This paper presents a multi-column deep neural network (MCDNN) approach that achieved a human-like recognition rate of 99.46% in the German traffic sign recognition benchmark (GTSRB). The method uses a deep neural network (DNN) that does not require pre-wired feature extractors, instead learning features in a supervised manner. By combining multiple DNNs trained on different preprocessed data into a MCDNN, the system becomes more robust to variations in contrast and illumination. The DNN consists of convolutional and max-pooling layers, which are inspired by the structure of the mammalian visual cortex. Each DNN is trained on raw pixel intensities, and the MCDNN combines the outputs of several DNNs, each trained on different preprocessing methods. This approach allows for better generalization and robustness, as demonstrated by the high recognition rate achieved in the GTSRB benchmark. The MCDNN was implemented on GPUs, allowing for efficient training and processing. The system was tested on a dataset of 39209 training images and 12630 test images. The images were preprocessed using various normalization techniques to enhance contrast and improve classification performance. The results showed that the MCDNN outperformed other methods, achieving a recognition rate of 99.46%, which is better than human performance (98.84%) and significantly better than the second best competing algorithm (98.31%). The MCDNN's success was due to its ability to handle variations in image preprocessing and its robustness to different types of noise. The system was able to reject only 1% of the images with low confidence, resulting in an even lower error rate of 0.24%. The results demonstrate the effectiveness of the MCDNN approach in traffic sign classification and highlight the potential of deep neural networks in image recognition tasks.This paper presents a multi-column deep neural network (MCDNN) approach that achieved a human-like recognition rate of 99.46% in the German traffic sign recognition benchmark (GTSRB). The method uses a deep neural network (DNN) that does not require pre-wired feature extractors, instead learning features in a supervised manner. By combining multiple DNNs trained on different preprocessed data into a MCDNN, the system becomes more robust to variations in contrast and illumination. The DNN consists of convolutional and max-pooling layers, which are inspired by the structure of the mammalian visual cortex. Each DNN is trained on raw pixel intensities, and the MCDNN combines the outputs of several DNNs, each trained on different preprocessing methods. This approach allows for better generalization and robustness, as demonstrated by the high recognition rate achieved in the GTSRB benchmark. The MCDNN was implemented on GPUs, allowing for efficient training and processing. The system was tested on a dataset of 39209 training images and 12630 test images. The images were preprocessed using various normalization techniques to enhance contrast and improve classification performance. The results showed that the MCDNN outperformed other methods, achieving a recognition rate of 99.46%, which is better than human performance (98.84%) and significantly better than the second best competing algorithm (98.31%). The MCDNN's success was due to its ability to handle variations in image preprocessing and its robustness to different types of noise. The system was able to reject only 1% of the images with low confidence, resulting in an even lower error rate of 0.24%. The results demonstrate the effectiveness of the MCDNN approach in traffic sign classification and highlight the potential of deep neural networks in image recognition tasks.