Deep Matching Network Based Image Auxiliary Localization System For Unmanned Aerial Vehicle

Localization of the Unmanned Aerial Vehicles (UAVs) have vital importance to provide with navigation information and avoid getting lost. However, the current localization technology mainly depends on GPS or wireless network, which is prone to be interfered by the environment and wireless attack. To solve this issue, we propose a vision based deep matching network for UAV by a image auxiliary localization algorithm. The proposed algorithm integrates vision information and traditional wireless localization information by an information fusion filter to realize more robust localization. The deep matching network realizes localization prediction of image stream by matching the extract features with stored memory segments. The image stream is firstly inputted to a convolutional neural network to extract image features. Then a LSTM is employed to extract localization features from image features. Finally, the matching network will output the localization prediction from the image features and stored memory segments. After experiment, compared to existing method, the mean localization error performance of the proposed approach can mostly improve 49.8%.