Surveillance camera and face recognition

As one of the most popular application area of deep learning, image recognition application like face recognition becomes more and more common in daily life. Some interesting applications of face recognition include identity confirming for financial apps on phones and clock in system. Also, the technique has been embedded in many newly introduced smart cameras for various applications.

One specific attention-getting application in China is for social security. Chinese government is replacing many of its current surveillance cameras in train station and airports with these smart ones to automatically recognize criminals. However, this kind of technique is also causing some concerns. In a recent article from New York Times claims “China is reversing the commonly held vision of technology as a great democratizer, bringing people more freedom and connecting them to the world. In China, it has brought control.” [1] It suggests that Chinese government is using the surveillance camera and face recognition technique to monitor people. On the one hand, I believe this kind of concern is not singly for Chinese government. On the other hand, I do not think this concern is necessary at least for now since the technique is simply not ready for monitoring and tracking people in the way people think in Person of Interest. Discussion is very welcome.

Firstly, in order to identity all the people show in the camera in the public area, the system will need clear and recent photos of all the people. The way the system works is that the embedded chip will locate a person’s face in the video, compare the face with all the photos in its database and find the most similar one, which means that the embedded system will need photos of all the people and these photos will need to be clear enough and updated frequently, letting alone that the requirement of storage and computational power is huge. Thus most common application for smart surveillance camera currently is to recognize a small group of people such as criminals or employees who need to enter their office.

Also, even if we assume the government maintains a huge database of everyone’s photo carefully, the surveillance cameras are still not working well. The key for the face recognition system to work is that it typical needs relative clear full-face photo of people in a relatively good or similar lighting condition. In reality, the videos from surveillance cameras in public areas are shot in various environments with various lighting conditions. The screen shots of people’s faces are commonly not good enough for recognition and the efficiency is still low.

There are indeed many companies world widely focusing on using deep learning for recognition and they are raising a lot of money from both private and public sectors [2]. Some of them like Sensetime in China work closely with government and got data from the government. It will be interesting to see how this technique will go forward. But for now, I will not feel being monitored in front of a surveillance camera in an airport.

 

Reference:

[1] https://www.nytimes.com/2018/07/08/business/china-surveillance-technology.html

[2] http://mini.eastday.com/mobile/171112234716142.html#