Traffic Sign Recognition Using Distributed Ensemble Learning

Traffic sign recognition is one of the active areas of research in recent years. The automotive technology is moving towards automation in most of the aspects including traffic sign recognition. In an attempt to focus on driving and concentrate on road the driver often misses out the traffic signs, results of which may lead to catastrophic events. This can be avoided by automating the tasks of traffic sign detection and recognition. In this paper, we implement the traffic signs recognition through distributed ensemble technique (DEL), which is an efficient method to automate traffic sign detection. The primary goal of distributed ensemble learning is to decrease the complexity, reduce the training load on each model and improve the convergence. The impact of load distribution with respect to the number of workers has been studied and thereby understanding the trends of a distributed ensemble. Here we use an ensemble of CNN models to train with standard German data set. Keras is used for implementation of distributed ensemble in CNN. Detailed analysis on data distribution between workers and how it impacts the model accuracy is discussed.