ABSTRACT Now a days Security and health care activities are the major issues in Smart homes.Vision based action monitoring has been a serious issue in smart homes.Activity recognition supports in various applications like health care,elderly monitoring,Safety,Social networks analysis,monitoring the environment,transportaion monitoring,surveillance systems etc.Hence the most booming technology ,deep learning is used to monitor the human activity .To perform the above experiment a Convolution Neural Network model is used to recognise the human acivity present in a Smart home.The model is tested and trained with a large data set which has large volume of video data collected .From the experiment a significant output is acheived for the inputs provided which ranges from low level to high level quality of inputs.DML smart actions is a popular data set and it used to test our model.The results obtained from the proposed system was around 82.4% in measuring the accuracy rate of the human activity ecognition in smart homes.