PROJECT TITLE :

Traffic Sign Recognition by combining global and local features based on semi-supervised classification

ABSTRACT:

The legibility of traffic signs has been studied from the start of the design process to ensure that they are easy to recognize by people. Identifying traffic signs, on the other hand, remains a difficult task for computer systems. Image-processing and machine-learning algorithms are constantly improving in order to solve this challenge more effectively. However, with the growing quantity of traffic signs, labeling a vast amount of training data comes at a considerable expense. As a result, an urgent research priority has been to figure out how to create an efficient and high-quality traffic sign recognition model in an Internet-of-Things-based (IOT-based) transportation system using a small quantity of labelled traffic sign data. In this research, we offer a novel Semi-Supervised Learning strategy for traffic sign recognition in an IOT-based transportation system that combines global and local data. To construct alternative feature spaces, we use histograms of oriented gradient (HOG), color histograms (CH), and edge features (EF). Meanwhile, on unlabeled data, a fusion feature space is discovered to bridge the gap between distinct feature spaces. Extensive tests on a set of signals from the German Traffic Sign Recognition Benchmark (GTSRB) dataset reveal that the suggested method surpasses the competition and might be used in real-world scenarios.