Application Of Model-Free Adaptive Control Method In Image Recognition

With the development of image recognition technology,the demands on accurately detecting dynamic foreground targets in the video have increased day by day.Detection of foreground targets is of important meaning for the further researches of target tracking and action recognition,etc.Through extracting the background images in the video,the dynamic foreground targets in the video can be obtained with application of background subtraction.Thus,the accurate extraction of background images of video is of significant functions for the foreground target detection.While it is difficult to extract real background images,as they might be disturbed by multiple aspects including changes of natural environment and detection errors of shooting equipment,etc.While problems like model distortion might happen under the influence of the above interference factors when using the traditional background extraction methods based on model,so the detected foreground targets are difficult to be recognized because the background cannot be accurately extracted.It needs not establish mechanism model when using the method of Model-Free Adaptive Control(MFAC),the control algorithm can be acquired simply through the changeable data of the system,which can well avoid the problems caused by dependence on models,and there are three control schemes with different structures available based on the degree of complexity of process objects.Thus,the applicable research on background image extraction and foreground target detection in the image recognition field based on MFAC were stressed on in this thesis.The main contents of this thesis are as follows:Firstly,theoretical analysis was made on the feasibility to adopt the MFAC method to background image extraction,and extraction of background images with MFAC based on three different formats,namely compact format,partial format and full format of single input and single output(SISO)was realized.It was shown through the comparison of background extraction effect of MFAC and traditional methods based on model that,the performances of MFAC method were better,the extracted background was more stable,and it was of stronger disturbance resistance.Meanwhile,the foreground targets detected through MFAC method were more accurate,and based on different degrees of complexity of environment,there were three patterns available.Secondly,to make full use of color information of the images and avoid the gap of some key color information caused by graying treatment for the color images,the multiple input and multiple output(MIMO)MFAC color background image extraction method was also established in this thesis.Color images could be directly disposed through the RGB tri-channel to effectively obtain the color background images.Meanwhile,the foreground targets detected based on color background looked clearer than those detected by gray background.The extracted color background could also provide basis for the shadow removal.Finally,improvement researches were made on dynamic background and shadow removal.In a dynamic background,the separation threshold value of foreground and background could be automatically set based on the dynamic degree of the background,which realized clearly detecting the foreground targets under the dynamic background.Meanwhile,aimed at the problem that there were shadows on the foreground targets caused by illumination,improvement was made on shadow removal based on color background images,the judgment method for shadow detection was acquired with the difference color characters of shadows and background images,which could well reach the effect shadow removal.