Detecting Momentary Shadows from Visible and Thermal Image Pair
Kazuya Fujita, Ryo Matsuoka, Takahiro Okabe · 2021 29th European Signal Processing Conference (EUSIPCO) · 2021
Outdoor shadows can be classified into two categories: continuous shadows caused by static objects and momentary shadows caused by moving objects. Since the momentary shadows such as shadows due to a photographer are annoying and do not exist in the original scene, they should be detected and removed for improving image quality. In this paper, we propose a method for detecting momentary shadows from a visible and thermal image pair. The key idea of our proposed method is that the continuous shadows have lower temperature than non-shadow areas, while the momentary shadows have almost the same temperature as the non-shadow areas. Therefore, our method combines the shadow areas detected by using an RGB image and the higher-temperature areas detected by using a thermal image, and then detects the areas of momentary shadows via image segmentation. Through a number of experiments using real visible and thermal image pairs, we show that the combination of visible and thermal images are effective for detecting momentary shadows, and that our method works well for momentary shadows with varying duration time.