FaceReader, developed by Noldus, is described as "the premier professional software for automatic analysis of facial expressions". The software works by automatically analyzing visual input for the 6 basic emotions I mentioned in my last posts (Anger, Disgust, Fear, Happiness, Sadness, & Surprise) as well as neutral and contempt. In addition it also assesses gaze direction, head orientation, and person characteristics. It also offers analysis of Action Units. The software was trained using more than 10,000 manually annotated images and offers accurate modeling of the face by describing 500 key points. It is presently in its 6th iteration and claims to have been used worldwide at more than 300 universities. See it in action in the above video.
The Computer Expression Recognition Toolbox , or CERT, is described as "an end-to-end for fully automated facial expression recognition that operates in real-time". It was developed at the University of California, San Diego, stemming from a collaboration between researchers Paul Ekman (mentioned in my previous post) and Terrence Sejnowski. The software works by automatically detecting frontal faces in the video stream and codes each frame to 40 continuous dimensions, including the 6 basic expressions, contempt, head pose, and 30 facial Action Units. In the above video, they show how this system is being used as an intervention for children with autism. Possible applications include automatic discrimination of fake or posed expressions, detection of driver drowsiness, and adaptive tutoring systems.
Emotient is described an on-demand web service where users can upload video from focus groups, online panels and other sources and quickly receive reports and data about the emotional state of individuals in the video, as well as information about their attention and engagement. It seems to be largely marketed for commercial use to provide this information to companies when customers are experiencing their marketing, products, websites, or retail location. While this does not work with live video, it may be an interesting alternative to when live video is not possible or sending recorded video is easier or preferable.
SHORE is an object and face recognition software developed for the Google Glass. Adapted to recognize facial expressions of emotion, the technology has been trained by accessing a database of more than 10,000 annotated faces, similar to FaceReader. The developers, researchers from the Fraunhofer Institute in Germany, claim that in combination with the structure-based features and learning algorithms, they can train "models" that boast extremely high recognition rates. They cite a 92% facial recognition detection rate, as well as highly accurate gender detection and age estimation. However the technology is still only currently limited ro the detection of four facial expressions: Angry, Happy, Sad, and Surprised.
And now, something for social media! Pocket Avatars, an app developed through Intel Labs,captures emotions in real time using facial-tracking technology to then send to your friends in the form of an animated avatar. Your friend then receives a video message where the avatar relays the emotion by mimicking your posed expression. These messages can run up to 15 seconds in length and are sent the same way as a Facebook or WhatsApp message. The app uses your smartphone's front facing camera, and allows you to also send a voice message that the avatar's movements sync to accordingly.
Based on this post it is evident there are quite a variety of facial expression recognition software available, ranging in price, functionality, and use. Some of these are free or offer free trials, so go ahead and try them yourself and let us know what you think!
No comments:
Post a Comment