Argus Algorithms
Argus uses a Template Based Neural Network feature classification technique to implement its face recognition algorithm. A Neural Networks Hidden Markov model is used to determine the facial features. Our models are trained for multitude of faces of all ethnicity, race and culture. This makes Argus, one of the comprehensive face recognition algorithm available in the market.
Argus uses a Deep learning Convolutional Neural networks based approach to implement its Human detection algorithm. Unlike image processing technique, which requires an excess CPU time, Argus human detection uses AI models to find and track human objects in a video frame. The algorithm is equipped to track human movements as-well. Argus uses human pose estimation algorithm and skeletal tracking techniques to determine the human motion and the direction of the movement.
People count is not an algorithm per se, it is an implementation of Human Detection algorithm and Human Movement algorithm. However, Argus has developed a niche technique to accurately determine the people count in a closed area, to the accuracy of 99.99%, at a given moment in time. Our customers have been using it to implement Emergency Assembly site management during an evacuation.
Detecting fire and smoke from the video frame is one of the difficult problem in Video Analytics. Argus uses a Deep learning convolutional Neural Network to detect the fire and smoke. We had collected samples from multitude of sources in different environmental condition to train our algorithm. Argus can predict the Fire and Smoke with 90% of accuracy.
Infra cost
Argus is ONVIF compliant. Hence it integrates with 3000 models of IP cameras. Customers who deploys Argus powered products don’t spend on new camera, network equipment or high end servers. Argus seamlessly plugs into the existing security infrastructure. Argus is also compatible with famous brands of NVR and DVR, as long as it is IP enabled and ONVIF compatible.
A huge share of capital investment in security infrastructure is on the storage. Storage is costly and maintenance of storage is even costlier. Argus understands this pain point. So, we uses the existing storage of the NVR/DVR for evidence retention. If it is possible, we use the software features of NVR/DVR to tag, star, label or lock (different brands uses different terms, but functional aspect of the feature is the same) to preserve a video segment of interest.
Argus, platform is operating system agnostic. It works on Windows or Linux. Some algorithms may require GPU support or specialized hardware. However, most Argus application only need an acceder appliance. Acceder appliance is powered by a single board computer with GPU and a Linux OS. And customers would connect 4-24 cameras at once to one acceder appliance.
The Argus engine is build with cloud in mind. Argus engine is deployed in Docker containers and works well with the AWS Serverless architecture. Customers can easily port the platform to AWS cloud instance and uses Argus RESTful API for integration. Further, Argus is integrated with Amazon Kinesis Video stream and can analyze kinesis streams as-well.
Integration
Argus powered hardware appliances (eg. acceder) have GPIO ports that can receive and transmit signals customized for the application. Appliance has the ability to integrate with Arduino NANO and Raspberry PI based boards. Argus can output signals to relays directly or drive an Arduino or Raspberry PIs. Argus also contains an inbuilt mobbus module that can drive a network PLC
(Programmable logic array).
You can integrate external software with Argus through Webhooks or WebSockets. Events are send out as web call backs or streams through Websockets. In case if you need help integrating to Argus, Perleybrook provides Professional Services to help with integration.
A RESTful API exposed by Argus can receive an image in jpeg or png format and can list objects in the image scene. There are two entry points for the REST API interface. One is a general image extractor and one for Face Recognition. The General image extractor classifies objects as Humans, Birds, Vehicles etc. Face Recognition end point returns the facial features. Embedded object and feature list are passed as json sting in a RESTful response object.
Argus provides output video streams in either RTSP, RTMP or Apple HLM format. The streams are never decoded before and re-streamed directly as it gets. It maintains all the source properties of the originating video stream, i.e a camera connected to the argus. External NVR/DVR can store record and store the videos as if it originated from a camera. In ONVIF parlance, Argus hardware apparatus will manifest itself as a n channel device.
Depending on how the Argus is deployed, you could use a full fledge RESTful APIs to integrate with Argus. Example, acceder; acceder has Argus running on a hardware edge IoT device. The device is hooked to an AWS backend. Customers uses the RESTful API of the cloud backend to integrate with the attendance and visitor management data with their in house software.