Summary: The advancing Safe City and Xueliang Projects have made video surveillance all over the streets. According to incomplete statistics, a medium-sized city has about tens thousands of road monitors, and some even have 100,000 roads. The daily generated video data is equivalent to 100 billion pictures. However, due to the limitations of early technology and other factors, traditional video surveillance cameras have been unable to meet the needs of the current socio-economic development. With the improvement of front-end computing power and the further advancement of intelligent analysis software, smart cameras will undoubtedly become the development direction of the next generation of cameras.
The continuous advancement of the safe city and the construction of the Xueliang project allow video surveillance to be spread all over the streets. According to incomplete statistics, a medium-sized city may have tens of thousands of roads monitored and some even reach 100,000 roads. The daily generated video data is equivalent to 100 billion. image. However, due to the limitations of early technology and other factors, traditional video surveillance cameras have been unable to meet the needs of the current socio-economic development. With the improvement of front-end computing power and the further advancement of intelligent analysis software, WiFi security cameras will undoubtedly become the development direction of the next generation of cameras.
Traditional surveillance security cameras can only provide real-time monitoring of the situation or look back at something that has happened. WIFI Smart cameras that incorporate artificial intelligence technology allow users to monitor conditions in real time and identify problems before they occur. Dr. Mahesh Saptharishi, a data scientist, said: “A monitoring system with video analytics can perform analysis of video content and detect anomalies that may pose a threat. Basically, video analytics can help security software ‘learn’ what is normal, so that it Anomalies can be detected, as well as potentially harmful behaviors that an individual may ignore.” This is one of the key drivers of the combination of artificial intelligence and video surveillance. The idea behind this is that advanced software can improve human judgment and provide more accurate and safer monitoring. But this does not mean to replace human monitoring, but to make this process more nuanced and more personalized.
Artificial intelligence landing on video surveillance
The Intelligent monitoring aims at intelligently analyzing and judging objects such as objects, behaviors, and events in video surveillance through visual pattern recognition technologies such as detection, recognition, and tracking, thereby reducing or replacing human intervention. Technologies covered include human faces. Visual objects, pedestrians, vehicles, logos and other visual object identification and behavior analysis, etc., its application is mainly divided into the following categories:
1. Face recognition. Face recognition systems have many potential values. They can be combined with video surveillance systems to help law enforcement officers identify and identify the faces of target people in the crowd. This may help the police to track down criminals in the future and even prevent them from happening. , from the source of organized crime. Can be used for face verification, security face search.
2. License plate recognition. First take a picture of the license plate that has stopped the car, and then use the image detection method to detect the location of the license plate in the image, then extract and identify the license plate text, and collect the video stream of the passing vehicle in the lane to achieve the same license plate. Many times of the recognition, the final output after the optimization of the selected results generally do not need external trigger signal, has a strong ability to adapt to the vehicle blocking situation has a certain degree of resistance. It is mainly used for the registration and inquiries of residential vehicles, and the capture of illegal vehicles on highways.
3. Speech recognition. Speech recognition is divided into specific human speech recognition and non-specific human speech recognition based on the speaker’s requirements. Specific person speech recognition means that the current speech recognition system is designed to recognize the speech of a specific user. In this case, the audio samples in the database are all from the user, so the vocalization habits and speech rate of the language in the database are used. Both the tone and the tone are consistent with the user, which can significantly increase the recognition accuracy. Non-specific person speech recognition refers to the use of a common system for all users, low barriers to use, and strong system promotion. The main realization of human-computer interaction.
4. Expression recognition. Expression recognition refers to the separation of specific expression states from a given still image or dynamic video sequence to determine the psychological emotions of the identified objects, realize computer’s understanding and recognition of facial expressions, and fundamentally change the relationship between humans and computers. Relationships to achieve better human-computer interaction. Based on people’s expressions, different products are recommended.
5. Age recognition. The intelligent camera can almost accurately determine the age by accurately checking the skeletal structure of the identification object, the position of the eyes and mouth, and the wrinkles around the eyes and nose. Precise marketing by identifying the age of the guests
Lack of Experience, Time-Consuming and technically demanding
Although the current artificial intelligence technology has greatly improved the application of video surveillance, it still faces many challenges from the perspective of practical applications. After all, whether it is the development of artificial intelligence or the application of security intelligence, its overall level is still in the early stage or the initial stage. The degree of intelligence of the system is still difficult to achieve in the short-term science and technology shows. Even though static face recognition is very mature at present, dynamic face recognition still faces enormous challenges. In addition to front-end high-definition cameras that must be able to capture high-quality face information, they also need powerful algorithms and computational support. Through the comprehensive improvement of neural networks, deep learning, big data autonomous training, and high-performance parallel computing capabilities, we can solve the current application problems.
Deep learning requires a large amount of big data training. At present, some artificial intelligence companies use manual labor to label data, which is time-consuming and laborious. Prof. Li Ziqing of the Institute of Automation of the Chinese Academy of Sciences stated: “I think there is room for further development in deep learning. Whether it is to improve algorithms, improve network architecture, or increase the number of data annotations, there is not much room for improvement. It is close to the ceiling. Exactly how much, I can not give a quantitative, we must break through in this area, as Kaifu teacher said, to form a closed loop of the data of the application scenarios, can use the big data in the production process for independent labeling, independent learning, regardless of You will not be marked, at least autonomous learning.”
Secondly, security video surveillance is a systematic project. The application of artificial intelligence technology in video surveillance also runs through the front-end, transmission, storage, and applications. For example, looking for targets in massive video data, a large number of image videos generated by the “Tianwang” video surveillance system are like finding a needle in a haystack for finding people and vehicles.
In addition, transmission limited warnings are not real-time. Especially for the large number of applications of high-definition and ultra-high-definition cameras, the amount of collected data is very large, the transmission cost is very high, and it is difficult to bring the data to the total platform at the first time, which causes difficulties in global early warning and search. In addition, for traffic congestion, the current major traffic data is mainly based on navigation maps, sharing of travel software, etc. The application of video intelligent analysis data is still relatively small.
Of course, with the advancement of technologies such as neural networks, deep learning, and edge technology, the constant optimization of algorithms, and the improvement of computer performance, various problems currently facing will gradually be effectively solved.
Analyze data and handle it accurately
With the continuous acceleration of the human social science and technology evolution, the advent of artificial singularity in the era of singularity may be far faster than we think. The research report “Smart Home Surveillance Camera Market Analysis and Prediction” published by Strategy Analytics, a market research organization, on April 12 pointed out that cameras with a full set of software and service functions below the US$200 price point will promote the growth of the smart home surveillance camera market. The report predicts that by 2023 consumers in the global market will spend more than $9.7 billion on smart home surveillance cameras.
Gao Wen, an academician at the Chinese Academy of Engineering and professor at Peking University, said at the 2018 AI Security Summit that the city’s brain needs an eye of wisdom. “Using surveillance cameras to make cities smarter is not just a matter of single video search and computer vision, but whether they can react quickly, reduce the amount of calculations, and whether they can deal with massive amounts of information and unexpected events. Effective identification and retrieval of a large number of system engineering.”
In the future, smart security ip cameras will no longer be invariable after they are shipped from the factory. Instead, they will implement different business functions by loading different software according to the needs of users, and integrate into the new era of software-defined products. Accurate data collection through large-scale and diversified smart front-end cameras, powerful cloud computing and video analysis systems in the background to accurately analyze and interpret the collected data, and large-scale big data analysis and mining system for efficient and accurate mass data The processing can truly allow video surveillance to assist users in accurately observing, recognizing, and responding to the surrounding things, so as to truly embrace the smart age.
Conclusion: Although video surveillance integrates artificial intelligence technology, while monitoring is also accompanied by some potential risks, the potential advantages clearly outweigh the disadvantages. Nowadays, the maturity and cost reduction of artificial intelligence in the field of algorithms and chips have made the commercialization of intelligent monitoring popularized more rapidly. At the same time, the intelligent surveillance market is seeking a differentiated competition and has formed a flourishing situation.
Post time: May-22-2019