an offline phase using collected training data from multiple locations. This preliminary action separates the time-consuming training process from real-time operation, so that during actual use, the pre-trained model provides rapid location estimates without requiring additional training time.
Solution Approach 2:
The patent divides the feature vectors into training sets and test sets, and further segments the training process by implementing clustering that divides the environment into multiple zones with separate models. This segmentation allows parallel processing and reduces the computational burden on any single model, enabling faster overall training and deployment.
Data Source
AI summary
Machine learning method and systems for estimating a location of a target wireless device in an environment are disclosed. A machine learning method comprises: receiving a plurality of training received signal indictor data sets for discrete locations in the environment, each training received signal data set comprising received signal indicator values and corresponding wireless transmitter identifiers for wireless signals received by a test wireless device at a respective discrete location; generating feature vectors from the received signal indicator data sets; training a machine learning model using the feature vectors to obtain a trained machine learning model; receiving a target received signal data set from the target wireless device, the target received signal data set comprising signal indicator values and corresponding wireless transmitter identifiers for wireless signals received by the target wireless device; generating a target feature vector from the target received signal data set; and estimating a location of the target wireless device as a discrete location output by the trained machine learning model in response to the target feature vector.