| Depending on the method we choose for human action recognition, some algorithms require the localization of human in action videos as a preprocessing step. This preprocessing is more challenging in real environments with noises or varying illumination, and may include background subtraction, shadow removal and noise removal.;Also a new dataset is provided containing several daily actions of people living in a smart condo. The dataset is used in order to evaluate the proposed method. Moreover, it can be useful in action recognition and motion analysis studies with health care applications.;This thesis concerns with shadow removal problem with the goal of extracting human silhouette properly. Detecting shadows can be considered as a labeling problem and can be transformed into an energy minimization problem in a Markov Random Field framework. The advantage of the algorithm is that it considers the neighborhood information as well as chromaticity values of the pixels. |
