Research On Object Detection Method Of UAV Landing Process Based On Ground-based Vision

Today,UAVs are no longer a mystery.More than a hundred years have passed since1921,when the UK first tested a UAV.UAVs are becoming more and more functional and powerful.They play an increasingly important role on today’s battlefields and in postdisaster reconstruction.The safe landing of UAVs is a pressing issue,with most accidents occurring at this stage of landing.A prerequisite for safe guidance of UAV landings is the need to know the location of the UAV in space,which requires object detection of the aerial UAV first.Object detection of aerial UAVs using ground-based vision has become a popular research.The research of deep learning theory also makes it possible to provide efficient object detection for ground-based vision.(1)Firstly,the background significance of ground-based visual guidance for UAV landing process object detection is described.The transformation between the four coordinate systems is analyzed by understanding the imaging model of the camera.The camera is then calibrated using Zhang Zhengyou’s calibration method to calculate its internal parameters.Then the binocular visual position measurement was analyzed.Finally,aerial UAV objective measure distance is achieved based on triangulation.In response to the actual error,the objective measure distance of the aerial UAV under non-ideal conditions was then analyzed in detail.(2)In order to get the position of the aerial UAV,object detection is needed first.In this paper,we use Faster R-CNN algorithm for single subject detection of UAVs.Self-built dataset and elaborated the structure of Faster R-CNN.The m AP of object detection was improved by improving the number and area of anchor frames for this algorithm.(3)When there are multiple UAVs in the air and one or more of them need to be specifically guided,then multiple UAVs in the air should be specifically detected to determine if they are the objects to be guided.In this paper,we use Mask R-CNN algorithm for multiple object detection of aerial UAVs.The two improvements,anchor frame improvement and embedded CBAM module,are made to the algorithm so that the m AP is improved.