Environment Recognition Based on Temporal Filtering SVM

Abstract

Since the signal quality of global navigation satellite system (GNSS) is extremely vulnerable to the surrounding environment, the environment-aware adaptive positioning algorithm has drawn wide attention. In order to select the suitable positioning method in different types of environment, the receiver need to recognize the type of surrounding environment in real-time. Targeting on the vehicle positioning applications in the city, this paper divides the urban environment into six categories: canyon, downtown, suburb, viaduct-up, viaduct-down and boulevard, and proposes a novel environment recognition algorithm based on the navigation signal characteristics. Firstly, a five dimension signal feature vector is proposed to describe the quality of navigation signal. The vector elements are signal power attenuation mean, power attenuation standard deviation, signal blocking coefficient, DOP value expansion ratio and power fluctuation coefficient. Then, taking this vector as environmental attribute, this paper proposes an environment recognition algorithm based on the temporal filtering support vector machine (SVM). In the experiment, the raw navigation signal data are collected for more than 100 thousand epochs in six types of urban environment, with no less than 10 thousand epochs for each type. In order to verify the validity of the proposed recognition algorithm, the five cross validation method is used to train and test all the collecting data. The testing results show that the recognition accuracy of the algorithm are higher than 90% for all types of environment.