EN

Machine Learning in Security and Traffic Monitoring

Machine learning is also a good way to count vehicles and people. We used the YOLOv5 model, trained with the KITTI dataset. This model accurately classifies and determines the exact location of vehicles (car, truck, bus, bicycle) and people.

The number of traffic participants passing by in front of the office was stored, grouped by type and direction of travel.

Smart camera systems can also be used to count traffic.

For example, we trained our traffic counting camera system to identify different types of vehicles (motor, bus, car, bicycle) and pedestrians passing on the main road in front of the office building, the direction of travel of all of them; and it also registers and counts the traffic, grouped by type and direction, and compiles real-time, hourly statistics from it.

Watch the live stream here

Computer vision is also a solution for finding and registering parking spaces, filtering through traffic, and detecting traffic or parking violations.

Traffic Counting

  • Base model: YOLOv5 (KITTI)
  • Input: 416x416
  • Output: position and recognition percentage of detected object

The model distinguishes between cars, trucks, buses, motorcycles, bikes, and people.

The traffic participants can be counted and grouped by type and direction of travel.

Remote monitoring of gates and other places

  • Base model: YOLOv5, Tiny YOLO
  • Input: 22×224
  • Output: position and recognition percentage of detected object

The model is suitable for monitoring car accesses, parking lots and detecting illegal parking. The Tiny YOLO is a low-resource model that can be used in a browser or on a smartphone.

Human movement analysis

  • Base model: PoseNET, MobileNET, ResNET
  • Input: 640×480
  • Output: 17 points on one body (nose, eyes, ears, shoulders, arms, wrists, feet, knees, ankles)

With the help of machine learning, we can observe human movement on surveillance camera recordings and receive notifications when typical gestures, posture, groupings, or movements indicating fights or vandalism etc. are detected.

Try it! (You'll have to turn on your camera.)