Video monitoring and security comes with a host of challenges. What are you looking for? Who will look for it? What video will be searched and when? Will the person looking for the information catch it? Can you afford to pay someone to spend countless hours searching for specified information? What if you need to be alerted to an action as soon as it happens? Relying on the human eye alone is no longer an efficient way to monitor video analytics. That’s why Equitus Video Sentinel was created.
Video Sentinel is a video analytics and security monitoring system that provides fully customized alerts and real-time analysis to easily detect suspicious patterns or behavior on video faster and more efficiently. It seamlessly integrates with third-party applications and supports forensic searching on recorded video footage.
Video analytics takes advantage of your existing cameras, your existing video management system, and your existing video archives. It makes video monitoring infinitely easier for anyone who has hours of film to pour through, or for those who need an alert system 24/7 without requiring human monitoring for that duration. With the growing need to be alerted to events quickly, Video Sentinel was created to save its users hours of pouring through film searching for events, people, and even vehicles.
Video Sentinel is fully customizable no matter what the use case. When you set up your own customized alerts, you can categorize them by priority type: low, medium, high, and urgent. All the alerts in the system have a thumbnail, and you can see details about each event, including the outlines of the binding box itself (x y coordinates). The system works on low resolution cameras, which saves money because it requires less GPU. There is an option within each alert to go back and play the original video to search for more detail. You can even create compound alerts within the system, for example, you can set an alert for a specific person and a specific car. You can then set the alert to “Urgent” so that these occurrences will be as “Urgent” priority as they arise.
When searching for events, you can search for information relative to the current time or look back to a specified time loop. You can select which cameras you want to search, select what areas you want to search (bridge, ramp, alleyway etc.) then specify what you’re looking for. For example, if you’re looking for a specific individual who has been put into the system, it will show which camera picked up those characteristics in the criteria. You can see the probability that the image matches your criteria. To “call out” a specific incidence, simply click on it. You can pull the metadata to the external data system as needed.
Video frames from existing equipment and archives are collected into a semantic streams engine. The semantic streams engine digests video frames. It performs interference through a deep learning engine and uses GPUs as required to validate alerts to confirm what is there. The final validated output is sent into a metadata engine. Metadata and alerting are either sent back to the video management system, or the external management system. It does not recreate the video archives; it collects metadata and allows you to watch a near real time video from the video management system. When you wish to replay the event, you can go into the alert system and click on one of the events to call the video from the video archives with no double storage. You can customize the system to recognize specific types of vehicles, uniforms, traffic cones, etc. by using the custom object modeler, or you can take external models and import them into the deep learning engine.
Let’s dive into an example of how you can easily use Video Sentinel to search for events and create alerts. (The images used below are fictitious portrayals for demonstration purposes only.)
Let’s say we have a client who wants to be alerted to know where a particular suspect is located. After we load their image into Video Sentinel, we can create a model from his image, and the system can easily identify them in future recordings. We can search for instances of the suspect in any and all recordings once this model has been established, meaning we can search for all events captured that include them.
We can set alerts for incidences of the suspect whenever they appear on cameras connected to the system. In this scenario, we can look even look at compound alerts, such as occurrences of this man and a specific car he drives. From the model we have created, we can set the system to alert us whenever he appears on one of our cameras. We can even set alerts for certain events as higher priority, such as setting an urgent alert for every instance that the man is recorded driving his car.
Once you customize the system to alerts of your choosing, the alerts come in as soon as they happen. The system will pick up instances of this alert on whatever cameras are connected to the system. This means you can expand your reach as you add equipment.
Video sentinel uses machine learning and AI to continuously update the knowledge base of the system as new footage gets digested. It goes beyond saving time as it helps connect information that is interrelated. Its truly a game changer for video analytics management. The possibilities for use cases with Video Sentinel are endless, to view a full demo:
USF English Major and current Copywriter at Equitus