Traffic-Sign Recognition Using Mask R-CNN
No Thumbnail Available
Due to the rapid developments occurring in the world, especially in the field of social media, we started to witness great impacts upon human’s cognitive systems. For example, concepts like short-term memories, distractions and forgetfulness became very prevalent. This study stemmed from these changes. We have developed a traffic-sign detection software model - usable in vehicles’ systems. Its main objective to help drivers remember traffic laws and aid them to focus while driving. The software is developed using Keras, which is a high-level deep learning API, and TensorFlow, an end-to-end open-source platform for machine learning. We used deep neural network Mask R-CNN as a model, which was trained and evaluated on a Small Traffic sign Dataset containing traffic sign images of three categories which this small data is trained on large dataset called COCO, with 5 epochs each epoch takes about 5 hours in training step. At the end of 5 epochs the proposed model training had a mean average precision (mAP) of 93% and the system was able to recognize input images based on the train model and the output was protected object in each image, we also used libraries of OpenCV for real time detection. Proving the positive outcomes of this model, we are exploring the potential of using it to improve upon driving schools’ systems, and car manufacturing systems. Moreover, we aim to explore a potential linkage between the vehicles that have the model installed in them and some governmental systems. This linkage works to alert officials regarding any violation of traffic laws, and will allow for officials to keep an online database of the vehicles and their behaviors on the roads.