基于声信号的履带机器人地面分类试验研究

In order to extend terrain classification methods and improve accuracy, a terrain classification method based on acoustic signal was proposed. An acoustic pressure sensor was installed to acquire acoustic signal resulted from tracked robot-terrain interaction. The modified MFCC+ΔMFCC feature vector was extracted. Finally, a tuned support vector machine (SVM) was adopted to perform classification. The results indicate that the information carried by the acoustic signal is able to characterize terrain type. The modified MFCC+ΔMFCC feature vector is obviously superior to features extracted from amplitude domain, frequency domain and time-frequency domain. The highest accuracy of 89. 5% is achieved in campus environment. When the SNR is higher than 20 dB, accuracies around 80% can be achieved in various background environments. Acoustic-based method is proved to be effective in terrain classification application.