Image segmentation of foliage and man-made objects in aerial RGB images

Unmanned Aerial Vehicle (UAV) technology has advanced significantly in recent years. With the aerial data, analysis provides solutions to government agencies and industries, including the energy and forestry sectors, as well as urban/rural planning. Among the many aerial data modalities, RGB images play an important role in the analytics process. Our research problem involves segmenting aerial RGB images composed of natural and man-made objects into two categories in real-time as the data, i.e., video frames, are collected. We will begin with an investigation of a combination of state-of-the-art computer vision and machine learning methods that work in real-time. This includes discovering training features and determining the correct method for training the model to achieve the best results. Another important objective is to develop a strategy which does not rely on a large training dataset. Improving the model’s accuracy and time performance will be the final focus of the project.

Faculty Supervisor:

Irene Cheng

Student:

Jilin Liang

Partner:

AERIUM Analytics

Discipline:

Computer science

Sector:

University:

University of Alberta

Program: