Deep learning algorithms for building opening detection from UAV images
Programming skill is also mandatory (preferably Python), experience with deep learning frameworks is highly preferred (PyTorch, Keras and Tensorflow)
Object detection is a fundamental topic in computer vision. Recently, deep learning has become an effective method and achieved great success in object detection. In this topic, we will focus on earthquake scenes and try to detect the openings of damaged buildings that are visible from drone images. An opening can be a window, a door, or an internal structure exposed by the collapse of a wall. The drone can enter the building through the detected opening and explore the interior of the building to search and rescue trapped people. The aim of this MSc topic is to design and implement an efficient end-to-end trainable deep network for opening detection using drone images as input. The combined use of object detection/tracking and monocular depth estimation algorithms will be considered to localise the opening and guide the movements of the drone in real time.
The student will initially revise the existing literature on single stage and two-stage detectors. The proposed network will iteratively predict the bounding boxes of openings in the drone images and will determine its position using a monocular depth estimation algorithm. Some datasets for understanding façades such as CMP-base, could be initially used to train the network.