An Unsupervised Learning-Based Spatial Co-Location Detection System from Low-Power Consumption Sensor.

Area of Science:

• Computer Science
• Geographic Information Science
• Sensor Networks

Background:

• Spatial co-location detection identifies objects in geographic proximity, often using mobile devices.
• Existing methods rely on GPS data, which is accurate but power-inefficient for this task.
• Absolute geographic location is unnecessary for inferring co-location.

Purpose of the Study:

• To propose and implement an unsupervised learning algorithm for spatial co-location detection.
• To utilize low-power consumption sensor data, specifically magnetometer readings, for co-location inference.
• To develop an efficient alternative to GPS-based methods for detecting human proximity.

Main Methods:

• Implementation of a convolutional autoencoder, an unsupervised learning algorithm.
• Training the model to infer co-location from magnetometer readings.
• Utilizing the Structural Similarity (SSIM) index (threshold > 0.5) to validate co-location inference.

Main Results:

• The proposed convolutional autoencoder effectively infers spatial co-location.
• The system successfully recognizes co-location using only magnetometer data.
• The method demonstrates the feasibility of using low-power sensors for co-location tasks.

Conclusions:

• Unsupervised learning with magnetometer data offers an efficient solution for spatial co-location detection.
• This approach reduces the reliance on power-intensive GPS for proximity sensing.
• The study validates the effectiveness of the convolutional autoencoder for inferring human co-location from sensor data.