Abstract
Automatic visual navigation flight of an unmanned aerial vehicle (UAV) plays an important role in the highway maintenance field. Automatic highway center marking detection is the most important part of the visual navigation flight of a UAV. In this study, the UAV-viewed highway data are collected from the UAV perspective. This paper proposes a model named the YOLO-Highway that uses an improved form of the You Only Look Once (YOLO) model to enhance the real-time detection of highway marking problems. The proposed model is mainly designed by optimizing the network structure and the loss function of the original YOLOv3 model. The proposed model is verified by the experiments using the highway center marking dataset, and the results show that the average precision (AP) of the trained model is 82.79%, and the frames per second (FPS) is 25.71 f/s. In comparison with the original YOLOv3 model, the detection accuracy of the proposed model is improved by 7.05%, and its speed is improved by 5.29 f/s. Moreover, the proposed model had stronger environmental adaptability and better detection precision and speed than the original model in complex highway scenarios. The experimental results show that the proposed YOLO-Highway model can accurately detect the highway center markings in real-time and has high robustness to changes in different environmental conditions. Therefore, the YOLO-Highway model can meet the real-time requirements of the highway center marking detection.
Highlights
A convolutional neural network (CNN) is a popular method in the field of object detection [5,6]
Is research aims to obtain a proper balance between the detection accuracy and the detection speed of an object—that is, reducing the processing time while ensuring the detection performance on the unmanned aerial vehicle (UAV) platform can meet the detection demand. us, a model named the YOLOHighway that is developed based on the latest research results of the YOLOv3 model using highway images obtained from the UAV perspective is proposed. e proposed model can save memory usage and have a faster processing speed than the YOLOv3 model while ensuring a certain accuracy, which means that it can be implemented in a UAV platform for accurate and efficient object detection
3.1. k-Means Clustering Method. e k-means clustering method [30] is used to determine the anchor box dimension in the UAV-viewed data. e k-means algorithm divides all samples into k clusters that are typically chosen to be sufficiently far apart from each other in space based on the Euclidean distance to produce valid data mining results. e dimension and number of anchor boxes are obtained by the final cluster centers calculated by the k-means algorithm
Summary
A convolutional neural network (CNN) is a popular method in the field of object detection [5,6]. E two-stage models mainly include the R-CNN [7], Fast R-CNN [8], and Faster R-CNN [9], which have good detection performance in terms of accuracy. Is research aims to obtain a proper balance between the detection accuracy and the detection speed of an object—that is, reducing the processing time while ensuring the detection performance on the UAV platform can meet the detection demand. Us, a model named the YOLOHighway that is developed based on the latest research results of the YOLOv3 model using highway images obtained from the UAV perspective is proposed. E proposed model can save memory usage and have a faster processing speed than the YOLOv3 model while ensuring a certain accuracy, which means that it can be implemented in a UAV platform for accurate and efficient object detection. (ii) An improved network structure of YOLOv3 is proposed to improve the detection efficiency and reduce the memory cost, which further enhances the performance in highway marking detection
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
