Research And Implementation Of Terrain Classification Based On Visual And Force Signal

With the rapid development of national economy and social advances in science and technology, driverless technology and vehicle passing ability evaluation have drawn attention of scholars around the world. Fast and accurate terrain classification is the premise study of vehicle passing ability evaluation. Classifying terrain is useful for vehicle to identify the driving condition and helps drivers make the best driving strategies. Common methods of the classification include Vibration-based approach, Lidar-based approach and Vision-based approach. Based on the research previously, The paper applied force signal and visual signal integration to achieve fast terrain classification. The paper used technology of visual image preprocessing and force signal analysis in frequency domain and time-frequency domain to extract features in visual images and wheel force signals. Dimensionality reduction technology and Category Training Recognition were also used in the classification. In order to get better results, the paper extracted both color and texture features of the images. As part of the force signal processing, signal was denoised before features extraction. PCA and SVM classifier were applied in dimensionality reduction and category training. For SVM classifier the paper also taken genetic algorithm to optimize the parameters of the corresponding classification. Finally, the paper selected several common terrain to verify the accuracy and reliability of the above-described algorithm. The results showed fusion algorithm achieved desired accuracy.