The Research On Object Recognition And Tracking In Anti-UAV System

The confusion created by unmanned aerial vehicle(UAV)has caused the rise of anti-UAV technology.As an important part of the anti-UAV defence system,the optoelectronic module detects and tracks suspicious targets through computer vision technology.But small and low contrast objects are hard to be extracted rich features in complex background,resulting in serious false alarm in dim target detection and tracking.In this paper,a lot of work has been done for target detection and tracking.The main research work is as follows:First of all,to decrease the probability of false detection of dim target,this paper present a novel method for target detection based on spatio-temporal continuity.Ten binary images are put into a FIFO(First In First Out)channel,and candidate targets are determined in first binary images.Windows are built around candidate targets and pixels in windows are accumulated in FIFO.The accumulation would exceed the threshold if there is a moving target in windows.Further,this paper uses the space-time continuity of the moving target to remove the noise in the window by carrying out the logic ‘and' operation between the adjacent windows.In order to make a distinction between UAV and other targets,such as birds,kites,we set up a data set and used it to train a neural network for target recognition.Then YOLO(You Only Look Once)is employed to identify the suspicious target which achieve a good result.Then,a test platform is built for UAV tracking algorithms,and more than 10 mainstream real-time tracking algorithms are analyzed qualitatively and quantitatively.STC(SpatioTemporal Context)algorithm is selected as the tracking algorithm in anti-UAV system.Aiming at the drawbacks of STC algorithm,such as wrong size estimation and model update mechanism,this paper put forward the improved STC algorithm.The proposed algorithm adds a scale filter for target scale estimation,and exploit the confidence map to find whether the target is occluded or lost.The model stop updated when the target is occluded or lost to avoid introducing error information.Compared with the original STC algorithm,the improved algorithm achieves a better performance in both tracking accuracy and success rate.Finally,an electro-optical modules of anti-UAV system is built,which can detect and identify suspicious moving objects and then track them automatically.The optoelectronic module uses two cameras and two processors.With two cameras,we gets both far and near field for tracking and recognition respectively,and the application of CPU&GPU dual processor improves the real-time performance of target recognition algorithm.