Research on Modern Population Management and Control Technology Based on Big Data and Artificial Intelligence

This paper proposes a modern population management and control technology based on big data and artificial intelligence analysis, studies the portrait recognition technology under complex conditions, and successfully applied the results to the grassroots public security combat, and explored a road of public security informatization construction integrating integrated reality. . Its achievements can effectively integrate personnel information resources across police types and across departments, effectively raise the level of informationization in the public safety field, and achieve outstanding results in actual combat.

(a) Background / Introduction:

With the rapid development of the economy and the acceleration of urbanization, China’s urban population has become increasingly dense and the mobility of urban population has greatly increased. However, various criminal activities have also grown explosively, and criminal means have become more and more abundant and secretive, such as through the use of multiple Identity cards and fake ID cards have been used to commit crimes, escaping from arrests, combined with convenient transportation and large population mobility, making it more difficult for public security staff to crack down on arresting criminal suspects.

The current urban public security video surveillance system is in the late stage of the large-scale construction phase. Local governments also invest a lot of money to build a safe city, but only rely on the captured video images. The current status of the current application is that only the monitoring centers at various levels take a look at the video of each monitoring point, and after the occurrence of the case, the criminal investigation department looks for clues to look through the video. This simple application is obviously not enough for tens or even hundreds of millions of construction investment, or millions of rental expenses per year. At the same time, in a large-scale video surveillance system, because of its lack of intelligent analysis of video, it is necessary to find out criminal suspects in massive surveillance videos and millions of photo libraries, which not only takes time and effort, but also may cause omissions. The efficiency of cracking the case is greatly reduced. There is no real-time recording and early warning of security incidents; long-term observation of surveillance video also highlights the limitations of personnel fatigue; and massive video data is difficult to manage and effectively view. The support for intelligent information technology is needed to solve the case of public security criminal investigation video, real-time deployment of suspects and key high-risk personnel.

Further, because there is no unified construction of an information platform, the information resources of various departments of the public security organs cannot be fully utilized, nor can they provide prompt, accurate and detailed reliable evidence for the decision of the leaders. The informationization of various departments at all levels is basically at the level of their respective applications. It has not reached the information sharing among the various police departments and the application resources of various types of information have not been fully tapped and the rational and effective comprehensive utilization has not yet been achieved. Providing scientific and timely services for the leaders' decision-making cannot provide front-line police with comprehensive information support for all-weather, all-around, and full-time processes.

In the era of artificial intelligence, face recognition technology addresses the structural changes in unstructured information such as videos and pictures, and photographs and identity information of people in the above situations. In response to the existing problems of social stability, counter-terrorism work pressure, limited police resources, and difficulties in supporting traditional work methods and technical means, it is necessary to establish a professional application platform for combating large-scale data on public security to assist the public security intelligence service. . For example, video investigation on suspects requires the timely collection and analysis of data on the dynamic trajectory and social relations of the suspects. The prevention and control of key venues cannot be separated from the in-depth analysis of the trends of pedestrians and risk factors in the venues. The service to key groups cannot be separated from the analysis of the composition and behavioral habits of key populations; the solution to social conflicts can not be separated from the behind-the-scenes push.

Therefore, combined with the actual combat of public security and research on modern population control techniques and applications based on big data and artificial intelligence analysis, we can maximize the role of various types of human image data resources, provide services for public security organs and police at all levels, and achieve " Face finding people, digging in the actual value of various types of image data, providing a simple, efficient and practical technical means for the actual combat of the public security.

At the same time, the “Embed” application of portraits is applied to the video surveillance system in the city of Pingcheng. In the policing and criminal investigation business, effective supervision of suspicious identities, evasion, etc. can be achieved, and existing video surveillance resources can be effectively used. By checking the personnel of the library after the query, confirming the identity of a person appearing under a certain camera can effectively deter crimes and eliminate social security risks.

(II) Key technologies

Portrait matching is to find a description of a portrait, which can be influenced by various factors. However, both the earliest used geometric description method and the more commonly used algebraic description method inevitably have various disturbances. Because of the above-mentioned various problems in the process of face recognition, in the actual detection and recognition process, when these factors are superimposed together, the situation becomes more complicated and minor light changes occur. It may cause recognition failure. The research in this paper is based on a deep neural network. Through the introduction of deep learning technology, the accuracy of portrait matching is improved, and it is adapted to the actual needs of public security.

First, hierarchical vectorization model

Figure: Flowchart of single layer feature coding

In order to solve the problem that deep neural network needs a large amount of data, we propose a hierarchical vectorized multimedia information expression system. Hierarchical vectorization is actually a multi-layer feature encoding process. A single-layer feature encoding consists of the following steps: First, all face images in the image library are partitioned; second, local features (such as LBP and SIFT) are extracted from each region to form local feature descriptors; then, All local features are quantified to form a dictionary; finally, according to the mapping of the dictionary information and the face image, the feature vector of the face image is encoded, and we define the feature vector as human face DNA.

The face DNA feature can describe the invariant of a specific face well. This feature has certain anti-interference to the face ray, angle, expression and various picture noises, and is then optimized by the two-layer heterogeneous neural network. Learning, face discrimination is stronger, better recognition.

For example: We recognize a person, the simplest from the person's height, body type, hairstyle, etc. to determine who (cognitive first); deeper from this person's face, bones, iris, fingerprint To confirm this person's identity (cognitive second layer); deeper, we can use this person's DNA to confirm the person's true identity (cognitive third). So to recognize a person, with a deeper layer, one layer is more reliable than one layer.

Face DNA is similar. In the computer face recognition process, we can put the most external features of the human face in eye size and shape (danhyphora, thick eyebrows, etc.), nose shape (hawk nose, flat nose), and the size and shape of the mouth. (Cherry mouth) is understood as the first layer; the distance of the eyes, the position of facial features, the contour of the face, etc. can be understood as the second layer; the face information is more abstract and the face is not affected by the light, angle, age, etc. The feature is a deeper layer that we define as face DNA.

Second, two-layer heterogeneous neural network

In order to compare the two photographs in the same feature space for comparison, we propose a two-layer heterogeneous neural network model based on heterogeneous neural networks. Each layer in this model is an in-depth network (two photos as input respectively). Using the classifier loss function in training and regularizing the difference in the corresponding weights in the two networks, different image spaces can be realized. Mapping of the same feature space. In the feature space, intra-class differences of face images of the same identity become smaller, and differences between faces of different identity face images become larger, thereby enhancing the discriminability of the features.

Let's take the example of a person's identification card: a person's passport must be compared with the on-site capture camera or the face on an ordinary photo. We can't use it directly for comparison. This is because of the influence of age, light, and other information, the recognition is not. quasi. We should send the passports to a layer of the deep neural network. The scenes are sent to another layer of the deep neural network. The two photos exchange information through two different networks (age gap, angle gap, lighting effects, etc.). Gradually remove these factors that are unfavorable to face recognition, and map the two faces into the same comparable space for comparison.

Figure: Schematic diagram of a double-heterogeneous deep neural network

(III) System Framework

The modern population management and control system researched and designed in this paper adopts a four-tiered structure of “Ministry-Province-City-County” and is divided into two parts, a video special network platform and a public security network platform.

Figure: Overall topology of the modern population control system platform

Video image processing module is deployed in the professional combat platforms of video private networks and public security networks at various levels to analyze the video stream, including the generation of face capture images, face modeling features, and portrait structure data, and modeling data. The structured data is synchronously pushed to the public security network via the security platform for aggregation and storage.

Video private network real battle platform mainly performs feature extraction modeling and structuring on the video stream and bayonet data collected by the platform at the same level, integrates the key features of the lower level platform and structured data to build a human face foundation database, and carries out non-compromised events. Face static comparison and face dynamic deployment. At the same time, it combines the multi-dimensional sensing data collected by the platform, such as alarms and wifi probes, to construct a big data collision database, focusing on the identification of target personnel information and activities, and the prediction of recent trends, and finally confirming the focus of the active area of ​​the target. , to achieve active prevention and control.

Construction Content:

1. In terms of front-end construction, building front-end face data collection points based on the Internet, video private network, and public security network in important areas of the city and districts, including: front-facing frontages, closed-cell communities, hotels, and merchants in all residents' residential buildings. The front of people's faces in important areas such as residential buildings, office entrances and entrances, and intersections, and meets the data requirements for personnel image information in various departments, police types, and intelligence research, investigation, crime prevention, and counter-terrorism and riot prevention;

2. To build a portrait big data platform based on the Internet, video private network, and public security network to complete the convergence, association, and collision of portrait data, establish the correlation between portrait data and public security professional data, and form a unified data fusion database. All kinds of business service applications that directly face users;

3. Build a batch of big data applications to greatly increase the effectiveness of public security departments in the use of video resources, including key personnel control and various types of big data techniques;

4, to achieve comprehensive application, the ultimate goal of all types of applications is: for the police to provide high-performance, high intelligence, high-performance, practical and strong face of the actual combat business;

(d) Typical Applications

Quanzhou Public Security Bureau Linjiang Police Station covers an area of ​​1.7 square kilometers. Linjiang belongs to the old city. There are a large number of ancient residential areas, old-fashioned self-built residential areas, open-style non-residential communities, back streets, and religious sites in the area. Major sites and locations such as hospitals, schools, and schools are concentrated, and public security management is complicated. At the same time, the detection of many cases now depends on the extension of video surveillance. It is simply relying on the naked eye to find valuable information in this massive video surveillance resource, which consumes a lot of manpower and energy.

Under the guidance of leaders of the municipal and branch offices, the Municipal Bureau, Yucheng Sub-bureau, Linjiang Institute and Yunshang Innovation innovate on the basis of actual practical police needs and public security prevention and control requirements of the Linjiang area. Video surveillance face recognition as the core, joint research and development of face big data platform, through the coverage of the key parts of the area of ​​video and intelligent face recognition analysis, combined with all kinds of daily policing combat applications, formed a set of face to the core "Initiate, prevent, manage and control" the intelligent active three-dimensional prevention and control system.

Quanzhou face big data platform is based on face recognition and big data analysis technology. Based on in-depth understanding of the actual needs of the public security business, it integrates ArcGIS and other offline map technologies, and accesses surveillance cameras, various types of human witness verification equipment, etc. With a precise search, city-wide tracking, one-click deployment, regional inspections, key personnel management, community floating personnel management and other types of face public security tactics as the core of the public security intelligent face recognition business combat system.

The Quanzhou system has been operating steadily since its construction in January this year, and the functions and performance of various systems have been continuously improved. Now it has formed a highly effective and mature face recognition combat system that is based on dual-network and dual-platform, with all kinds of rich business combat methods, and is well received by first-line grass-roots police.

The Quanzhou system has successfully alerted the two branches of the country’s fugitives to the Sub-bureau and assisted the Sub-bureau and Linjiang Institute to quickly crack down on “3.3 Yi Jianfu Stealing Car Case”, “Huang Yongcheng Stealing Electric Vehicle Cases”, and “Xie Shaohua Stealing Electric Vehicles” and other series. Theft case. The recognition rate and function of the system are also constantly improving. The Linjiang police station also incorporated key daily patrols into its daily work through the police service APP daily alarms and the statistical analysis of frequency of strangers in the community. Based on the Quanzhou system, the Xinfeng community "smart guard" system has greatly improved the residents' sense of security and is well received by the people.

(5) Conclusion

This paper proposes a modern population management and control technology based on big data and artificial intelligence analysis. Its research achievements have been successfully applied to grassroots public security combat systems. Through the sharing of image database information, the integration of personnel information resources across police and inter-departmental polices has been strengthened. Comprehensive utilization will effectively improve the tracking ability of special populations in the public security field, such as monitoring of special groups, deployment of terrorists, and pursuit of people involved in crimes. It will play a positive role in social security management, criminal investigation analysis, and anti-terrorism maintenance.

Wen/Li Xiafeng Yun, Director of Technology R&D