Pavement pit and pond detection method based on machine vision in complex environment

The invention provides a pavement pit and pond detection method based on machine vision in a complex environment. The method comprises the following steps: firstly, training a semantic segmentation model by utilizing a traffic monitoring image subjected to pixel-level labeling; secondly, carrying out background modeling on the traffic monitoring video to obtain a traffic monitoring background image, utilizing a semantic segmentation model to segment the background image, and extracting a road in the image; then, carrying out binarization on the extracted road, segmenting areas with deep colorsand large areas on the road surface out, adopting a support vector machine to classify the areas, and obtaining candidate areas of the road surface pit pond; and finally, outputting candidate regionswhich are divided into the pavement pit and pond sub-regions by the semantic segmentation model. According to the invention, the interference of a complex background on a detection task is effectively reduced, the robustness