In this paper, we focus on the human activity recognition module of a homecare system that consists of wireless sensors developed for remotely monitoring patients with cognitive disorders, such as Alzheimer. To this end, as an initial study, we designed a smart couch that is equipped with accelerometer, vibration and force resistive sensors to monitor how much time people spend while sitting, lying or napping on the couch and to recognize the drifts from their daily routines. In order to distinguish these activities, features of the data collected from the sensors is first extracted and then different classifiers, such as k-nearest-neighbor and Naïve Bayes, are applied. The impact of sensor type, feature type and classifier type on the performance of classification is experimented with four test subjects and experiment results reveal that k-nearest- neighbor classifier with maximum, minimum, mean and standard deviation features and accelerometer and force resistive sensors exhibit the best performance with 95% F-measure.