Automated Detection for Road Marking Quality, using Visual Based Machine Learning

Most of the cities road marking inspection is performed manually and considered an ideal candidate for automation because it is a labourintensive process. We propose a solution based on automated quality assessment tool for road marking to detect and qualify road marking characteristics. Our data inventory and data collection approach work on images collected from a camera mounted on vehicle or drone. We use an automated procedure to choose images suitable for inspection based on road marks conditions. From the selected data we segment the ground and detect three different parameters; road markings conflicts, missing road markings, and road marking reflectivity. We describe convolutional neural network and image classification algorithms that identifies road marks conflicts, visibility, missing road markings. We also discuss the problem of evaluate the quality of the existing road markings, conserve human and financial resources that is caused by on-site inspections, reduce improper manipulation of road marking (human errors), and prompt jobsite management to focus on promoting fine workmanship of road marking during the execution and train the classifier. We present results from the prototype that shows the detection of the three trained parameters mentioned above. Finally, we show results computed using our method over a subset of a citywide urban and non-urban road network.