Vehicle classification
Recognition and counting of vehicles in automatic mode allow assessing road congestion and analyzing traffic density both along the entire highway as a whole and in individual lanes.
Solution metrics
97%
Recognition accuracy
Accuracy measured on test videos in the daytime
- < 1
Number of false positives per day per camera
≥ 10 FPS
Frames per second
camera requirement
≥ HD
Recommended video stream quality
23
Vehicle types
20 px
Object size requirements in the frame
per vehicle
Technology Description
Vehicle detection and counting is based on a set of neural networks optimized for CPU and GPU operation. A neural network for object detection is the basic element of the solution; most other video analytics are based on it.
Roadly is distinguished by high quality of detection in difficult conditions, a low percentage of false positives, and the ability to train the system as it works (due to the built-in algorithm for collecting target datasets).
Vehicle counting statistics are kept by traffic lanes and vehicle types (it is also possible to record statistics by vehicle make and model).
Roadly is distinguished by high quality of detection in difficult conditions, a low percentage of false positives, and the ability to train the system as it works (due to the built-in algorithm for collecting target datasets).
Vehicle counting statistics are kept by traffic lanes and vehicle types (it is also possible to record statistics by vehicle make and model).
- High accuracy in simple and difficult conditions
- Fast object recognition
Screenshots and demo videos
Features of the technology
High accuracy out of the box
Additional training in the process of work
GPU and CPU optimization
-
Cost Savings for Video Surveillance
Proven technology