A Research Of Unmanned Aerial Vehicle Vision Algorithm

The research and application of Unmanned Aerial Vehicles(UAV)have been greatly improved in recent years,the UAV can be used from the military fields such as early warning and reconnaissance to the civil fields such as weather monitoring,disaster detection,and search and rescue.Due to its compact structure and variable operation,miniaturized drones can adapt themselves to more complex environments,but they also impose certain restrictions on their onboard load bearing capacity,energy supply capacity,and endurance capacity.This means that sensors such as heavy weight and high power consumption are no longer suitable for drones in some environments.So the camera as the main sensor of the drone has begun to become popular,and with the combination of machine vision technology,drone research and application began to diversify.There are many researches and applications of UAV-based visual algorithms,such as positioning and tracking,obstacle avoidance and navigation,3D reconstruction,augmented reality,virtual reality,and so on in different environments.However,due to the complexity of the environment and tasks,the research of UAV vision algorithms is more inclined to achieve the feasibility of a certain task under certain circumstances.Therefore,this paper proposes and implements a binocular vision-based UAV indoor collision-free flight strategy,which is mainly divided into three parts:First,the binocular stereo vision system is used as a sensor.Compared with a monocular camera,two different images can be generated in the same scene,which can provide more visual information for the subsequent image processing and can also perform effective depth estimation.Second,An artificial neural network for image classification in indoor environment is constructed.The indoor environment of the UAV is divided into corridors,stairs,and corners.Compared with the k nearest neighbor classification method and support vector machine classification method,it has higher Classification accuracy,and greatly reduced the time required for each scene classification.Third,Different drone flight strategies were designed for different indoor environments.In the corridor flight environment,corner environment,stairway environment and mixed environment continuous flight experiments,drones all havehigh success rate of safe flight.Experimental results show that our proposed binocular vision-based UAV indoor collision-free flight strategy is feasible.