Accurate position information is a key element to ensure the efficient operation of inspection UAVs,but due to the widespread distribution of transmission and distribution lines,the traditional GNSS-based UAV positioning method is very likely to be obstructed and difficult to provide stable position information.In this paper,the monocular camera and IMU carried by the machine patrol UAV platform are utilized.Based on the traditional visual odometry model utilizing convolutional neural network,combined with the long and short-term memory neural network and IMU information,a deep learning model based on the segmentation of visual inertial instances is proposed,which effectively improves the robustness of the system and the accuracy of the motion solution.Through experimental evaluation of the proposed self-localization model,the training effectiveness of the model is demonstrated.Field experiments are designed to address the application environment of UAVs.The final average localization error under the VIPS-Mono model is 0.058 m,which is better than that under the CNN-LSTM-VO model of 0.234 m.The results show that the model proposed in this paper can provide effective support for the self-localization of the UAVs for power transmission line inspections.
维普
万方数据