In Mobile Robotics, localization is a primordial task, as it makes possible the navigation of the robot, thus enabling it to carry out its activities. From the emergence of the Internet of Things (IoT), there was a different approach of interacting objects with each other, as well as between objects and humans. Based on the context presented, this article evidences the utilization of IoT for the development of a system aimed for localization mobile robots employing Convolutional Neural Networks (CNN)in the process of feature extraction of the images, according to the concept of Transfer Learning. The mechanism uses the topological mapping method to orient themselves in the exploration environment considered. The effectiveness of the approach is demonstrated by parameters such as Accuracy, F1-score and time of processing. The IoT system confers the centralization of processing, reducing costs and allowing reuse of the robot's idle computing power. Combined with this benefit, CNN still achieves 100% Accuracy and F1-Score, proving to be an effective technique for the required activity. With this, the proposed approach demonstrates to be efficient for the use in the task of locating mobile robots. © 2019 Elsevier B.V.