Ultra-precise road marking vertices extraction using UAV imagery and deep learning algorithms.
Programming skill is also mandatory (preferably Python), experience with deep learning frameworks is highly preferred (PyTorch, Keras and Tensorflow)
When infrastructure companies refurbish asphalt, a road-marking inventory needs to be created in advance to allow for accurate road marking replacement. Because the measurements need to be very precise (< 5 cm accuracy), infrastructure companies in The Netherlands usually need to manually place a several GCP’s on the ground, which is time-consuming and expensive. On the other hand, UAVs allow to acquire high-resolution images useful to generate accurate orthophotos. Therefore, the aim of this project is to develop an algorithm that can precisely measure in road markings from UAV-imagery with only few GCP’s in a completely automated way, delivering their positions and allowing their periodic re-location after maintenance.
The student should start with a literature review, investigating the state-of-the-art methodologies for infrastructure inventory generation. The student will then develop an algorithm for road marking extraction from images. In particular, different semantic segmentation algorithms (adopting deep learning approaches) will be considered with this aim in order to reliably detect markings and exclude similar objects in the scene. In addition, as the algorithm is supposed to run on-board of the UAV, the student should consider real-time processing on on-board processing unit (i.e. NVIDIA Jetson NX).