Image Anomaly Detection And Quality Assessment In Viedo Monitoring System

With the increased awareness of security, video monitoring is applied in the vastfield of manufacture and our daily life, such as traffic cross, market, bank, etc. As thelarge number of the cameras, it is hard to maintain all these device. Therefore, thetechnology which combines the intelligence image processing and pattern recognition isvaluable in the aspect of application and scholar investigation, so that the computers cancover the work of device detection for the human.In this dissertation, the conventional faults of the video monitoring system areinvestigated and discussed from the aspects of signal faults, camera noise detection andthe assessment of image quality. In this context, the common faults which affect thenormal operation of the monitoring system have been discussed. This article focuses onresearch of anomaly detection and image quality assessment with digital imageprocessing technology for video monitoring system.The faults in the video monitoring system are usually presented as video signallosing, image frozen and color disorderliness. The video signal losing can be detectedby the variance of the gray value in the image, since the image is usually a black screenwhen the video signal is lost. The image frozen can be detected by the differential of thetwo image frames in the time sequence and calculate the standard variance of the grayvalue of the differential image. The result is obtained by the comparison between thethreshold and the standard variance value. The hardest problem is the detection of thecolor disorderliness in the image. The solution is that we can calculate the ratio betweenthe average value of the color and the variance value of the image in Lab color space.The effectiveness of this method is demonstrated by some experiments.Camera is the "eye" of monitoring system, and it is easy to be affectby someunexpected factors. According to scale invariant feature transform (SIFT), the variationof the feature between the normal image and the distorted image is very large. Therefore,based on this principle, this dissertation proposed a SIFT based algorithm which is usedto detection the interruptions to the camera. The effectiveness of this algorithm dealingwith interruptions is demonstrated by experiments, the normal interruptions are omittedin this algorithm, so that the robustness of this algorithm is also very good.The definition fault, brightness fault and the snowflakes fault in the video imagecan be detected by the estimation of the quality of the video image. The quality estimation of the video image algorithms are investigated in this dissertation. And asupport vector machines (SVM) based algorithm which is applied to estimate thequality of the video image is proposed in the last part. For the drawback of the SVMalgorithm in the computation performance and the large marked samples,semi-supervised learning method and incremental learning method are introduced in thisalgorithm to improve the performance of the SVM algorithm. The effectiveness of thisalgorithm is carried out by some experiments to ensure this algorithm satisfy thereal-time processing and high accuracy detection requirements.In the last part of this dissertation, the conclusion of this investigation is given.And the future work in this investigation is also discussed.