One of the most researched fields over the past decade has been Human Action Recognition (HAR). This is because there is no single model that can adequately reflect the wide variety of difficulties presented by the mechanics of the human body. As a result, many techniques, applicable to a wide range of data, have been created. The difficulty in accurately representing data is linked to the recognition difficulty. This issue was gradually fixed by increasing the dimensionality of the data, which led to inexpensive 3D RGB-D data in place of 2D RGB films. The purpose of this paper was to develop a 4-stream convolutional neural network (CNN) for use with RGB and depth data. Each of these 4 CNNs is fed this multi modal data as spatial and motion data. Optical flow maps are used to display the RGB and Depth motion data. The network was created to be completely re-programmable.