Facial Gesture Recognition

Abstract

Facial gesture recognition systems can have a variety of uses. The expression recognition is newlineone of the most popular applications. A rather critical application associated with the driver s newlinesafety during driving a car is considered. The Driver s behaviour monitoring system is an newlineapplication of gesture recognition system which detects facial gestures such as the driver s newlineyawn, talk and certain expressions such as happiness and anger. The intention is to monitor newlinethe alertness of the driver during a driving session. There is a certain amount of commonality newlinein these gestures but they are also person-dependent. The objective of the work is to develop newlinenovel extraction and detection techniques for monitoring the driver s behaviour. newlineFor any application to work in real time, constant tracking is important. The image newlineframes of the driver have to be monitored. The Mean-shift object tracking method was newlineimplemented for the purpose of tracking the driver s head movement and face. The tracking is newlinefaster as the region to be tracked is small. This makes the tracking not only effective but at the newlinesame time less time consuming. newlineThe region of interests to be extracted are the face and the mouth region. The approach newlineadopted for extracting the region of interest associated with face in the research work employs newlinethe YCbCr and the RGB colour spaces from the tracked output frame. The two agent scenario newlinefor skin-tone detection is novel. This is the stage which segments only the face region from newlinethe rest of the image contents. The segmented skin tone region is further applied to a series newlineof image processing transformations for a noiseless output region. A number of colour space newlinetransforms were explored for this stage. The YCbCr colour space was chosen based on a newlinenumber of factors. The algorithm makes use of the Connected Component Analysis for the newlineface boundary detection. The algorithm is computationally less complex when compared to newlineother skin-tone detection algorithms. newlineiv newlineThe region of interest associated with mouth is extracted using th