Security Industry: Unexplored Big Data Treasury

Millions of cameras have been connected to the Internet, which has brought us a lot of convenience for people’s financial security and transportation. However, this is only the beginning. In fact, the current technology can easily carry out intelligent identification, flow counting and even behavior recognition. And so on, this huge video data analysis is closer to the real world than the traditional Internet data. What will this Pandora's box bring?

I remember seeing a movie a few years ago and starring Will Smith as the "National Enemy". All the corners of the United States are filled with cameras. The CIA can call these cameras at any time to track. Countless cameras form a super network... ... At that time it felt a bit sci-fi and a little cold.

But in a few short years, this has become a reality. Millions of cameras have been connected to the Internet, which has brought us a lot of convenience for people's financial security and transportation. But this is only the beginning. In fact, the technology is now able to It is easy to carry out intelligent identification, flow counting, and even behavior recognition. This massive video data analysis is closer to the real world than traditional Internet data. What will this Pandora's box bring?

Video Surveillance will be the World's Largest Data Generator In 2012, the world added 2.8ZB of data, equivalent to 3 billion TB. If video data is also stored, this data will double.

Take a province in China as an example, public video surveillance has more than 1 million cameras, and the total number of cameras is nearly 4 million. It is estimated that the number of cameras in the country is no less than 40 million. An investment bank’s report states that the industry still uses 20% each year. Accelerated growth, the industry’s leading Hikvision (12-year-old) annual report disclosed sales of 5.7 million units (including front and rear), an increase of 37%. The other data is also very alarming. The ratio of the number of cameras in the UK to the population has reached 1:15. At least roughly 100 million cameras in the world looked at the corner quietly.

If this data is stored, it will be super data. Assuming all cameras are changed to 1080P, the daily data volume is 64G, the annual data volume is 23T, and the world will generate 2.3 billion T of video data each year.

This is a terrible number. Fortunately or unfortunately, most video surveillance data storage will not exceed one day, because from the current point of view, these data are worthless compared with the stored price.

The super video surveillance cloud video surveillance that is being formed is no longer the same as before. Several cameras are connected to the security room of the building. Only by entering this mysterious room can one get a glimpse of what happened. In fact, surveillance cameras have become smart terminals, with network ports, controllable and even two-way audio, and the interconnection of cameras has become a trend.

For example, public video has evolved from an interconnected city to a provincial interconnected network, gradually to a country's interconnection. Whether it is a new webcam or an old analog camera + DVS/DVR, it has started accelerating access to a huge Cloud platform, from simple video streaming data to cloud platform operations, each terminal has become a smart terminal that can be remotely controlled. At the other end, these data are packaged into a variety of public applications and open to the public to see how many mobile phones in the media are full of real-time news on urban road traffic videos.

In addition to public videos, numerous private videos are also forming private clouds. Branches of chain supermarkets, chain hotels, logistics companies, financial institutions, and even industrial production lines are gathering data from countless endpoints. Many countries also require private video data to be open and can be called and operated.

This is exactly the same thing as the Internet of Things Cloud Network, from perception to interconnection to application, but this perception is to see the world with countless “eyes”.

The value of video big data has not yet been explored. Generally speaking, there is too much redundant data in the video. Only when an incident occurs, the records are reviewed. Therefore, the storage of most video data is very short, and it is very rare for more than a week. On the other hand, due to high data read and write requirements, cloud storage is extremely expensive (a few thousand dollars 1TB), so most of the data is front-end. The original purpose of video intelligence analysis is to solve the problem of redundant data, liberate people from boring monitoring, and help people to better look at the six ways. The core of so-called intelligence is target recognition and behavior analysis. Rules are set according to the requirements of applications. Alarms and records are recorded when the information in the video meets the rules. Common applications include intrusion detection, perimeter alarm, vehicle identification, and traffic violation monitoring. Wait.

At present, the mainstream intelligent analysis vendors in the industry, such as foreign ObjectVideo, ioimage, Emza, domestic Zhuoyang Technology, Wen'an Technology, Zhianbang Technology, etc., are basically around the word “monitoring”, which belongs to the primary processing of video data. Only the video information of a single camera is processed in real time, and event data (such as alarm events and vehicle counts) are generated according to certain rules. The secondary processing analysis across space and time ranges is still relatively small. Therefore, the field of video surveillance is not really formed. Big data in the internet sense.

However, video surveillance data can certainly become the treasure house of the next big data. On the one hand, it has typical big data 4V features, huge amounts of data, diversification, superficial disorder, but implies the behavior of countless people and things. On the other hand, it is a portrayal of the real world, which is very different from the big data obtained on the Internet. The real world contains countless information that is difficult to express in formatted text. For example, people can quickly form judgments through vision, and one place is prosperous. Still recession, the atmosphere is tense or cheerful.

Of course, the premise is that the cost of storage can be reduced and data processing capacity can be upgraded.

For example, for a shopping mall, in addition to the security needs, the second excavation of video data can collect the customer's gender, age, wearing information, can count the customer's shopping path, stay mode, gather hot spots, or even two times or Repeatedly returned to the store cycle. This is similar to website visit analysis, which can provide basic data for the optimization of shopping malls. If it is a chain-type enterprise, it can combine data from multiple stores to obtain regional or even national data.

It also zooms into a city. The data on the flow of people and traffic in each street are all in it. The data is collected as a whole. It is a map of the distribution of people and cars in a city. The characteristics of people in different regions, characteristics of cars, and dynamics, such as The path of people and vehicles and the mode of stay are extremely valuable to the planning and management of the city.

If combined with the timeline, information gathered from countless cameras can also be seen in a country, a region, a city change, as "big data" said, and even predict the trend, such as whether more shops In the newly renovated or more closed business, more hotels or more clothing stores, etc. These changes bring together changes in population, economic trends, trend trends, changes in the natural environment, and even people’s Happy and nervous.

This is not a fantasy. The technology is now fully implemented. On the one hand, the popularity of high-definition video cameras, the quality of video information is more excellent; the other hand, the level of intelligent analysis has been quite high, the identification and separation of objects, the recognition of the face, the identification of color characters, changes in the object The analysis even monitored the violence.

However, to realize data mining in this sense, a large amount of metadata records are needed, even metadata that is not related to monitoring purposes. To analyze the data, a large amount of data is required to store the data and the data processing is very large-scale. Finally, it is also necessary to carry out correlation analysis and integration based on location and time. This huge resource and cost is an obstacle to the opening of big data.