A Novel Shadow Detection Method using Fuzzy Rule based Model

Abstract

Computer vision applications such as object classification, human detection, action recognition, gait recognition, etc. are often facing challenges in terms of improper segmentation and tracking due to shadow effect. However, conventional shadow detection algorithms highlight the shadow variant and invariant features. The limitation comes from the fact that many approaches are not applicable for both outdoor and indoor shadows. They fail to detect shadow in different illumination conditions as well as a different geometric position such as ground shadow, vertical shadow, self-shadow, etc. Moreover, the limitation includes shadow detection in video sequence, where different threshold values have been computed for each change of frames due to the dynamic nature of the video sequence. As a result the complexity of the system increases. To overcome the above challenges, this paper proposes a fuzzy rule based model for cast shadow and self-shadow detection using three premises, variant properties such as R-channel spectral ratio from RGB, invariant properties such as a difference in chromaticity color space, and average image intensity. Fuzzy rules are employed using some training data sets. Then, another validation data set is used to check how the application of fuzzy rules reproduces the threshold values for various illumination conditions, as well as different environments (indoor, outdoor) and different texture based background. The proposed framework has been compared with other state-of-the art methods.