Dynamic Intelligent Identification Of Video Streaming Algorithms Evolutionary Algorithm

With the advancement of technology and the urgent needs of intelligent identification, alarge number of investigators have devoted into the face recognition research areas. Facerecognition technology has broad application areas. It can be not only used in video surveillance,access control systems, but also can promote the multidisciplinary interactions. The method canbe easily accepted by the public due to its nature of concealment, non-contact and intuitive.This paper discussed the algorithms that can be used for dynamic video stream from fouraspects including pre-processing of the image, face detection, feature extraction, and facerecognition. The image geometry normalization, light compensation, and color histogramequalization preprocessing method were introduced in the image preprocessing part. We used thetraditional principal component analysis method to extract the image data for subsequentidentification in the human feature extraction part. Based on results from previous studies, newalgorithm was promoted in this study. The main innovation points of this paper are as follows:(1) A new face detection method taking advantage of both the color of the face detection methodand AdaBoost method were raised. Skin color detection method has a good advantage inoperating speed, but may detect human skin outside the face and provide unsatisfactory results.The misjudgment rate sometimes is very high for AdaBoost method. In this paper, the color of theface detection method and AdaBoost method were combined in face detection. The color of theface detection method was used first to roughly locate the skin color area, and then the AdaBoostmethod would be used to accurately locate the face within the skin regions. This study verifiedthat this combined method had a better detection results than the original methods.(2) An integrated kernel function was raised. The paper discussed and analyzed the performanceof both the radial basis function (RBF) kernel and polynomial kernel. In order to overcome thelimitations of single kernel function to get a kernel function with better performance, the methodof RBF kernel and polynomial kernel were combined together. This study verified that theintegrated kernel function was more efficient than single kernel function.(3) Parameter settings are the main factors affecting the performance of support vector machine(SVM). SVM of the RBF kernel has two parameters need to be set, while the polynomial kernelSVM has three. Five parameters needs to be set for the SVM of the integrated kernel functionused in this study The traditional parameter selection methods relied on experience or a widerange of grid optimization, and it is very difficult to find the optimal classifier to set theparameters. This paper solved this problem by presenting a real-coded niche evolutionaryalgorithm, and the5parameters of the SVM of the integrated kernel function could be set withhigh accuracy.