Realization And Exploration Of Binocular Vision Odometry System For Small UAV

With the rapid development of modern science and technology,mechanical automation has become the requirement of the era.As an automatic tool,UAV become more and more popular and develop rapidly due to the properties of light weight,low cost,good concealment,high flexibility,strong adaptability to a variety of harsh environments.An advanced navigation system plays an important role in the whole UAV system,while the accurate positioning function is the key to achieve autonomous navigation system of UAV.The existing positioning scheme cannot meet the needs of UAV in indoor applications due to the limitation of many problems.Such as the signal shielding problem of GPS,serious drift problem of inertial navigation system,high cost and hard to deploy problem of laser ranging and so on.On the contrary,the visual positioning scheme is easy to deploy,cost less,is immune to wireless signal shielding,and hence has become hotspot in the field of UAV positioning and navigation research in indoor application.This paper firstly introduces the stereo odometry algorithm of UAV in detail.Based on the algorithm,a Pixhawk flight control board,an ODROID XU4 ARM board,two mvBlueFOX-MLC camera and a quad rotor frame are used to build the test platform.A SIFT algorithm is used as main method for feature detection and matching and to achieve stereo odometry function,and some flight task are further deployed for testing.From the experiments result of this paper,it can be found that the cumulative error significantly affects the positioning accuracy.Focusing on the problem,this paper considers an indoor cruise application of UAV and proposes a cumulative error suppression method based on historical visiting information(called as CES-HVI for short).Like the landmark method,this method also uses some known global information(historical location results)to eliminate the cumulative error,but it does not need to set up landmarks in advance,therefore it suitable for the unknown environment.Aiming at a typical application scenario of UAV-indoor cruise,this method first defines a neighbor visiting area in current location,then calculates the feature point matching rate of current image and historical visiting points within neighbor area.For these neighbor points with high matching rate,a trusted historical visiting point set is defined.Further,the current coordinate can be estimated by current image and each historical image of trusted point set.After traversing the trusted point set,the weighted sum of all estimated coordinates is outputted as corrected location result,where we adopt the matching rate as weight.Experiments show that this method can effectively improve the positioning accuracy.