Object detection
Deep learning based object detection technology Object detection of 95% or more using AI technology
LPR(License Plate Recognition)
AIBIS deep learning-based license plate recognition technology Recognizes small-sized license plate letters from a single camera's long-distance image and supports 99% or more high number recognition rate even in degraded image quality due to image compression
- Over 99% Plate Recognition Rate For Identifiable Size or even Distorted Characters
- Real Time License Plate Recognition of Over 15 Frames per Second, regardless of Camera type
- Acquired 100.0% performance certification from Korea Road Traffic Authority (KOROAD) in a wide angle of Surveillance Camera
-
Character Candidate Detection(CCD:Character Candidate Detection)
-
Character separation(CS : Character Segmentation)
-
Character recognition(CR : Character Recognition)
-
Object Recognition/Multi-Attribute Classification
Attribute based object classification Based on AI algorithm, quickly and accurately identify attribute values for detected objects
Object
Recognition
CNN
Classification
by Attribute
- Direction
- Face
- Gender
- Female
- Hair
- Short hair
- Top shape
- Long sleeve
- Top color
- Pink
- Bottom shape
- Long pants
- Bottom color
- Black
- Shoe production
- Black
- Bag
- UseX
- Glasses
- UseX
- Pedestrian property classification
- Type of top (long sleeve, short sleeve, sleeveless, long padded)
- Bottom type (long pants, shorts, long skirts, short skirts)
- 12 clothing colors
- Hair type (short hair, long hair, Bobbed Hair, bald head)
- Hood type (cap, helmet)
- Accessories such as bags, shoes, and glasses
- Vehicle property classification
- Classification of 5 vehicle types (passenger car, van, bus, truck, taxi)
- 7 vehicle colors
Recognize events and behavioral patterns
Based on various environmental learning data Recognize events and behavior patterns with high accuracy through building learning data in various environments
- Pedestrians: Trespassing, Loitering, Fight, Falling, Fire, Crowding
- Vehicles: Driving in Reverse Lane, lane violation, center line violation
Object tracking
Attribution and Re-ID technology-based tracking algorithm It is possible to track the trajectory by applying multiple search terms to the image-analyzed object and event information.
- State-of-the-art technology to find the same object (person, vehicle) in different cameras
- Re-Identification AI model of person attribute vector in cropped image
- FairMOT backbone-based model development
-
CCTV
-
Image acquisition
-
Detection Model :
Darknet-19
Object detection network model
-
-
Person detection
-
Cropping people images
-
Feature Extraction
Model : Darknet-53
Human Recognition Network Model
-
Detected object tracking Multiple search word-based search and result analysis
Car
Pedestrian
Two-wheeled
Vehicles
Person (black)
Bus
