Obstacle detection in dangerous railway track areas by a convolutional neural network. (14th June 2021)

• Record Type: Journal Article
• Authors: He, Deqiang; Li, Kai; Chen, Yanjun; Miao, Jian; Li, Xianwang; Shan, Sheng; Ren, Ruochen
• Is Part Of: Measurement science & technology. Volume 32:Number 10(2021)
• Publication Date: 2021-06-14
• Subjects: deep learning -- rail transit -- obstacle detection -- Mask R-CNN -- trajectory extraction; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/abfdde
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: The obstacle detection in the dangerous area of railway track is an important research direction in the field of the driverless train. Traditional obstacle detection methods have many issues, such as complicated steps, low detection accuracy, and slow inspection speed. To overcome these defects, a detection network based on deep learning, named Mask R-CNN, was adopted, as described in this paper. The detection network uses the Mask-RCNN model with ResNet101 as its backbone feature extraction network, which has deeper network layers. Therefore, this network has high detection accuracy for small targets. Data from a subway obstacle test was used to train this network. In addition, data augmentation and transfer learning were adopted to improve the efficacy of the training. To improve the detection speed, the technical framework of the detection was also improved. The test results showed that the precision of the Mask-RCNN model with ResNet101 as its backbone feature extraction network reached 95.7% and that it required an average time of 0.18 s. The proposed model shows satisfactory performance when used for obstacle detection in the dangerous area of the railway track, compared with other networks.