Action detection using deep learning on IoT devices

Implementation of the Internet of Things (IoT) is becoming wide-spread, particularly in smart city applications. Due to the high amounts of raw data gathered by enormous of IoT devices, the Deep Learning (DL) method has become necessary to further develop intelligence and application capabilities. In particular visual action detection is one of the critical components of a smart city. It is challenging to use standard Deep Learning techniques for action detection in IoT devices because Deep Learning applications need high CPU, RAM, and storage. To use the standard DL techniques in IoT devices some of DL models shrinked. In this master's thesis, Deep Learning Lite and Micro techniques were applied on real IoT devices. Comparison of IoT devices and Deep Learning Lite and Micro techniques was performed in terms of parameters such as accuracy, delay, and temperature of the devices, applying these Techniques in IoT devices.