Deep Learning-Driven CNN Approach for Accurate Traffic Sign Recognition in Intelligent Transportation Systems

Аннотация

Accurate and robust traffic sign detection is crucial for the development of advanced driver-assistance systems (ADAS) and autonomous vehicles. This paper presents a review of recent advancements in intelligent traffic sign identification using Convolutional Neural Networks (CNNs). The article surveys various CNN-based architectures, methodologies, and optimizations employed to address the challenges of traffic sign detection, including variations in illumination, weather conditions, and sign degradation. The performance and limitations of current approaches, along with potential future research directions, are discussed. Keywords: Convolutional Neural Network, Traffic Sign Recognition, Intelligent Transportation Systems, Deep Learning, Image Classification, Road Safety, Autonomous Vehicles, Computer Vision, Real-Time Detection, Smart Mobility. Introduction: Traffic signs play a vital role in conveying TSDR systems. essential information to drivers, ensuring road safety, METHODS and regulating traffic flow. However, traditional This review focuses on studies published in recent years methods of traffic sign recognition rely on human (2023-2025) that utilize CNNs for traffic sign detection. perception, which can be fallible due to factors such as The research methodology involved a systematic driver fatigue, distraction, or poor visibility. To mitigate search of academic databases, including IEEE Xplore, these limitations, there has been a growing interest in ScienceDirect, and Google Scholar, using keywords developing automated systems for traffic sign such as "traffic sign detection," "convolutional neural detection and recognition (TSDR). networks," "CNN," "deep learning," and "autonomous In recent years, Convolutional Neural Networks (CNNs) driving." have emerged as a powerful tool for image recognition 1. Dataset Selection and Preprocessing and object detection, demonstrating remarkable success in various computer vision tasks. Their ability to The performance of a deep learning model is highly automatically learn hierarchical features from raw dependent on the quality and diversity of the dataset image data makes them particularly well-suited for the used for training and testing. For this study, we utilized task of traffic sign detection. CNN-based TSDR systems the German Traffic Sign Recognition Benchmark can analyze images captured by onboard cameras to (GTSRB) dataset, which is one of the most widely used accurately detect and classify traffic signs in real-time, datasets for traffic sign recognition. The dataset even under challenging environmental conditions. consists of over 50,000 labeled images, representing more than 40 different classes of traffic signs. These This article provides a comprehensive review of recent images were collected under varying conditions, research on intelligent traffic sign identification using including different lighting, angles, and levels of CNNs. It examines the various CNN architectures, occlusion, making the dataset highly representative of methodologies, and optimizations proposed to real-world driving conditions. enhance the accuracy, robustness, and efficiency of American Journal of Applied Science and Technology 1 https://theusajournals.com/index.php/ajast American Journal of Applied Science and Technology (ISSN: 2771-2745) Before training the CNN model, the images were making predictions about the traffic sign class. preprocessed to ensure that the system could learn • Softmax Output Layer: The final output is from high-quality, normalized data. The preprocessing produced by the softmax layer, which assigns a steps included: probability score to each class. The class with the • Resizing: All images were resized to 64x64 highest probability is selected as the predicted label. pixels to ensure consistency across the dataset. 3. Model Training and Optimization • Normalization: Pixel values were normalized to To train the CNN model, we used the Adam optimizer, a range of [0, 1] to standardize the input data. which adapts the learning rate based on the gradients, • Data Augmentation: To enhance the model’s and the categorical cross-entropy loss function to ability to generalize, we applied various augmentation calculate the error between the predicted and actual techniques, such as random rotation, flipping, and class labels. The model was trained over 50 epochs, color adjustment, which helped simulate different with a batch size of 32 images per update. Early weather conditions, times of day, and other factors. stopping was used to prevent overfitting, with the model being evaluated after each epoch to monitor its 2. CNN Architecture performance on the validation set. The deep learning model was built using a The model's learning rate was initially set to 0.001, with Convolutional Neural Network (CNN) architecture adjustments made through a scheduler to improve designed to automatically extract relevant features training efficiency. The performance was tracked