Narrow down by specifying conditions
現在1888件がヒットしています。check
In general, there is a perception that security cameras are used for real-time surveillance, but there are still many situations where past events are recorded and played back only after receiving a notification from the outside. The reason for this is that even though the number of cameras installed in large commercial facilities and public transportation has increased, the number of security guards has not increased. Even if we install cameras for various purposes such as security, safety management, and equipment maintenance, and invest a huge amount of money into the surveillance infrastructure, the reality is that we are not seeing any concrete results.
"Even if we reduce the number of patrols and patrol guards, we want to realize enhanced security in real time."
Isn't there a lot of companies who think?
Recently, AI image analysis has been attracting attention, and the number of systems that can detect "certain events" by analyzing camera images with AI has increased. For example, intrusion detection, lost item detection, people counting, face recognition, etc., the types and services are increasing explosively. On the other hand, I started to hear such voices from people who actually introduced it.
I tried to introduce it, but it does not meet the accuracy required in actual operation, and it is not possible to detect specific movements in crowded places.
Based on these experiences, there are cases in which people come to the conclusion that it is difficult to completely entrust monitoring to AI.
In this article, we would like to introduce the latest "anomaly detection" technology and its scope of application, which we would like people who have experienced this kind of experience to know.
Issues noticed after introducing a rule-based monitoring system
Rule-based monitoring is the detection of specific events and behavioral patterns. Deep learning technology is used to detect specific objects that have been specified in advance. It is a mechanism to issue an alert when an event occurs.
It has some weaknesses. It is difficult to detect events that cannot be defined in advance. For example, unexpected events, confusing events, and events that rarely occur cannot be ruled by humans. It is very difficult to fine-tune the camera installation location and angle. If you want to detect people or objects, it is necessary to adjust the camera accordingly if the target is different. Furthermore, since the behavior and methods that should be monitored, such as nuisance and dangerous behavior, are constantly changing, it is necessary to keep updating the rules. The background image that is the target object also changes depending on the season, and the layout of the store may change. The vulnerability to such changes was also a weak point of rule-based image analysis.
Monitoring required in the future
Efficiency of monitoring that has been given up with rule-based image analysis. On the other hand, improvement of the shortage of personnel is not expected. Based on these backgrounds, the monitoring that will be required in the future is monitoring that supports the discovery of "potentially abnormal events" rather than detecting specific events based on rules. By actively focusing on and observing abnormal events, you can concentrate on your core business and improve monitoring efficiency. There is a solution called "icetana" that we can introduce as a solution to realize it. I will explain what kind of anomaly detection is realized by icetana.
What is anomaly detection realized by icetana?
Rather than "detecting specific events" based on rules, it detects "events that can be abnormal". Instead of recognizing people and objects, it focuses on groups of pixels to detect position, density, direction of movement, and speed. By pre-learning camera images for a certain period of time and recognizing them as normal conditions, abnormal conditions are judged to be abnormal.
Features of icetana
1. One system can handle all events
Since the target to be detected is not a person or object, but a collection of pixels, it is possible to respond to various situations such as fires, intrusions, reverse driving, falls, and crowds.
2. No need to set rules
Since it can learn the normal state and detect abnormalities from the difference, there is no need to collect a large amount of data or define detailed rules at the time of initial setting. In addition, since there is no need to configure the settings for each camera, it is possible to detect hundreds of cameras.
3. Resistant to environmental changes
During operation, it learns based on the video of the most recent fixed period (eg, 2 weeks) and updates the normal state update, so it can respond to environmental changes in the season and background video. Therefore, there is no need to adjust the angle of view of the camera each time when the target object changes, or to re-learn according to environmental changes.
Using icetana, which realizes these technologies, there is no display on the monitor screen during normal times, and images are displayed only when an abnormality is detected. The administrator does not have to keep a close eye on it all the time, and can use it to judge whether it is necessary to rush to the detected image immediately after seeing it.
Widespread use industries and usage scenes
Since security patrols can be reduced and security can be strengthened, attention is increasing not only for security companies, but also for railway companies, commercial facilities, and the developer industry that operates these. Furthermore, by combining the data analysis results from anomaly detection with big data, it is expected to be added value to the company as a data asset for building and facility management and improvement.
There are four specific examples of usage:
“icetana” is recommended for anomaly detection
The "anomaly detection realized by icetana" introduced above is an optimal solution using the latest anomaly detection technology that can be provided as a means to accurately capture current anomalies in real time with surveillance cameras. It has been introduced in more than 35 countries around the world. With many years of AI/ machine learning development experience, we can quickly incorporate and introduce it to our customers.
Since it is provided in the form of anomaly detection software, there is no need to set each camera individually, and it can be easily connected to an existing surveillance camera system. In addition, with general rule-based video analysis, the number of cameras would be a dozen or so at most, but icetana can detect dozens to hundreds of cameras, enabling real-time processing of an extremely large number of camera images. It has features that make it possible.
Would you like to use it as a material for your consideration?
Click here for the solution page
Contact Us
In this article, we have introduced what is anomaly detection realized by icetana, the issues faced in the field, and their solutions. How was it?
The goal of icetana is not to completely entrust monitoring to AI, but to help human cognition and nurture it into an AI that interacts with humans. We will help you implement the solution in the field.
Please contact us if you are interested.
