The aim of active and assisted living (AAL) is to develop tools to assist the elderly people in the ageing status. Human posture recognition algorithms can help monitor aged people in home environments. Different types of sensors can be used for such a task. A case in point is the RGBD sensors, which are cost-effective and provide rich information about the environment. This work aims to propose a posture recognition approach exploiting skeleton data extracted from Kinect. Our approach is based on the pose prediction using key joints features. We exploit the Convolution Neural Network for pose estimation and a multiclass Support Vector Machine to perform posture classification. The proposed approach has been tested on a publicly available dataset for activity recognition, namely CAD60. Our approach compares favorably previous works for both human pose estimation and posture recognition