Overview / Usage

Nowadays that security comes into more prominence every day, it is necessary for people to keep more passwords in their mind and carry more cards with themselves. Such implementations, however, are becoming less secure and practical, thus leading to an increasing interest in techniques related to biometrics systems. Biometrics systems are the systems which store physical properties of people in the electronic environment and enable them to be recognized by the stored electronic information when needed. Biometrics is the identification of human. It works on the principle of identification of physical properties of a person which he or she cannot alter, are distinctive from others, can be used for identification, and are in his or her possession only. Extensive studies are conducted on biometrics techniques such as fingerprint, hand, face, iris, retina and voice recognition. Some systems have been developed, tested, and results have been obtained. Face recognition systems are among the most important subjects in biometrics systems. These systems, which are very important for security, in particular, have been widely used and developed in many countries.

Methodology / Approach

Face recognition systems’ senses of work are exactly the same as the biometric recognition systems. Face recognition system is based on the idea that each human being is different and unique in creation. If we elucidate this, the facial structure has parts that are unique to each person like fingerprints. As a result of the biometric matching carried out in a computer environment, facial structure is compared to the faces on the database and the personnel who has the matching facial structure is allowed to log in.
When choosing the real-time image detection algorithm, most algorithms perform convolution throughout the input image, thereby taking in the input images multiple times, which increases computational cost. The only deep learning algorithms that take the input image once are the Single Shot Object Detection (SSD) and You Only Look Once (YOLO) algorithms. But for this project, I will be using the SSD algorithm because based on the paper, it performs better than YOLO.
The SSD algorithm uses the pretrained VGG16 convolutional neural network to extract feature maps. Then it detects the face by applying convolution filters. For each location, it makes 4 face predictions. Each prediction composes of a boundary box and 21 scores for each class (one extra class for no face), and we pick the highest score as the name for the detected face.
Firstly, the system is launched and then accessed so as to use the suitable camera in the system by checking the camera devices connected to the computer. The system will start after it accesses to a camera without a problem. A short time after the system is launched; the camera detects a human face and eye in front of it. It will put the detected face in a boundary box to show proof of face detection. If the face moves, the box follows its movement in order to process on the face. Afterward, it scans whether such a face is recorded by searching the detected face in the previous records. While doing this process, it uses the SSD algorithm. When the matching record is not found, persons can be prevented from logging into the system. When the matching record is found, user information recorded in the system is written on the face beforehand.

Technologies Used

I am using Pytorch Deep Learning Framework and based on an image detection algorithm called Single Shot Object Detection (SSD) and comparing the result with pre-recorded face samples.
Also, I intend on deploying my model on the Intel Movidius chipset with a Raspberry Pi device.