A Study Of Aerial Target Detection And Recognition Based On Image Sequences

Along with the growing concern for public safety problems, video monitoring systems, which still need artificially surveillance, are quietly going into public places such as the airport, station, bank and road transportation. In order to save the human and material resources inputted, emerges as the requirement of the intelligent video surveillance technology which is based on computer vision and artificial intelligence. And its purpose is to try to automatically analyze the image sequences captured by the monitor and to realize the target location, recognition and behavior understanding, thus completing the automatic feedback and early warning about the abnormal situations. Such a technology has great practical value in the market, commercial and military fields.This paper presents researches on the key intelligent video surveillance problems related to the moving objects detection, feature extraction and target recognition. Firstly, common target detection methods such as background subtraction, frame differential method and optical flow process are introduced, and their applicability and drawbacks are summarized in detail. Considering that the C-V model requires an initial contour in its segmentation process, a clustering method based on the frame differential result is applied to gain the suitable contour, which can be used for object detection very well combined with the C-V model. Secondly, common characteristics description methods such as color feature, texture feature and shape feature are introduced briefly. According to the fact that global features can reflect the intrinsic shape properties of the object and the overall distribution of the gray image, Zernike moment is detailed analyzed and its robustness to the translation, rotation and scaling is verified using the experimental data. Then features of three types of aerial targets are described by Zernike moment, and the SVM classifier is taken into account to separate these features. In the experiment, no public flying objects database can be taken in use, so 3DSmax software is used to establish three types of flight objects database. Video resources on the network are used as test samples to verify the classification process. The final classification results can explain the validity of our aerial target identification scheme.