Politehnica University of Bucharest, Romania
Our team is consisted of three senior students. We wanted to use our background in Electrical Engineering and Computer Science to create a device that automatically monitors and detects depression. Niculescu Vlad is student at faculty of Electronics and Telecommunications. As former participant in high school Olympiads in mathematics and physics, he redirected his interests towards research projects involving a lot of electronics and programming. During the bachelor’s years, he has worked at numerous electronic projects and competed and won awards in plenty of national and international student competitions. His main fields are digital signal processing, machine learning and embedded programming. Moreover, he has been working at Microchip Technology for two years as an Embedded Software Engineer. From September, he will continue his studies at ETH Zurich with a master in Robotics, Systems and Control. More about him here: www.vlad-niculescu.com . Isac Andrei is studying computer science at the Politehnica University of Bucharest. He is passionet about machine learning, cloud computing and web applications. He has previously developed a distributed image aggregator like Pinterest hosted in the cloud, a GitHub project analysis platform for assisting professors in grading students homework and last, but not least, a smart robotic arm. All available at link GitHub. My future plans are to continue my masters in computer science at ETH Zurich. Reference: https://github.com/isacandrei Alexandra, also a senior student at Politehnica University of Bucharest, is passionate about product design. Combined with her studies in computer science, she has developed many user interfaces and software product strategies in the past.
According to statistics, the depressive disorder affects around 8 percent of children and 6 percent of adults. Detpression is a device that can not only detect the depression, but also monitor it. It allows the doctor to monitor how the disease evolves for his patients. The patient simply put the device in his room corner, and it will start monitoring and send the results on cloud, where the doctor will access them. Detpression runs two powerful algorithms that allows it to accurately give a result (for voice and face). For both algorithms, the device does not send voice or images to cloud. It actually sends specific features extracted from those. Our research challenge consisted in finding those features. We collaborated with professors from Imperial College and University of Southern California during our research. Professor Sharon Mozgai from USC gave us a database with 192 interviews with depressive people, which was crucial in developing and testing our learning algorithm. They also suggested us what features they found to be the most relevant for detecting depression in their research work. Furthermore, we spoke with psychologists who gave us a lot of precious advice regarding the behavior of depressive people or key aspects that have to be tracked when it comes to depressive people. They offered to help us with the testing step. Even if our device provides very good results on the data set, is also has to be tested on real patients and this is the next step. The monitoring device will have a low price and will be used with ease by both the doctor and the patient. Our product is designed to be user friendly and very efficient. It will be a small box that encompasses raspberry pi, an external battery, camera and microphone. The physical part will have a button that starts recording data that will be stored in a database and passed through a neural network to get the results. The results will be displayed in an eye-catching web interface using a lot of visual representations to have a better understanding of the result. The web server in charge of the interface with the used is hosted on Microsoft Azure. The final predictive analysis is implemented using Microsoft Azure Machine Learning service. Technologies used: Cognitive Services API’s are used as follows: Speaker API - to identify if the current speaker is known and proceed accordingly Emotion API - to detect emotion on the recognized face Face API - to identify the faces in the current frame Text Analytics API - to detect the level of sadness in the transcripts of the subject's voice. Presentation video: https://www.youtube.com/watch?v=vqQgS-N31Mw&feature=youtu.be