A Lightweight Convolutional Neural Network (CNN) Architecture for Traffic Sign Recognition in Urban Road Networks

Abstract

Recognizing and classifying traffic signs is a challenging task that can significantly improve road safety. Deep neural networks have achieved impressive results in various applications, including object identification and automatic recognition of traffic signs. These deep neural network-based traffic sign recognition systems may have limitations in practical applications due to their computational requirements and resource consumption. To address this issue, this paper presents a lightweight neural network for traffic sign recognition that achieves high accuracy and precision with fewer trainable parameters. The proposed model is trained on the German Traffic Sign Recognition Benchmark (GTSRB) and Belgium Traffic Sign (BelgiumTS) datasets. Experimental results demonstrate that the proposed model has achieved 98.41% and 92.06% accuracy on GTSRB and BelgiumTS datasets, respectively, outperforming several state-of-the-art models such as GoogleNet, AlexNet, VGG16, VGG19, MobileNetv2, and ResNetv2. Furthermore, the proposed model outperformed these methods by margins ranging from 0.1 to 4.20 percentage point on the GTSRB dataset and by margins ranging from 9.33 to 33.18 percentage point on the BelgiumTS dataset.

RF-CNNS: Thin Deep Learning Networks For Accelerating Traffic Signs Recognition

“…While a higher accuracy of 98.37% is achieved, by the Enhanced LeNet-5 which uses just 0.38 million parameters. In the same context, we find that the CNN adopted by Khan et al (2023) achieves an accuracy of 92,06% while using a higher number of parameters (2.63 million).…”

“…This gating network with switching system, for multi-experts' recognition, achieves a high accuracy of 99.10% using the GTSRB dataset. Khan et al (2023), adopt also an approach based on CNNs. A preprocessing stage is applied, including rescaling (100100), data normalization, and augmentation.…”

“…In fact, these parameters exceed in many cases millions of parameters. For state-of-the-art methods, we find for example that, DenseNet121 includes almost 8 million parameters (Lim et al, 2023b), ResNet-50 25 million, AlexNet 61 million, ResNetv 40 million, VGG16 134 million and VGG19 138 million (Khan et al, 2023).…”

“…From the presented works, we find that hardware optimization is generally used to accelerate the recognition process because the complexity and the size of DL networks affect considerably the processing time and memory requirements (Khan et al, 2023). For that, real-time recognition, using constrained resource environments, still presents a huge challenge for DL based approaches (Lim et al, 2023a).…”

Exploring Explainable Artificial Intelligence Techniques for Interpretable Neural Networks in Traffic Sign Recognition Systems

“…They used the focus loss to control the network of regional proposals and the use of three convolutional and one fully connected layers for the detection of road signs. The authors of [17] presented a lightweight CNN architecture for TSR and achieved accuracy rates of 98.41% and 92.06% on the GTSRB and BelgiumTS datasets, respectively. Similarly, in [18], an attention-based deep CNN is presented and and achieved a testing accuracy rate of 98.56% on the GTSRB dataset.…”