Video Tracker
The module detects faces on an image (in a video stream), performs multiobject object tracking and filtering of a series of images by quality with sending the best of them to an image processing server.
Watch a video about Cyber Data Control features
"You lose clients and money every day due to employees' violations of rules, regulations, and even theft"
Cyber Data Control is a neural network that sees violations and provides you with conclusions for making management decisions
How it Works
The CDC neural network automatically analyzes the video stream from cameras, selects events, and provides you with facts in a convenient way
Industry solution
System interface
Evidence
Due to the accuracy of event detection and minimal false "noise" in the data provided by the system, you can view violations from 5 hours of video in 1 minute. It is convenient to extract any types of events for all controlled objects.
Reports
CDC reports are clear and useful to the security Manager, Department managers, marketer, Director, and owner. Reports can be accessed via secure channels from anywhere in the world. The system promptly signals critical events.
OCR and video analytics
Document detection and recognition
- Detecting documents in the image (selecting an area) Field recognition and text reading (translation of typewritten text images into text data)
- Identification of faces in the image
- Face Image Personality Verification (KYC) Support of joint work of several algorithms on processing of the received image.
Detection and classification of objects
- Control presences/ lack of people/ motor vehicles under surveillance
- Classification of vehicles to determine their categories and legality of being on the roadway
- Detection of different classes of objects in a video stream: human, airplane, dog, car, train, bird, motorcycle, boat, horse, bike, coach, cow
Detection and recognition of fractures
- The technology allows detecting the state registration marks in the frame and recognizing symbols (letters and numbers).
- Working under real conditions: dusty rooms, angled view, poor light, best picture selection
- Quick learning for different fracture formats
Traffic rules violation
The technology allows you to build an accurate trajectory of all moving objects in the frame and vectorize it. Pedestrians and vehicles trajectories are separate classes.
This makes it possible to automatically detect violations such as turning in a wrong place, reversing, obstructing a pedestrian at a pedestrian crossing, etc.
Calculating pedestrian traffic
The solution allows:
- Get the exact distribution of pedestrian traffic for the selected period (day/week/month) taking into account the direction of movement
- Learn about the social demographics of clients Get real data fast (requires only video recording, no infrastructure investment).
- Evaluate the potential revenue and optimal time of the outlet in a particular place even before its opening
Counting the number of people in the hall
- occupancy tracking
- determining queue length, number of customers in the trading floor
- graphical and tabular reporting
- video frame count zone setting
- threshold alarm
Face Detection
Processing Module pictures
Image vectorization, recognition (identification and verification, gender and age classification) Convolution NN (Cellular Neural Networks) + FRIS classifier (a classifier based on the function of competitive similarity)
Bioidentification
Face
Qualitative snapshot, setting of algorithm accuracy thresholds. Additionally - control of liveness (object alive).
Voice
Verification by voice + check of the set sequence of numbers (output on the screen, send to the phone).
Portrait segmentation
Portrait segmentation technology is essential for intelligent photo editing, such as creating a "bokeh effect" in portraits or replacing the background.
Characteristics:
- neural network - "end-to-end"
- accuracy = 98.4% (IoU)
- model size = 24 Mb
- processing time = 180 ms on a Snapdragon 835 processor (1 core)
The traditional binary segmentation approach does not allow for high-quality segmentation with fine image detail, resulting in edge effects and causing customer dissatisfaction.
Technology Expasoft with alpha-matting based on a single neural network (end-to-end), which takes into account the transparency of the hair, features of the background and allows you to create effects of excellent quality.
.