The Robot Vision System Of Indoor Static Multi-target Recognition

Target recognition and tracking has a very important practical value in many fields,such as intelligent surveillance,medical research,human-computer interface,virtual reality,motion analysis,so more and more domestic and foreign researchers are in interested in it.In this paper,a large number of experiments were carried out on the recognition of different target.Based on these,we studied the image data pre-processing and image feature extraction.And we designed corresponding real-time target recognition systems.The main work is as follows.Firstly,we learned that image noise pollution is serious in the most target images through the analysis.On this basis,de-noising methods of target images are studied and analyzed.Experimental results of several common filtering methods are compared and analyzed.Secondly,taking the variability of illumination into account,we designed a new illumination compensation algorithm based on correction of Gamma.The adaptability of Gamma correction for illumination changes is improved by superimposing nonlinear functions.The automatic Gamma correction in uneven illumination conditions is achieved by analyzing the parameters of nonlinear functions.Thirdly,we need to identify the different types of targets,so we use the method of image recognition based on scale-invariant feature.The characteristics of the target are stored in database.Common methods of scale-invariant feature extraction are introduced and a large number of experiments are done with these methods.We select the best algorithm for each target by measuring the time matching and accuracy matching between the various extraction algorithms.Fourthly,the target of real-time recognition is achieved by combining the algorithm of feature extraction and Mean Shift.Firstly the target is located by using the feature matching algorithm.Secondly initialize the target template of Mean Shift according to the target position in the image,and update the target real-time.Finally,transplant from the indoor static multi-target recognition system of PC to embedded Linux systems.