Automatic Identification and Vectorization of Traffic Infrastructure Features from Orthophoto Images

Orthophoto imaging of the Earth's surface using unmanned aerial systems have in recent years become a common and efficient method for acquiring highly detailed orthophoto maps. These are widely used in transportation and civil engineering fields. In the context of traffic accidents and technical documentation, such imagery can be applied for accurate reconstruction of the scene. However, this process often requires manual vectorization of selected road infrastructure features. This task is time-consuming and demanding, especially in more complex scenarios. The presented paper introduces a newly proposed method for semi-automatic vectorization of road infrastructure features from raster imagery. The method was implemented in MATLAB and consists of several sequential steps. These include selection of the area of interest, colour identification, noise reduction, clustering, and generation of vector contours. The entire process emphasizes simplicity, computational efficiency, and ease of use without the need for machine learning or extensive training data. Statistical evaluation using a paired t-test (p = 0.0022) confirmed that the automated approach is significantly faster than manual processing. On average, the proposed semi-automatic vectorization process was 2.15 times faster. In realistic scenarios, such as entire intersection areas, a speed increase of up to 3.1 times was achieved. These results confirm the practical benefit of the proposed method for efficient and rapid processing of traffic infrastructure image documentation.