Lemma, Dagmawi (PhD)Gezahegne, Melak2020-10-202023-11-292020-10-202023-11-292020-05-25http://etd.aau.edu.et/handle/123456789/22855Internet of things (IoT) often uses devices that are resource scared having very limited processing power and storage capacity. However, some scenarios demand resource intensive technology such as camera technology that usually requires high computational resource due to the images captured by camera requires an image processing. One of the scenario is on-site solid waste segregation. To provide an alternative solution for such scenario, a framework is developed based on the technology of IoT, Image processing, and Deep Learning. The framework is presented based on four-layered IoT architecture and Faster-RCNN architecture. In the framework, the camera (used to capture real-time visual information) and actuator (to perform physical actuation) are placed at the perception layer. The captured video transmitted to through wired or wireless via transport layer to processing layer. In the processing layer, the real-time object detection facilitated based on Faster-RCNN objected detection architecture and represent the detected object to the lightweight information(title) after successful detection. The title which actually represent the object is communicated through serial communication to the Actuation service provider placed at the application layer. The Actuation service provider running on the low-performance IoT device at application layer and delivers actuation service to physical actuation (perception layer) based on the title it receives from processing layer. The Physical actuation makes a physical action using Actuator IoT device. For the scenario of on-site waste segregation, the actuator is attached to the dustbin hatch and performs opening and closing of the dustbin to realize the segregation based on the command it receives form Actuation service provider. The implementation of the thesis based on Tensor Flow object detection API framework (to develop real-time object detection) and the Arduino Uno IDE(C++) to develop the actuation service provider embedded system. In addition, the IoT device hardware circuit was designed and simulated using Proteus Simulator. During the evaluation, the concept is simulated and experimentally tested based on the scenario of on-site solid waste segregation. We have trained the object detection model to detect the class of plastic and glass objects. The result shows that real-time actuation can be successfully accomplished if the object detection model successfully trained to detect the corresponding object (visual information).enInternet of ThingsImage ProcessingVisual InformationFrameworkThesis