Facial Emotion Recognition Application In Emergency Prevention

The impact of emotion on human actions is a discussion subject in psychology, where some claim that emotion has a merely-there impact on human action and reaction. This misconception may have contributed to the disregard of emotional impact in the area of computer research where existing emotion recognitions system are more or rather exclusively focused on how well to detect emotion using sophisticated algorithms and hardware. The few today's systems that recognize the existence of emotion impact work as human companion but lack pro-activeness towards preventing impairments that come as results from emotion. Statistics show emotion as cause to almost as many as emergencies caused by driving-under-influence. That shows that the emotion-disregard era in computer-based prevention system should come to an end. It is time to not only recognize emotion but also exploit advantage it presents. It is in this regard we resolute to use facial human emotion for emergency prevention purpose by predicting emergency using emotion state. Our case study focuses on driving emergency prevention. This work is divided into two big tasks:facial emotion recognition and emergency prevention.The Facial emotion recognition goal is achieved through three sub-tasks:Face detection using Principal component analysis, Action Units computations and facial emotion recognition using Hidden Markov Model. The face detection sub-tacks consist of classifying a frame as a face or not. It is implemented using many Haar features combined into a strong classifier 90-95% reliable. The second sub-task consists of automating facial Action Units recognition by calculating main distances'ratio between the neutral and current state. The Third sub-task recognizes emotion based on Maximum Likelihood probability. The model is a set of three elements:initial probability function, inter-states transitional matrix and output observation matrix. It has also 7 states namely Neutral Happy, Angry, Sad, Surprise, Disgust and Fear. Emergency and Non Emergency represent the two possible observations. The parameters of the model are tuned using Baum-Welch re-estimation method constructing Maximum Likelihood classifier. The calculation of emotion probability is achieved using Forward-backward procedure. We used a homemade face database of 420 images for system training and achieved 89.2% of desired results using real-time video in emotion recognition.Emergency prevention can be achieved by predicting the future probability of emergency occurrence which can be done by extending the model of emotion recognition and consider Emergency and Non emergency as two possible output observations and the future occurrence of emergency can be calculated using backward procedure. The algorithm can be used to find which emotion state is more likely to contribute to emergency situation and to estimate which level (threshold) emotion intensity is safe or not. This is our direction for the future work. However, like in prevention systems the impact of the proposed algorithm depends not only on the system but also the user'cooperation which makes their impact hard to measure.