Figure - available from: Wireless Communications and Mobile Computing
This content is subject to copyright. Terms and conditions apply.
Source publication
For terrain classification tasks, previous methods used a single scale or single model to extract the features of the image, used high-to-low resolution networks to extract the features of the image, and used a network with no relationship between channels. These methods would lead to the inadequacy of the extracted features. Therefore, classificat...
Citations
... Various technologies were applied in relevant studies to identify the terrain type using mobile robots. Methods based on LiDAR [3] and camera [4][5][6][7][8] data are widely used, but these systems require embedded systems with high computational capacity, and they also have high costs. Beside the large amount of data which have to be processed, complex classification algorithms must be used, such as convolutional neural networks [8]. ...
Terrain classification provides valuable information for both control and navigation algorithms of wheeled mobile robots. In this paper, a novel online outdoor terrain classification algorithm is proposed for wheeled mobile robots. The algorithm is based on only time-domain features with both low computational and low memory requirements, which are extracted from the inertial and magnetic sensor signals. Multilayer perceptron (MLP) neural networks are applied as classifi-ers. The algorithm is tested on a measurement database collected using a prototype measurement system for various outdoor terrain types. Different datasets were constructed based on various setups of processing window sizes, used sensor types, and robot speeds. To examine the possibilities of the three applied sensor types in the application, the features extracted from the measurement data of the different sensors were tested alone, in pairs and fused together. The algorithm is suitable to operate online on the embedded system of the mobile robot. The achieved results show that using the applied time-domain feature set the highest classification efficiencies on unknown data can be above 98%. It is also shown that the gyroscope provides higher classification rates than the widely used accelerometer. The magnetic sensor alone cannot be effectively used but fusing the data of this sensor with the data of the inertial sensors can improve the performance.
... However, due to its low computational efficiency and slow learning speed, it is easy to fall into local minimums [16]. To improve the accuracy of traditional machine learning classifiers, these weak classifiers can be integrated, and the accuracy of classification results can be improved using integrated learning [17]. ...
Terrain recognition exerts an extremely important role in outdoor mobile robot gait planning, speed control, environment perception, etc. Compared with the traditional terrain recognition process that uses color, texture, and other underlying features to describe terrain images, the present study starts from the perspective of transfer learning. MobileNet and DenseNet are employed for high-level feature extraction, and the voting integrated learning algorithm is used to classify high-level feature data sets. In the meanwhile, we have established an outdoor terrain data set that conforms to the traveling process of outdoor mobile robots, and processed the collected video data with key frames and sliding windows. The accuracy of the classification results reached 97%, basically satisfying the needs of actual terrain recognition.
