An algorithm for surveillance cameras to automatically detect faces in the video stream has been developed at Tomsk Polytechnic University. The algorithm allows processing high resolution videos in real time and also evaluating the number of people, their gender and even approximate age.
The supposed algorithm allows working with an Ultra HD (4K) format, which is increasingly gaining popularity in the professional closed-circuit television (CCTV) sector. Earlier the Full HD resolution (1920×1080 pixels) was believed ideal, now 4K is 3820×2160 pixels. To date there are surveillance cameras supported this format. But a problem arises: processing and storing these high-quality videos require very powerful servers and data transmitting channels with high throughput as well. The polytechnicer's algorithm allows working with video 10 times faster than existing analogs.
Professor Vladimir Spitsyn from Department of Computer Engineering said: "This original method can work with a stream video in ultra-high resolution 4K Ultra HD format with acceptable quality and is much faster than existing analogues in the world. This will allow automatic monitoring of the environment and quickly identifying emergency situations without operator attention and will provide highly detailed videos."
The algorithm is referred to one of machine learning technologies, a neural network convolution — an image classification technique based on neurobiological model of cat visual perception.
"In order to detect a person on his face image in the crowd, like on stadium or in subway it is required to work with an image in scene with very high detail. 4K resolution is suitable for this purpose. Because of the large amounts of data it is difficult even to store these videos not only to process them, not to mention intellectual analysis," Said Ilya Kalinovskii, the developer, a PhD student of Department of Computer Engineering.
Now the young scientist has to develop a CCTV camera for 4K video stream in real time. The device will automatically detect faces and transmit only significant fragments of the scene.
The development of a camera with biometrical analytical system has been recently awarded a grant under the UMNIK program of the Fund for Assistance to Small Innovative Enterprises in the scientific and technical field. 400,000 rubles grant is given for two years.