AI Magic Behavior Award: Nine Application Scenes of Final Machine Vision

Guide: This paper mainly introduces the main application scenarios of machine vision. At present, most of the digital information exists in the form of pictures or videos. If we want to analyze and utilize these information effectively, we need to rely on the development of machine vision technology. Although the existing technology can solve many problems, but it still needs to be improved. It is far from solving all the problems, so the application prospect of machine vision is very broad.

We eagerly hope that more readers will devote themselves to this field and join us in exploring the limitless potential of image data.

Author: Wei Xi with Tu Ming Zhang Xiupeng

For reprinting, please contact Big Data (ID: hzdashuju)

Image.png

_Fig. 1-1 Artificial Intelligence Related Domain Diagram

What is machine vision?

Machine vision is an important branch of artificial intelligence, and its core is to use "machine eyes" instead of human eyes. Machine vision system uses image/video acquisition device to input the captured image/video into the visual algorithm for calculation, and finally obtains the information needed by human beings. There are many kinds of visual algorithms mentioned here, such as traditional image processing methods and in-depth learning methods in recent years.

Fig. 1-2a shows a classified data set Cifar10 consisting of color images. There are 10 categories of airplanes, cars, birds, cats, deer, dogs, frogs, horses, boats and trucks. There are 1000 32 x 32 color images in each category. Figure 1-2b shows the classification effect of different algorithms on the Cifar10 dataset.

Image.png

_Figure 1-2a Cifar10 Data Set Display

Image.png

_Fig. 1-2b Comparison of Traditional Image Processing Method and Deep Learning Method on Cifar10 Data Set

We can see that before the emergence of in-depth learning, traditional image processing and machine learning methods can not do such a simple classification task well, and the emergence of in-depth learning makes it possible for machines to reach the human level. In fact, the emergence of AlphaGo has proved that machines have the ability to surpass human beings in some areas.

With the development of in-depth learning technology, the improvement of computing power and the growth of visual data, visual intelligent computing technology has made remarkable achievements in many applications.

Classical and new problems such as image and video recognition, detection, segmentation, generation, super-resolution, captioning, search and so on have made great breakthroughs. These technologies are widely used in urban governance, finance, industry, Internet and other fields.

The following will take 9 scenarios as examples to introduce some common application scenarios, so that readers can intuitively understand what problems machine vision can solve.

01 Face Recognition

Face Recognition (Face Recognition) is a biometric technology based on human facial feature information for identity recognition. It collects pictures or video streams containing faces, and automatically detects and tracks faces in pictures, and then recognizes the detected faces. Face recognition can provide the functions of face detection and location, face attribute recognition, face comparison and biopsy detection in image or video.

Face recognition is the most mature and popular field of machine vision. In recent years, face recognition has gradually surpassed fingerprint recognition as the dominant technology of biometrics. Face recognition is divided into four processes: face image acquisition and detection, face image preprocessing, face image feature extraction and matching and recognition. Its main applications and explanations are as follows:

Face Payment: Binding the face to the user's payment channel, the payment stage can be brushed face payment, without showing bank cards, mobile phones, etc., to improve the payment efficiency (Fig. 1-3)

Face Card Opening: When a customer opens a card in a bank or other department, he can verify his identity through ID card and face recognition to prevent borrowing ID card to open a card.

Face Logon: Enter face pictures in user registration phase, and start face logon verification in high security scenarios to improve security.

VIP Face Recognition: Automatically identify customers by face recognition and provide differentiated services

Face Check-in: Enter face pictures before the start of the activity, and check-in can be done by brushing the face on the day of the activity, so as to improve the efficiency of check-in.

Face attendance: Using high-precision face recognition and comparison capabilities, build attendance system to improve attendance efficiency and anti-cheating ability (as shown in Figure 1-3)

Face Brake: Identifying Passengers by Face Recognition in Airport, Railway and Customs

Membership identification: Members do not need to show their membership credentials when they come to the store, but only brush their faces to complete the membership identification, and realize the card-free identification and flow statistics.

Security monitoring: monitoring the crowd in the dense public places such as banks, airports, shopping malls and markets, so as to realize automatic statistics of the flow of people, automatic identification and tracking of specific persons.

Photo album classification: through face detection, automatic recognition of personas in photo library, and classification management, enhance the user experience of products

Face Beauty: Based on Face Detection and Key Point Recognition, it realizes the interactive entertainment functions such as special effect facial beauty, special effect camera, patch and so on.

Image.png

_Fig. 1-3 Face Recognition Application Scene

Due to the strong demand of face recognition industry, many large-scale technology companies and AI start-ups are involved in this field. At present, the technology is in a large-scale commercial stage, and will continue to grow at a high speed in the next 3 to 5 years.

Video surveillance analysis of 02

Video surveillance analysis is a technology that uses machine vision technology to quickly retrieve, query and analyze specific content information in video. Due to the wide use of cameras, by

Please read the Chinese version for details.