Zařízení pro detekci objektů přechodného dopravního značení

Abstract

This master's thesis deals with the design and implementation of a low-cost device for automatic monitoring of temporary traffic signs. Temporary traffic signs are exposed to the risk of damage or displacement due to traffic accidents or vandalism, while their inspection is currently carried out mainly manually. The device was built on the Raspberry Pi single-board computer platform. Since no available model was capable of detecting temporary traffic sign objects, a custom dataset was created and a YOLO11n model was fine-tuned with a single object class. The model was exported to the NCNN format, optimized for inference on ARM processors. The system operated by comparing the current frame with a reference image and upon detecting an anomaly triggered a series of actions including lens self-cleaning. A protective enclosure designed for outdoor use was modelled and 3D printed, equipped with a self-cleaning mechanism using a transparent film on a spool. A server-side component using MongoDB and Metabase was prepared for data storage and visualization. Two hardware sets differing in performance and camera were assembled and tested, with one set deployed in real traffic conditions.