Semantic Based Abnormal Motion Detection Using Deep Learning in Decentralized Fog Environment

Abstract

Intelligent abnormal motion detection systems have become widely popular newlineresearch topic in recent research trends, as a result of the growing worry over high-profile newlineacts of violence. The popular abnormal motion detection model detects for unusual human newlinebehaviour. Unfortunately, the lack of a universally accepted definition for abnormal newlineactivities makes it difficult to reliably identify unusual human behaviour from the video data newlineand is highly dependent on contextual information. The crucial phase in abnormal motion newlinedetection is video analytics. It is the process of inferring useful information or sensible newlinepattern out of video data. Analysing such video data for abnormality remains complex. The newlinesuggested research work starts with the inclusion of deep learning based optimal preprocessing newlinelayer. Pre-processing layer has the capability of refining the raw video data into newlinea suitable form which makes the further reasoning part easier. Key frame extraction and newlinecontour-based background subtraction are two crucial pre-processing steps that the newlinesuggested system uses to make the raw video data as quality one. The refined video sequence newlinecan be used for further reasoning process. newlineThe adaptive key frame extraction system chooses potential key frames from the newlinevideo sequence by using a sliding window technique. The visual geometry group-16 Transfer newlineLearning (VGG-16 TL) technique is used to comprehend the high-level semantic newlineinformation of video frames in order to better characterize the video material. The adaptive newlinecontour-based background subtraction method separates the target foreground pixels from newlinethe background scenes, making it easy to spot odd motions in the video frames. To call the newlineproposed abnormal motion detection system as intelligent, the proposed work suggests newlinesemantic approach newline