The Design And Implementation Of Moving Small Target Recognition System Based On Template Matching

The small target recognition has become a hot spot in the field of machine vision.In the intelligent monitoring or more advanced human-computer interaction system,the detection and recognition of small space targets has become an indispensable part,which provides information for the specific advanced visual understanding and analysis.The small target in this paper refers to an object with a pixel size between 10* 10 and 40 * 40 pixels in the video image,which can be used by the algorithm with less information,no obvious texture or enough feature points.The existing traditional target detection algorithm can not achieve a good result to detect small,often incomplete contours and other issues,resulting in the follow-up to the small target tracking process,often the target loss and other issues.In view of the above problems,this paper mainly studies the detection and tracking algorithm of small target of visible light.On the basis of the classical algorithm of moving target recognition,this paper focuses on the camera small target detection and tracking technology under static background to improve the robustness,accuracy and real-time performance of the algorithm.The main contents of this paper include three aspects: image preprocessing,small moving target detection,small moving target tracking.In the image preprocessing stage,based on several commonly used algorithms,the effects of mean filtering,median filtering and Gaussian filtering on video image noise are studied,and the effect of noise elimination is verified by experiments.The effect of mathematical morphology filtering on binarized image is improved,and the reliability of small target detection is enhanced.In the stage of small target detection,this paper proposes a new method to apply small target extraction based on the characteristics of moving small target in video and small inter-frame displacement change.This method will extract the small target contour based on the background subtraction method and the edge detection method of the mixed Gaussian model,mark the target area in the original image,obtain the small target initial template,and prepare the template matching tracking for the next step.In the small moving target tracking phase,in order to avoid the phenomenon of target loss due to the "degradation" of the target template,it is proposed to combine the template update,Kalman motion prediction and template matching technologybased on the unit gray scale entropy to form a complete Matching tracking algorithm.The Kalman filter is used to predict the detection area of the next image,and then the exact location of the small target is determined by using the template matching based on the gray value entropy of the image unit to realize the fast and accurate tracking.To verify the effectiveness of the proposed detection and tracking algorithm,OpenCV open source computer visual library and Visual Studio 2012 were used to set up the experimental platform,and a set of small target recognition system based on template matching was developed.Experiments show that the proposed small target detection and tracking algorithm can detect and track small moving targets well,and it is robust to rotation,occlusion and size change.This has certain reference value for the design of moving target recognition system in video surveillance field.