Date: 19 November 2019
Venue: School VI
Automatic detection of animals from aerial photography
Aerial photography is increasing available to ecological researchers from a variety of platforms such as drones, light aircraft and satellites. This generates large amounts of data which is usually examined manually by researchers to detect and locate the animals. This is a time consuming process and often a bottleneck in the pipeline that prevents these data sources being used more widely. Modern computer vision techniques hold out the hope that this could now be automated. I will discuss a dataset that I have created to test algorithms for detection of animals in aerial imagery. I have applied classical computer vision and machine learning techniques to the dataset as well as convolutional neural net approaches. The results of these techniques on the dataset will be presented. The results are now approaching human accuracy for recall, now the focus is on reducing the number of false positives. The location of the animals detected can then be used to model the spatial distribution of the animals over large areas of land.
Christina Fell is in the final year of her PhD in the School of Mathematics and Statistics at St Andrews. She is jointly supervised by Monique Mackenzie from the School of Mathematics and Statistics and David Harris-Birtill from the school of Computer Science.