05.10.26 (inclusive)

Payment under the development agreement based on the results of successful pilot tests at the facility

MVP

ICT tasks

Car industry

Intelligent control systems

Software/ IS

The object is the entrance to the parking lot. One 4 MP (2688×1520) IP surveillance camera is mounted above the entrance at an angle to the traffic lane. There is a speed bump in front of the barrier, so the car drives slowly or stops. At night, the camera switches to black and white mode with IR illumination. What prevents recognition at night: — glare from headlights, especially LED ones: auto exposure adjusts to the headlights, the number turns out to be dark or overexposed, there are halos around the headlights; — light spots from the headlights on the asphalt knock down the exposure of the entire frame; — blurring of the number when driving; — matrix noise and low contrast of symbols in IR mode; — the glare of IR illumination from the retroreflective coating of the license plate. What we have already tried: we adjusted the camera settings, changed the IR illumination (partially blocked it, left one backlight instead of two), compared the frames with headlights, without headlights and with dimensions. The quality has improved, but there is no stable recognition — about 87% of driveways are correctly read at night compared to 98% during the day. The numbers are standard Kazakhstani, "NNN ABC NN" format: 3 digits, 3 Latin letters, region code.

Automatic entry without operator intervention at any time of the day: night recognition accuracy is at least 95%, fewer manual barrier openings and queues at the entrance. The ability to install the system on objects with conventional video surveillance cameras without expensive specialized cameras with automatic license plate recognition.

Andrey Guchakov