Easy and quick video analytics configuration
Virtual sensors (“rules”) can be configured on the acquired images, or portions of the images, by using a web-based configuration interface supported by the commonest browsers and therefore easily accessible by workstations with different operating systems.
The interface’s ease of use allows to quickly change the number and type of sensors. The number and type of sensors can be quickly changed, allowing several rules to be configured on each single video camera to simultaneously detect different types of events without interfering with each other.
Furthermore, default values and/or suggested ranges are available for all parameters and thresholds, allowing even video processing novices to perform analytics configuration with no need for complex measurements or evaluations.
Post-event analytics on recordings
The video analytics software modules can also be applied to recordings, allowing to quickly and automatically reconstruct events of interest, without having to examine hours of recordings. Events can be quickly detected just by drawing the virtual sensors on the images: post-event analytics are performed at the highest possible speed, allowing to process hours of recordings in a few minutes!
Real-time alarm notification
All notifications generated by sensors, with the corresponding images, can be automatically sent to a control center provided with Aitek’s AiVu-VMS video management portal or to third-party supervisory systems, using the AiVu format or Onvif specifications. The control center automatically receives alarm notifications from video cameras any time unpredicted events occur, allowing the security personnel to view the recordings associated with the event and trigger the appropriate response procedures.
Deep learning: the evolution of video analytics
Aitek’s video analytics have evolved using deep learning, the artificial intelligence technology for implementing algorithms capable of learning from experience by analyzing data extracted from processed images. Deep learning is based on training highly-sophisticated neural networks to achieve extremely high reliability in processing images and video streams.
Deep learning provides numerous advantages. Using neural networks provides robustness to rapid variations in the monitored scene such as changes in weather, lighting or video camera orientation, all of which may easily interfere with earlier video analytics systems. Furthermore, algorithms can process in real time every single image with no need to employ a reference model of the scene and can detect with extreme reliability even partially overlapping or occluded objects or objects which have long been abandoned.
This is why neural networks can be trained to perform a wide variety of tasks, such as object detection based on object type or shape (vehicles, pedestrians, animals as well as everyday objects such as shopping carts, luggage, etc.), face detection and recognition, object state detection (open/closed door, raised/lowered barrier, standing/lying person, etc.), smoke fire detection, and much more.