Existing methods have extensively addressed the issue of detecting abnormal events in a smart home environment through wearable sensors in the past years. However, the limitations of wearable sensors include the limited battery power as well as the use and adoption challenges of wearable activities on a daily basis. The use of non-wearable and non-intrusive sensors is necessary for providing better user experiences and achieving a sustainable and reliable detection model. However, it is still very challenging to analyze such non-wearable sensor data with a high level of accuracy. In this paper, we present a continuous deep learning model which receives a set of consecutive images for classifying posture types using a Microsoft Kinect as our non-wearable sensor. We adopt a deep learning technique called the recurrent neural network (RNN) using the long short-term memory (LSTM) architecture to construct our detection model by identifying human postures in fall detection. Furthermore, we investigate the inputs for our model by extracting the features from the pre-processed high-resolution RGB images, including body shape, depth and optical flow. As a result, the body shape with genuine motion and depth information are considered. Finally, we present the experimental results to demonstrate the performance and novelty of our approach.