Customized Object Proposal Algorithm Based Traffic Sign Detection

Convolutional neural networks help in extracting information from the camera images. In the present work, traffic signs were recognized by a deep neural network architecture from a camera image. The customized object proposal method is proposed for localizing the region of interest. The number of regions generated by the proposed method is five times fewer than selective search segmentation, this makes the method faster as compared to selective search segmentation as the time spent on false regions proposed by selective search is saved. The proposed method is computationally efficient and simple as compared to modern faster regional convolutional neural networks, where Convolutional layers (RPN) are embedded in the network to determine the region of interest. The proposed system can be utilized for developing a smart driver assistant system and self-driving cars.