Embedded Based Monitoring System Through DNNs

Abstract

In a world where machine replaces human through the Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL). We must keep all things in mind that these technologies work fine on all sort of platform and gives us benefit one way or an another. Our project is also using these technologies to perform a monitorin' system in which we are applying the pre-trained models to get the desired result on the embedded devices but in such a way that it can be implemented in real-time without degrading the performance of the Deep Neural Network (DNNs) on the small devices like embedded systems. DNNs are extremely computationally intensive which result hanging in resource-constrained device like embedded systems, so we cannot get into the field of real-time and make some application out of it. As we are talking about the real time, so we need our DNN to be optimized that its takes millisecond to compute prediction on every frame and give us results on the run time that is nearly impossible on the embedded system due to DNNs structure. Of course, DNNs will give us the prediction on the frame but that results in the frame degradation in the real-time which will be of no use as we need the result on the run time or real time. In our Project we described such a system that it will give us the desired output without degrading our model performance and real-time speed.