Research On Image Transmission And Object Detection Acceleration For Tethered UAV

With the rapid development of the industry of unmanned aerial vehicles(UAVs),tethered UAVs have been widely used in many fields including the patrol inspection,security,traffic,aerial surveying and mapping,etc.To improve the airborne image processing capability of UAVs on the airborne platform with the restriction of power consumption and load capacity has been the research hot issue in the current industry of UAVs.In this study,the real-time image processing was applied as the main research direction,and the hardware circuit of the high-speed image transmission and the acceleration method of the object detection on the airborne platform were mainly investigated.At first,the comprehensive analysis was performed and the design indicators were made according to the requirements of the real-time processing of the airborne image,and the content in this study was investigated quantitatively.FPGA in ZYNQ series was selected as the embedded processing core,and the lightweight models of MobileNetV2 and the algorithm of YOLOv3 were selected as the core research objects of the object detection acceleration.Secondly,two hardware architectures of the high-speed transmission board and the test board of the airborne image were proposed in this study to develop the high-speed transmission circuit of the airborne image.According to signal integrity theory and the high-speed signal wiring skills,the PCB design and manufacture of Gbit/s orders of magnitude were completed.Meanwhile,the research on the object detection acceleration proposed in this study was mainly performed from three aspects: 1)the network pruning,normalized layer processing and the introduction of attention mechanism were applied to improve the object detection algorithm,balancing the lightweight,the detection speed and the detection accuracy.2)the hardware accelerator was built on the basis of the development platform of Vivado and high-level comprehensive technology,achieving the hardware acceleration of the algorithm.3)the pipeline,memory access and the optimization strategy of adder tree were proposed respectively aiming at the problems of cache shortage and resource occupation of the accelerator.Finally,this study tested the function,power consumption characteristics of the airborne image high-speed transmission board,and the acceleration performance of the improved object detection algorithm through building the experimental platform.The test results indicated that:(1)The data interaction rate and the inter-machine-ground transmission rate of the high-speed transmission board of the airborne image were both up to the Gigabit order of magnitude.The average power consumption,size specification and the weight of the board were only 4.2 W,60mm×70mm×15mm and 54 gram respectively,ensuring the high speed transmission performance and reducing the deployment pressure of the airborne platform effectively.(2)The enhanced YOLOv3 algorithm was accelerated via the hardware and tested via the test data set which was constructed on the basis of VisDrone.It could applied to process about 11 pictures per second on average with a power consumption of 3.3W,and the accelerated energy efficiency ratio was 7.6 times higher than that of GPU.Compared with the related acceleration studies,it was found that the object detection acceleration method proposed in this study had the obvious advantages of the efficiency and the embedded mobile terminal.The research study in this work would provide the technical support for the improvement of the aerial operation efficiency of tethered UAVs,and would maximize the effectiveness of UAVs within limited airborne resources.Meanwhile,this work could provide the reference for the image processing of other unmanned platforms and promote the development of the unmanned technology in the future.