The models are the artificial neural networks, trained to solve a certain type of object detection or classification. The standard pre-trained models are listed here.

• Audio recognition - The module classifies audio information
• Common objects detector - Common objects detection and classification
• Face detector - This model detects all faces in the frame
• Face detector standard - A simplified version of face detector
• Face recognition - Face recognition and matching model
• Age gender emotions recognition - Age, gender, attitude and emotions recognition
• Fire detector - The model detects open fire
• General objects recognition - This model is trained to recognize several hundreds of general objects types
• License plates detector - This model detects all license plates in the frame
• License plates recognition - This model converts images of the license plates into symbols
• Text detector - This model detects all text areas in the frame
• OCR - OCR model saves recognized text to object's properties under the 'textbox' key
• People recognition - The model detects people
• QR-Codes and Barcodes - This model detects and recognizes all QR codes and Barcodes in the frame
• Vehicle color model make recognition - The model recognizes 3000 car models and 14 colors
• Vehicle detector - This model detects vehicles and classifies them into 5 categories

Audio recognition

The module classifies audio information.

Common objects detector

This is a neural classification model for detection of a wide range of common objects.

Face detector

This model detects all faces in the frame.

Face detector standard

This is a light version of face detector. It scans each incoming frame and saves information about all detected faces in the frame meta-data. In order to reduce HW resources consumption,you can adjust the parameter resize_factor.

The face detector allows reducing the detection zone to a certain area. On one hand this will keep the processor from detecting faces in unwanted areas, and on the other hand - it also reduces the load on the computing hardware.

Face recognition

The face recognition neural module extracts facial feature that can be used to search for particular persons and to do real-time matching against list of known person’s faces. The photos of known people need to be saved in the lists. In this case the system will be comparing the faces, detected in the frames, and, if matched, will add additional properties to the object as well as generate a face_match event. This event can be used to generate pop-up notifications for an operator.

This module does not work with the frame. It works with the list of face objects that need to be already detected by a face detector. If frame's meta-data does not contain information about detected faces, the recognition model will skip such a frame.

Age gender emotions recognition

This neural module analyses facial features to estimate person's gender, age, attitude and emotions. The module doesn't require any special configuration.

Fire detector

The model detects open fire.

General objects recognition

This model is trained to recognize several hundreds of general objects type.

License plates detector

This model detects all license plates in the frame.

License plates recognition

This model converts images of the license plates into symbols.

Text detector

This model detects all text areas in the frame.

OCR

OCR model saves recognized text to object's properties under the 'textbox' key.

People recognition

This module detects people in frames by their shapes. It is designed to work effectively at low-light conditions, with blurry objects, with partly occluded peoples' figures. The neural network distinguishes people from animals and other moving objects.

The typical use cases are pedestrian traffic counter or advanced CCTV analytics

QR-Codes and Barcodes

This model detects and recognizes all QR codes and Barcodes in the frame.

Vehicle color model make recognition

The model recognizes 3000 car models and 14 colors.

Vehicle detector

This model detects vehicles and classifies them into 5 categories: bicycle, motorbike, car, bus and truck. The model works effectively during both daytime and nighttime. It allows detecting blurred objects and partly occluded objects.

The typical use cases for this recognition model are car counting software or vehicle type recognition