Detecting Dangerous Scenarios from Body Language and Emotion Using a 2D CNN

Kryzsko Commons Ballroom

Using computer technology and deep learning methods, it is possible to analyze data more quickly and accurately than ever before through training convolutional neural networks (CNNs) to detect certain characteristics from a range of media including videos and images. These machine learning technologies and methods can be applied in real-world scenarios to detect, and possibly even help deter, dangerous situations or events. This work proposes using a 2D CNN to classify images including people based on the emotion that is being presented in the image. The emotions that will be primarily focused on in this study will be pain, anger, fear, and suffering. The dataset used for the training and testing for this work will come from the EMOTIC database which contains approximately 24,000 images. Success in this endeavor may result in a CNN model that could be installed into security cameras, allowing them to monitor and alert whenever a possible dangerous situation is about to occur, or is already in the process of happening. The time that could be gained from detecting these situations early may save people from injuries or possibly even save lives. This approach will attempt to improve upon existing emotional detection research by focusing less on just faces and more on the person's body and environment as well.