Nurses in a hospital are responsible for the monitoring and care of a large number of patients. Regularly checking if patients are sufficiently covered by their blankets is part of a nurse's normal workload. In this paper, an algorithm based on 3D images is proposed that can automatically and unobtrusively detect if a patient is covered with a blanket or not. The depth images for this study was recorded by a Microsoft Kinect™ sensor in a simulated hospital environment. The training dataset consisted of 20 images with blanket and 10 images without blanket. Additionally, held-out test data (20 with and 10 without blanket) was created in both good lighting conditions and under a greater variation of lighting conditions to test performance of trained classifiers. We first extract traditional features (textural features and a statistical measures) and new features based on a cross-section profile of the data from a training data set. The skewness of pixel values over the region of interest and the minimum of the differentiate of smoothed cross-section profile were selected as the most important feature using mutual information. Several classifiers, including support vector machines (SVM), k-nearest neighbors, and logistic regression were built with the selected features. The best performing was linear SVM classifier with overall accuracy of 93% on entire held-out dataset, with sensitivity of 90% and specificity of 85%.