Supervisor: Roman Pflugfelder
Object detection, i.e. the recognition and localization of objects, is very important in aerial imaging and remote sensing. Demanding applications can be found such as urban planning for smart cities, environment monitoring to reduce traffic and pollution.
Object detection in aerial and satellite data is still challenging due to their tiny appearance compared to the size of the images. Classical methods of object detection very often fail in this scenario due to violation of implicit assumptions made such as rich texture, small to moderate ratios between image size and object size. We developed an approach (see illustration above) that performs in two stages: First, a CNN predicts a heatmap, which indicates the likelihood that an object is present at a given image coordinate. Second, vehicles are detected by thresholding (segmenting) the heatmap.
Improve the given approach by including the second segmentation step of the heatmap into the network to enable end-to-end learning.
The thesis can be combined with a preceding Informatik Praktika.
- Review literature
- Create training and validation dataset
- Implement training and test algorithms
- Optional: Improve algorithms for better results
- Basic knowledge in computer vision
- Basic experience in Python
- Interest in Machine Learning, Pytorch, maths, statistics
- Interest in GPU programming
LaLonde, Rodney, Dong Zhang, and Mubarak Shah. “Clusternet: Detecting small objects in large scenes by exploiting spatio-temporal information.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.
Roman Pflugfelder, Axel Weissenfeld and Julian Wagner. On Learning Vehicle Detection in Satellite Video. CVWW, 2020.