A Novel Method for Traffic Sign Recognition Based on DCGAN and MLP With PILAE Algorithm

Abstract

This paper centers on a novel method for traffic sign recognition (TSR). The method comprises of two major steps: 1) make strong representations for TSR images, by extraction deep features with the deep convolutional generative adversarial networks (DCGANs) and 2) classifier defined by multilayer perceptron (MLP) neural networks trained with a pseudoinverse learning autoencoder (PILAE) algorithm. The PILAE training process is considered efficient in which it does not require the number of hidden layers specified nor does it need the setting of the learning control parameters. This results in the PILAE classifier attaining a better performance in terms of both accuracy and efficiency. Empirical results from the German TSR (GTSRB) and Belgium traffic sign classification (BTSC) have proved that TSR achieves excellent results with other algorithms and reasonably low complexity. INDEX TERMSDeep convolutional generative adversarial networks (DCGAN), feature extraction, pseudoinverse learning autoencoder (PILAE), traffic sign recognition (TSR).

Robust Semi-Supervised Traffic Sign Recognition via Self-Training and Weakly-Supervised Learning

“…In conducting the experiments, the evaluation metrics employed to evaluate the model were classification accuracy, F1-score, precision, and recall. The classification accuracy was the commonly used evaluation metric to evaluate most of the classification algorithms mentioned in the literature [1,16,30,31]. In simple terms, precision denotes the percentage of relevant samples among the retrieved samples and recall the percentage of relevant samples retrieved over the total relevant samples.…”

“…The result in Table 3 provides a comparison of the classification accuracy between ROSST, the proposed method, and some state-of-the-art supervised learning algorithms, which include single CNN with three STNs [29], DCGAN-PILAE [30], traffic sign classification based on pLSA [16], multiscale CNNs [31], BAGAN [32], traffic sign recognition with hinge loss CNNs (HLSGD) [49], and residual blocks CNN [50]. All these state-of-the-art methods were evaluated on the GTSRB dataset, making it fair to compare the proposed method ROSST with them.…”

“…A recognition accuracy of 99.71% was obtained for precision, recall, and F1-score for the GTSRB and the BTSC; 98.95%, 98.87%, and 98.86% were obtained for the precision, recall, and F1-score, respectively. The recognition rate was improved by the study conducted by Mahmoud et al [30]. Mahmoud et al combined features learned by deep convolutional generative adversarial networks (DCGAN) and pseudoinverse learning autoencoder (PILAE) supplemented with the softmax classifier method to obtain excellent performance with a recognition rate of 99.80% on the GTSRB and 99.72% on the BTSC, as compared to handcrafted features and other methods that were DNN based.…”

“…Furthermore, Sermanet et al applied the convolutional network architecture to achieve a better result on the GTSRB dataset after experimenting with the Energy-based learning (EBLearn) open-source library [31]. However, their result of 99.17% was later improved by the work of Mahmoud et al [30]. Another study proposed Balancing GAN (BAGAN) [32] as an augmentation tool to restore balance in imbalanced datasets.…”

Multi-Feature Fusion and Enhancement Single Shot Detector for Traffic Sign Recognition

“…The faster region-based CNN (Faster R-CNN) [18] is a representative two-stage target detection framework that has become a popular object detection framework, but it still has difficulty detecting small objects. In recent years, some new methods have been proposed to identify traffic signs [19]- [21].…”