Design Of An Image Data Acquisition Module For The Visual Obstacle Avoidance System Of Agricultural UAVs

In order to improve the safety of production operations,real-time obstacle avoidance is one of the development trends of intelligent agricultural UAV.The obstacle avoidance system based on visual sensors is more intuitive and clear in identifying the type and depth information of obstacles,recording flight data and playback.The existing obstacle avoidance system of agricultural UAV has some problems,such as high modularization,API privatization deployment and other security measures,not supporting secondary development,poor platform versatility and so on.There is no image data collection module for the obstacle avoidance system of agricultural UAV sold separately in the market.The existing modules can not meet the requirements of lossless compression,complete preservation,fast reading and so on.The research and development of vision obstacle avoidance system of agricultural UAV is inseparable from the storage,processing and analysis of massive image data,and the validity and reliability of image data are crucial to the accuracy of obstacle recognition and analysis.Based on the embedded development board NVIDIA Jetson Nano,a real-time image data collection module is designed for the vision obstacle avoidance system of agricultural UAV.The module is divided into data collection sub module and data storage sub module.The data collection sub module includes Intel Real Sense D430 3D depth camera module and USB3.0 data line.The data storage sub module includes the embedded development board NVIDIA Jetson Nano and SD memory card.After the completion of hardware selection and construction,the package shell of the data collection module is designed through Solid Works 2014.The design strives to realize the integration and lightweight of the module,and facilitate the assembly and use with other modules in the obstacle avoidance system,such as the self built image transmission module,multi visual image acquisition module,DJI NAZA-M Lite flight control module,obstacle avoidance system power supply module,etc.Finally,based on the volume and weight of the module,a 4-rotor UAV platform(the flight control system is DJI NAZA-M Lite)with a endurance of about 15 minutes is built to collect and store real-time image data.During the test,the image of Intel Real Sense D430 binocular camera was released in the form of "topic" in ROS(Robot Operating System).Through C++ programming,after starting the binocular camera,the current image of the lens is transmitted at the speed of 10 FPS.When the data collection sub module fetches the image of the binocular camera,the current image of the lens is displayed by continuously subscribing to the "topic" of the lens.The data storage sub module stores the remap image and depth image collected by the data collection submodule in real time.According to the actual requirements of the application scenario,this study compares the frame loss rate of PNG format and BMP format,and selects the storage mode with lower frame loss rate.The frame loss rate in this study refers to the ratio of the number of lost image packets to the sent image data group in the working test of the image data collection module.Its size reflects the integrity of the system data storage,and has a negative correlation.Complete data preservation is more conducive to timely and effective obstacle avoidance feedback,providing real-time guarantee for safe operation of agricultural UAV.The collection and storage functions of this module are realized by functional testing.Finally,the designed image data collection module is mounted on the quadrotor UAV according to the functional requirements,and the on-board test and indoor test are carried out.The two storage methods of PNG and BMP are calculated and analyzed.The test data shows that the frame loss rate of the two formats is basically unchanged and the frame loss rate of image data collection module is about 1% in PNG format storage mode,and about 10% in BMP format storage mode,so the PNG format storage mode with good real-time performance can meet the requirements of image data real-time storage module Block design requirements,and can ensure the integrity of picture information.Then the effect of uneven illumination is eliminated by homomorphic filtering,so as to improve the image quality.This research can provide accurate data source for the image processing and analysis of the follow-up obstacle avoidance system,and provide technical support for the realization of autonomous obstacle avoidance technology of agricultural UAV.It has the characteristics of low module cost,high integration,strong practicability and wide platform versatility,and supports the advantages of two times development of software.