When Premier Li Keqiang of the State Council delivered the government work report at the Fifth Session of the 12th National People's Congress, he stated that "we should fully implement the development plan for strategic emerging industries and accelerate the R & D and transformation of technologies such as new materials, artificial intelligence, integrated circuits, biopharmaceuticals, and fifth - generation mobile communications." This was the first time the term "artificial intelligence" appeared in the government work report.

In the field of public security, especially in the application of security product technologies, the application of artificial intelligence has been put into practice. At the end of 2016, at the Beijing Security Expo, many security enterprises successively proposed solutions such as high - performance chips + machine vision and deep - learning algorithms. At the same time, solutions based on GPU computing, face recognition, and big - data applications have become the consensus among manufacturers.

So, what is the current development level of artificial intelligence in the security field? What obstacles and difficulties need to be overcome? What is the future development trend? These are well worth our attention and discussion. Based on this, Mr. Gu Changhai, the R & D director of Beijing ZhongShengYiHua Technology Co., Ltd., elaborated on the following questions:

What is the current implementation status of artificial intelligence in the security field? At what stage is its development level?

1、Artificial intelligence in the security field mainly focuses on people, vehicles, and behaviors. For people, artificial intelligence mainly includes face recognition and pedestrian recognition. Face - recognition features include gender, age, ethnicity, glasses, smiles, and facial feature data. Pedestrian - recognition features include backpacks, satchels, suitcases, skirts, hats, umbrellas, hair, and scarves. Currently, in the security field, for face applications, cameras installed according to face - capture requirements at fixed points can achieve practical application. However, for general - purpose security cameras, their application level is greatly reduced and not suitable for practical use. Regarding pedestrian - feature recognition, due to the influence of camera resolution, light, and angle, high - precision human - body recognition cannot be achieved with the current technological level. Vehicle - feature recognition is relatively mature and can be put into practical use in toll - gate/micro - toll - gate systems. However, for security cameras, the accuracy is affected by light and angle, and the accuracy drops rapidly, making it unable to be put into practical use. Abnormal behaviors such as crossing a virtual line, entering a restricted area, and object - leaving can be put into practical use. Other behaviors such as loitering, gathering, and detecting flames can only be put into practical use in specific scenarios and cannot be used in general scenarios.

2、Some people in the industry believe that security is the application field with the most market potential for artificial intelligence. What's your view on this?

I agree with this view. Video is the most widely used data in the security field, and video structured description is the most direct expression of artificial intelligence. With the country's emphasis on maintaining stability, video structured description is facing an explosive growth pattern. Therefore, artificial intelligence has the most market potential in the security field.

3、What do you think are the key technologies that support the implementation of artificial intelligence in the security field? Why?

The key technologies that support the implementation of artificial intelligence in the security field are deep learning and efficient computing. As we all know, the emergence of deep learning has led to rapid development in face - recognition technology, which has changed from the laboratory stage to a technology that can be used on - site. However, another negative effect of deep learning is the extremely large amount of computation. Since traditional CPUs are not suitable for parallel image processing, face - recognition solutions come at a high cost. The emergence of high - density computing devices such as GPUs (or TPUs) has greatly alleviated the demand for computing resources in deep learning, enabling the final implementation of artificial intelligence.

4、From the perspective of market application, which fields do you think have a stronger demand for security artificial intelligence?

As mentioned above, artificial intelligence has emerged prominently in the field of video analysis, enabling the practical application of relevant intelligent - analysis functions such as face recognition, vehicle recognition, and abnormal - behavior recognition. This also means that fields with a strong demand for video also have a strong demand for artificial intelligence. Fields such as public security, transportation, finance, industrial parks, and prisons have a stronger demand for security artificial intelligence.

5、With the affirmation of artificial intelligence at the national - policy level, what opportunities do you think there are for artificial intelligence in the security field?

In March 2017, "artificial intelligence" was written into the government work report for the first time, and many artificial - intelligence concept stocks also performed well during the Two Sessions. Video is the core data in the security field, and the actual security needs are met through the intuitive display of videos. As the number of videos in the security field gradually increases, it is no longer possible to view and monitor them in real - time with the human eye. At this time, through the structured description of unstructured video data, the goal of quickly locating the area of interest in the video and quickly retrieving and searching can be achieved, realizing data analysis and information collision similar to that of the human brain and forming industry - data applications in the security field. This will lay the foundation for the final realization of intelligence and intensification in the security field. All these will greatly enhance the application prospects of the video - security field and create various opportunities for artificial intelligence in the security field.

6、What challenges, technical or market bottlenecks do you think artificial intelligence still faces in the security industry that need to be overcome?

The primary challenge that artificial intelligence faces in the security industry is the universality of technology, such as the influence of light, resolution, and environment. Taking face recognition as an example, there is a huge difference in the quantity and quality of face capture between cameras set up according to face - collection standards and ordinary security cameras. Secondly, there is a demand for technical computing. As we know, artificial intelligence requires a large amount of machine computing, which places high requirements on system construction and maintenance. Still taking face recognition as an example, the currently popular CPU (i7) can only handle face capture and feature extraction for 3 - 4 channels of 1080P - resolution video. If a large amount of face data needs to be collected and stored, a large amount of computing resources are required. Now, many companies use GPUs to improve computing efficiency, but how to ensure the 7×24 non - stop and stable operation of GPUs remains a difficult problem for each company. Finally, in the market, artificial intelligence is still in the early stage of application, which means it has a relatively high price. If it is to be widely applied, there will inevitably be a demand for a more reasonable price.

7、What is your prediction for the future development trend of artificial intelligence in the security field? In which technological or application fields will there be major breakthroughs and progress?

With the wide application of artificial intelligence in the security field, it will inevitably give rise to technological innovations more suitable for application scenarios, enabling artificial intelligence to adapt to various application scenarios and truly achieve practical use.