Approaches to Video Analytics Processing
Depending on the goal, video needs to be processed using different methods in order to deliver relevant results. Gorilla has categorized the most widely used types of analytics into five fundamental IVA groups which are described in more detail below.
Behavior Analytics
These analytics use algorithms that are designed to look for a specific behavior.
Thinking more deeply, a behavior could be defined as action over time. With that in mind, each Behavior Analytic needs more than one frame from the video to determine if an event or behavior has occurred. So it follows that the algorithms in Behavior Analytics look for changes from frame to frame over time to identify a very specific and predefined event or action. We've broken down and classified the Behavior Analytics that are used in our solutions here:
People Counting
The People Counting IVA does just that, it detects and counts people for a specified amount of time as they enter a zone and/or cross a line which users define in the software.
Line Crossing
This IVA detects when people cross a line (or lines) of user defined length and position.
Intrusion Detection
Intrusion Detection monitors user created zones to detect any activity or entries by moving objects (like people).
Direction Detection
This IVA monitors a user created zone for people moving A) within the zone AND B) in the marked direction. Movements in the opposite direction do not trigger an alert.
Direction Violation Detection
Same as the direction detection IVA but detects and alerts to movements in the opposite direction. As an example, security checks at airports and other transportation hubs stand to benefit from this type of IVA.
Loitering Detection
The Loitering Detection IVA monitors figures or people entering and then remaining in a user created zone for a specified period.
People/Face Recognition
People and Face Recognition could easily be sliced into two core groups, but we keep them as one since they are so closely related. As Behavior Analytics need to detect human shapes to perform effectively, People/Face Recognition IVAs are next.
Human Detection
The Human Detection IVA detects human figures within the video. Once detected, features like clothing color, gender, eyewear, masks, and age group can be detected as well.
Face Recognition
This IVA recognizes and identifies faces. This is used in conjunction with Gorilla's BAP software and its facial recognition database. While uses for this are myriad (and often in the news), we most often see Face Recognition used for Watch Lists, VIP identification, Attendance Systems, and Black Lists.
License Plate Recognition
Currently, we generally see this IVA added as a customized feature because adding all the different plates into the general release of the software would require too much space.
There are currently two approaches to LPR.
- Parking LPR detects the license plates of parked vehicles in user created zones, vehicles travelling slowly, or vehicles stopped at boom gates.
- Road Traffic LPR detects the license plates of moving vehicles, or vehicles stopped at a stoplight.
Object Recognition
Replace the Face Recognition IVA with any given object and you'll get the Object Detection IVA. This is where algorithms are used in training the software to detect and recognize a specific object, like a hot dog. There are a lot of different objects in the world, so the training and size requirements add up quickly.
Business Intelligence
Dashboards in software showing data about various business activities are a valuable asset in just about any retail or enterprise setting. Using video analytics from within a dashboard to enrich and increase results should be a part in everyone's toolbox.
While the IVAs in numbers one through four above are widely used for surveillance scenarios, there are magnitudes of business scenarios that can reap the benefits that video analytics offers. To see some examples, check out how Gorilla is applying these to create intelligent solutions for multiple business markets and industries.
Putting the Video Analytics Idea Together
As you read above, these IVAs all orchestrate various algorithms to achieve and deliver results. Essentially though, IVAs detect for and determine if a defined event or behavior has been found or occurs within a video camera's field of view and then notifies the designated user of the finding.
In a similar manner, most of us go through varying processes depending on if we're searching for keys at home or for our friend in a busy station.
Video Analytics Processing Power
Thinking about the entire process, could there be a single solution that can do everything effectively? It seems like an insurmountable amount of tasks: from processing each single frame's analytic data to displaying it together with the video, into creating a complete video system with an array of user selectable & customizable IVA in a building or any other scenario, all the way to putting multiple systems together that report back to a central control center.
It's not impossible. To demonstrate this, let's look at what IVAR™ from Gorilla can do and how it operates.
CPU and GPU Video Analytics Processing
Video analytics as a whole requires a lot of dedicated processing power. We should keep in mind here that before optimization and edge devices with capable CPUs, video analytics was processing both video and analytic data on one machine and required additional GPUs to do most of the work. Technology and the ability to split these two up has advanced to the point that it's now possible to keep the video data at the edge while pushing the analytic data up the network for quick processing.
One technology, which Gorilla was the first to adopt, is the Intel® distribution of the OpenVINO™ toolkit. Using the OpenVINO™ toolkit to optimize IVAR keeps deployment and upkeep costs low while decreasing operating temperatures by minimizing the need for expensive GPUs.