Employing support vector machines in microcontrollers and the real-time performance

It is difficult for support vector machines to be employed in microcontrollers because of limited hardware resources. A framework based on cloud computing conception is presented for solving this problem. In this framework, the support vector machines are not trained in the microcontrollers, but trained in the SaaSproviders. Microcontrollers gain or update models from SaaS providers and store them in the ROM, and thenput them in use. The time complexity of using support vector machines in microcontrollers is analyzed. Experiments have been performed to verify the practicability of the framework. The run time of the experiments shows that models with a dozen of features and dozens of support vectors running in 32-bit microcontrollers with dozens of MHz could meet the realtime need for most mechatronics systems. Meanwhile, the result of complexity analysis is confirmed.