Real-Time Transportation Mode Detection Using Smartphones and Artificial Neural Networks: Performance Comparisons Between Smartphones and Conventional Global Positioning System Sensors

Traditionally, traffic monitoring requires data from traffic cameras, loop detectors, or probe vehicles that are usually operated by dedicated employees. In efforts to reduce the capital and operational costs associated with traffic monitoring, departments of transportation have explored the feasibility of using global positioning system (GPS) data loggers on their probe vehicles that are postprocessed for analyzing the traffic patterns on desired routes. Furthermore, most cell phones are equipped with embedded assisted-GPS (AGPS) chips, and if the mode of transportation the phone is in can be anonymously identified, the phones can be treated as if they are probe vehicles that are voluntarily hovering throughout the city, at minimal additional costs. Emerging cell phones known as “smartphones” are equipped with additional sensors including an accelerometer and magnetometer. The accelerometer can directly measure the acceleration values, as opposed to having acceleration values derived from speed values in conventional GPS sensors. The magnetometer can measure mode-specific electromagnetic levels. Smartphones are subscribed with roadside Internet data plans that can provide an essential platform for real-time traffic monitoring. In this article, neural network-based artificial intelligence is used to identify the mode of transportation by detecting the patterns of distinct physical profile of each mode that consists of speed, acceleration, number of satellites in view, and electromagnetic levels. Results show that newly available values in smartphones improve the mode detection rates when compared with using conventional GPS data loggers. When smartphones are in known orientations, they can provide three-dimensional (3-D) acceleration values that can further improve mode detection accuracies.