Based On The Research And Implementation Of The Statistics On The Number Of People Through Image Processing And Neural Network

How to detect personnel and know the number of people from the image is alwaysan important and popular topic in the field of digital image processing and patternrecognition. The subject of statistics of the number of people has high values not only inresearch also in practical. Especially it has a great value in campus teaching sites suchas classrooms, auditoriums and so on.For the physical environment varies widely and interface from a variety of factorsto the number of image statistics, to improve the accuracy of the statistics, the expertsand scholars gradually raised a lot of relevant statistical methods. In these methods, thatintegrated image processing technology and machine learning has gradually become themainstream method. Therefore, this paper is trying to use digital image processing andthe machine learning in artificial intelligence and pattern recognition to do statistics onthe number of still images.First, this article makes a research in and analyzes this subject in the developmentstatus quo at home and abroad, also does analysis and comparison for some maturemethod such as face detection, head detection, to select a more appropriate techniqueand method for classrooms or other campus teaching places etc.Then, it makes the principle analysis and concrete realization of the practical digitalimage processing technology. Not only for the research and implementation of theclassic image processing method such as image denoising&smoothing, imageexpansion and corrosion, color image to grayscale, gray image and color imagehistogram equalization, also it puts forward that a targeted light compensation algorithmand image enhancement algorithms according to the actual occasion’s imagecharacteristics studied in this paper. Through these image pre-processing means, it caneffectively reduce the adverse impact from a number of confounding factors on thesubsequent processing of the image, ready for further detection and identification ofhuman characteristics.Then it analyzes and implements the most commonly used cascade classifier in facedetection based on Haar-like features, where it first introduces Haar-like features, integral image, and illustrates with practical examples the acceleration principle for thevalue of the integral image features calculation. Next, it introduces the weak learningalgorithm and the strong learning algorithm, Boostring algorithm and AdaBoostalgorithm. And ultimately composes the functional cascade classifier with the strongclassifier. In order to further improve the accuracy of face detection, skin colorinformation to verify the face region is used. For head detection, it attempts to use theBP neural network to detect. With BP neural network’s learning on multiple samples,heads in image can be eventually well detected.At last, because the unsatisfactory result of the face and head detection in theunclear images, it uses a reference check to determine the possible area for persons inaccording to classroom image features. After that, it does analysis and testing of theseareas to improve the detection rate of people counting, which makes the better practicalvalue to the system.