Object Detection at Edge Using TinyML Models

With the penetration of IoT across sectors, image classification becomes a critical issue if the computations have to be done at the edge. The evolution of low-cost devices with powerful processing for any vision-based systems leads to the next requirement of machine learning for imaging with reduced latency and reliability along with data security. Running the energy hungry computer vision techniques which need frequent memory access may not be the solution. This paper deliberates the works carried out which can be deployed at the end devices, such as mobiles, microcontrollers for image computing at edge. The approach used is by leveraging the TinyML optimized features for low latency and energy efficiency. This paper deals with sample data worked for classification on ML methods, namely, FOMO and MobilenNetSSD which are converted from the Tensorflow to lite version, using the edge Impulse platform. The results discussed are taken from the deployed TinyML model on mobile phone. The outcomes along with the accuracy are also discussed.