Pakistani Traffic-sign detection using Deep Learning

Abstract

Traffic-sign Detection is one of the major aspects of the working of a modern car, more so in the case of a self-driving car. They need to be detected and recognized up to a certain degree of accuracy. This research revolved around the detection of Pakistani Traffic-sigs. The research was conducted in 3 phases; firstly, a fixed camera was used to collect video feed from real-world car rides. These videos were then extracted and manually annotated using pertinent software tools. This helped create a dataset of images of traffic signs which was important as the model being used is Deep Learning-based, which require colossal amounts of data to function properly. Secondly, this data is used to train a Deep Learning model to detect and classify the type of traffic sign. The trained model produced a mean average precision (mAP) of 75.636% on the training dataset and 49.699% on the validation dataset while the mAP stood at 43.453% for the test dataset. All these results are state-of-the-art and strong enough for implementation as real-world models. The model was cross-validated and regularized to help improve the model’s working. The final model was tested in real-world scenarios and tweaked according to requirements.