Types of services
A table with a summary of the accuracy and speed of the services offered by AISP.
| Type of processing | Measurement error** | Processing speed * |
|---|---|---|
| Traffic measurement (basic) | 3.00% | x5 |
| Speed measurement | 5.00% | x5 |
| Pedestrian measurement**** | 10.00% | x5 |
| Traffic measurement - intersections*** | 10.00% | x3 |
| ALPR measurement + Traffic | 10.00% | x3 |
| Parking ALPR (50-60FPS FHD video) | 10.00% | x1-x2 |
| Photo anonymization (for 1000 photos) | n/a |
Notes on the above table and results
- *processing speed calculated for a 20fps video,
- ** measurement error determined on the basis of a test set of 70 films,
- *** the measurement error for intersections is greater due to the large number of intersecting vehicle trajectories,
- *** the measurement error for pedestrians is larger, more intersections of traffic trajectories and greater similarity of objects
Traffic measurement cross-sections/intersections
The basic measurement of road traffic is the counting of vehicles in the cross-section of the road. Historically, the first service offered by our company, derived from General Traffic Measurements.
Speed measurement
For each tracked object in the video, we know its location in each image frame (motion trajectory). Marking a measurement section with a known length in meters allows you to calculate the time of covering the section by the object. Each measurement section consists of two cross-section lines and the real distance [m] between the lines. The obtained results are saved in the result sheet Velocity.csv and in the form of graphs. It is also possible to receive data in the format used by SCPR.
Pedestrian measurement
Pedestrian measurement is defined as a separate measurement category due to the fact that pedestrians are more difficult to count than vehicles. There are many trajectory intersections, obscurations and the similarity of the tracked objects is high. At the moment it is a separate measurement, in the future we will measure pedestrians together with vehicles.
Types of pedestrian categories recognized by AI:
- Pedestrians
- Bikers
- People riding a scooter (TODO),
- Persons driving a baby stroller (TODO),
ALPR measurement + Movement
From 2022, we introduced the license plate recognition service (ALPR). For each counted vehicle we are able to recognize the license plate. Counting is possible from both fixed (stationary) and stationary cameras.
Fixed cameras (e.g. mounted on a pole).
Video requirements for ALPR during the day are the same as for traffic. However, after dusk and at night, the main problem is the sharpness / shutter speed of the image. Research and individual adaptation of the camera are necessary.
Moving cameras (e.g. mounted inside a moving vehicle) .
The requirements for measurements from inside the vehicle (e.g. parking time measurements in zones) are 50-60FPS and the speed of the moving vehicle < 30 km/h.
Anonymize photos and videos
All categories recognized by AI can be anonymized as agreed with the client.