Research On Air Target Recognition Algorithms Based On Video Stream

Target detection is a prerequisite for target extraction and target recognition.In this thesis,we analyze the advantages and disadvantages of traditional three target detection methods through simulation.When testing the background subtraction method to detect targets,we observe that the process of background modeling updating is complicated and very slow.Since the background updating is not timely,the error of detection results is relatively large.Experiments of the optical flow method show that the detection effect of this algorithm is good,but the time complexity of this algorithm is important and it is time-consuming,It can not meet the requirements of the subject for real-time.According to the experiment results by testing the frame difference method,we find that the detection result is prone to duplication and it is greatly affected by the ground background.Although the duplication problem is solved improving the frame difference method,but the improved frame difference method will be affected by the ground background when they are used for target detection,and result in false detection.In order to solve the influence of ground background on target detection.This thesis present a detection method to remove the interference of ground background.Firstly,the video frame image is preprocessed by threshold segmentation,morphological processing and edge detection technology,then use the Hough transform line detection technology to divide the video frame image into sky area and ground area By combining the block frame method and the morphological processing method for the sky region,we successfully reduce remove the interference of the ground background and detect the target.Aiming at the segmentation and extraction of aerial targets,we propose a marker controlled segmentation method to effectively extract target.To choose the eigenvector which can effectively represent the target,two kinds of features are deeply studied: Hu invariant moment and HOG feature descriptor.In order to classify the features,we propose a multi-class support vector machine model and inputs the HOG features and Hu features into the SVMclassifier for classification.The experiments show that the HOG features is better for the identification of the air target.