Shadows in drone images commonly appear in various shapes, sizes, and brightness levels, as the images capture a wide view of scenery under many conditions, such as varied flying height and weather. This property of drone images leads to a major problem when it comes to detecting shadow and causes the presence of noise in the predicted shadow mask. The purpose of this study is to improve shadow detection results by implementing post-processing methods related to automatic thresholding and binary mask refinement. The aim is to discuss how the selected automatic thresholding and two methods of binary mask refinement perform to increase the efficiency and accuracy of shadow detection. The selected automatic thresholding method is Otsu’s thresholding, and methods for binary mask refinement are morphological operation and dense CRF. The study shows that the proposed methods achieve an acceptable accuracy of 96.43%.

This study shows a preliminary investigation of shadow detection in drone-acquired images using a deep learning method with minimal labelled shadow images. The aim is to discuss how the selected U-Net architecture performs in a small-sized dataset consisting of various types of shadow brightness and objects. Two types of data augmentation methods, which are shadow variant and geometric transformation are implemented, aiming to improve the segmentation accuracy. Several experimental procedures are performed to observe the model performance. The study shows that adding images for training increases the accuracy of shadow detection in drone images from 0.95 to 0.96, and geometric transformation data augmentation method increases the accuracy from 0.961 to 0.963, while the shadow variant method increases the flexibility of detection.

This study evaluates the performance of shadow detection using different image color models. The pixel-level supervised classification procedure employed in this study includes filtering images, creating a trained shadow model, obtaining shadow masks and post-processing of the output masks. Considering the advent of drones usage, we discuss the results of shadow detection on aerial images. Based on the results, the method using YCbCr color features yielded 92.71% average accuracy. The low performance of shadow detection on images with small shadowed regions and images under various weather conditions indicated that additional investigation is necessary to create detection schemes for challenging input images with high spatial resolution.