Rhombus Systems is excited to announce Unusual Behavior Detection (UBD), a collection of enhanced machine learning algorithms to enable human stance and unusual behavior detection. Built on a convolutional neural network that identifies key body parts, Rhombus now has the ability to distinguish human stances and alert you for out of the ordinary behavior.
Since this is brand new technology for the space, this feature is in Beta but fully operational. Please give it a try and let us know if you encounter any issues or have any feedback.
Identifying people lies at the center of almost all security systems, with features such as facial recognition or people counting being highly desirable. Rhombus is already able to detect humans and faces, but we wanted to take it one step further by making our algorithms more powerful than ever.
With human stance detection, our cameras have the ability to understand a person’s movements. For example, if someone is walking through the office, the cameras will gather their positional data as someone who is walking and upright. If for some reason this person collapses, our cameras can then interpret this scene as a person who has fallen and is now laying on the ground. This Fall Detection feature combined with our Unusual Behavior Detection can then send real-time alerts to the system manager to handle the situation appropriately. This feature offers organizations unparalleled insights into their employees’ health and well-being and serves as a proactive tool that can monitor if accidents, injuries, or other medical situations were to occur.
We want to provide you with analytical information on what’s going on at your business at all times. By exploring deep learning methods of identifying human poses from video, we engineered a method that will work with multiple people in the same frame. Our first version gathers key body points and detects whether a person is upright, sitting, or laying. In addition, we have laid the framework to expand these stances for later versions as we gather more data.
We saw human stance detection as only the first step in a process of providing actionable intelligence from human poses. It occurred to us that on a given day, each camera witnesses people going about their normal behavior 99% of the time. This is not terribly exciting, even if we can determine their stance with good accuracy. So, we set out to develop a system to identify people with unusual behavior making up that last 1%.
With unusual behavior detection, you can receive alerts if a camera detects abnormal behavior including people falling, climbing fences, altercations, thefts, and much more. Each camera learns what normal poses are for their specific location and then computes a probability that a newly detected pose is unusual. Poses with high probabilities of unusual behavior trigger an alert so that you can be better informed of events under the watch of your cameras without having to monitor a video feed. This is unique from other behavior alert systems that only look for a specific action. Our solution is much more flexible and can be customized from camera to camera. After all, it doesn’t make sense for a camera inside an office to be looking for people climbing fences!
Check out how UBD works on our office test subjects interacting with the office dog:
We’re proud and excited to reveal this initial version of UBD! As we continue to gather more information, we will refine stance detection labels and provide better human behavior reports of a given event. Our goal is to be as accurate in detections and alerts as possible to provide you the best information in the most easily accessible way.
If you're interested in implementing the UBD license for your organization, please contact us at firstname.lastname@example.org.