Cognition based Recognition of Partially Occluded Traffic Signs

Computer vision based traffic sign detection and recognition is an active field of research but the task becomes challenging when interest partially occluded by nearby objects like a tree, pole or vehicle. Another difficulty posed especially in developing countries lost colors problem that arises due to aging poor maintenance. This work presents automatic technique focuses on visible parts only suppresses portions. Features are collected using convolutional neural network inspired invariant feature extraction augmented with interaction dimensionality reduction. Further, use dynamic parameter estimation, adaptive system for continuous learning also proposed. Since effect partial occlusion has not been thoroughly studied, there no benchmark database available this purpose. We have prepared two datasets combining originally synthetically images taken from surveys famous GTSRB database. Experiments revealed our outperforms state art methods previously used signs obtaining 0.81 precision 0.79 recall values average. The proposed method shows remarkably low error rate as amount increased.