Utilizing Capsule Networks in Deep Learning for Traffic Sign Detection
R Giridharan, P Mariush Rufin, Ramya Navaneethan, Chaitra Pranesh, Tammugonda Rahul · 2024
Convolutional neural networks (CNNs) have been the predominant Current Advanced neural network techniques and Methods for Traffic Signal Classification. Despite their capabilities, image recognition models are restricted in their ability to capture aspects like pose, view, and orientation because of the limitations of the max pooling layer. A new approach has been introduced in this paper to identify German traffic signs utilizing capsule networks, which have displayed exceptional performance. Caps nets consist of capsules, which are clusters of neurons that encode object instantiation parameters such as position and alignment. This is achieved through algorithms for dynamic routing and agreement-based routing. In contrast to prior techniques that depended on manual feature extraction and the utilization of several complex neural networks with numerous parameters, our approach minimizes the necessity for manual parameter configuration and bolsters resilience against spatial variations. Moreover, CNNs are susceptible to various adversarial attacks, whereas capsule networks can effectively counter such intrusions, enhancing the dependability of traffic sign detection in autonomous vehicles.