An Efficient Malicious Activity Detection Model for Video Surveillance with Deep CNN model Using Positional Inclination Features

The problem of activity classification towards video surveillance has been well studied. Various approaches are recommended by researchers in literature but struggle to meet the accuracy constraints. To enhance efficiency, this article introduces the PIFMAD-CNN model, which leverages a Positional Inclination Feature for detecting malicious activity using CNN. The model utilizes the Mean Color Approximation technique to standardize the image and enhance its visual quality. With the normalized image, Regional Intensity Segmentation algorithm is applied which groups the various objects and features. Further, human feature are extracted and applies template matching to identify the human feature. From identified human feature, the method extracts positional inclination features. The deep CNN model is trained using derived features, incorporating multiple convolutional and pooling layers. The output layer neurons compute Positional Activity Support (PAS) and Inclination Activity Support (IAS) measure against different activity classes. Using the PAS and IAS values, the approach determines the Malicious Activity Score (MAS) to categorize the activity. The PIFMAD-CNN model hikes the accuracy in video surveillance.