Transportation Mode Recognition Algorithm Based on Multiple Support Vector Machine Classifiers

Abstract

Transportation mode recognition is a special sub-field of activity recognition, it is also an important user context information in pervasive computing which can not only be used to conduct unobtrusive monitoring of human behavior, but also to provide intelligent information push service. Aiming at the situation that traditional GPS-based transportation mode recognition system suffered from high power consumption, limited usage scenario and modest accuracy of motorized transportation mode recognition, a lightweight transportation mode recognition algorithm with high accuracy which is based on multiple support vector machine classifiers is proposed. The algorithm uses a variety of sensors integrated in intelligent terminals such as acceleration sensors, gyroscope sensors, magnetic sensors, and pressure sensors. Based on the mining of behavior features of vehicular movement patterns, a number of support vector machine classifiers are designed to conduct transportation mode recognition. The experimental results show that the transportation mode recognition algorithm based on multiple support vector machine classifiers has better universality and robustness, and can obtain more than 95% recognition accuracy for motorized transportation mode recognition.