Affirmation of human faces out of still pictures or picture progressions is an as of now making research field. There are an extensive variety of engagements for structures adjusting to the issue of face limitation and affirmation e.g. exhibit based video coding, face conspicuous confirmation for security structures, look area, and human-PC connection. The acknowledgment and region of the face, and furthermore the extraction of facial features from the photos, are fundamental. In view of assortments in illumination, establishment, visual point and outward appearances, the issue becomes complicated. This paper presents a novel method to extract human facial features for the detection of human emotions (such as "sad", "happy", "sorrow" etc.) under noisy conditions. This whole work constitutes better working of a video surveillance system. For detection and extraction of facial features simple formulae are used to represent skin color models depending on the range of HSV (Hue, Saturation, Value) values used for the detection of human skin. Here HSV color model is used because it is fast as well as compatible with human color perception. Additionally, implementation of Probability Neural Network (PNN) enhances the working of the surveillance system. Utilization of PNN expands the ability of surveillance framework as it can give the yield image regardless of whether the information image contains noise in it. The proposed algorithm for the entire task is developed using MATLAB software along with suitable Image Processing Toolbox (IPT).