حميدرضا حيدرزاده مقدم

Artificial intelligence and computer vision are among the leading technologies of today, playing a pivotal role in enhancing the accuracy and speed of visual data processing across various industries. In the railway industry, these technologies are utilized in autonomous train systems for obstacle detection, signal recognition, and level crossing eva‎luation. The proper functioning of these systems relies heavily on the quality of input images fed to neural networks, and any degradation in quality can lead to issues such as detection errors, incorrect eva‎luations, and compromised safety. In the railway industry, factors such as high train speeds, track conditions, and weather changes like wind and rain cause severe and random vibrations in cameras, resulting in blurred images entering the neural networks. These conditions necessitate an effective solution to maintain the quality of input images. To address this, the DeblurGAN-v2 network with the MobileNet-DSC architecture, known for its effective performance in mitigating mild to moderate motion blur, was selected as the base network. Considering the complex and random nature of vibrations in the railway industry, targeted modifications were made to the generator architecture of the network. These modifications included increasing the number of convolutional layers to enhance feature extraction, enlarging filter sizes to capture more details, adding Attention layers to focus on critical regions of the image, and optimizing the downsampling and upsampling layers to improve image reconstruction. These adjustments significantly enhanced image quality metrics, such as PSNR and SSIM, while key industrial metrics, including processing time and FLOPS (representing computational operations), remained within industry standards. The proposed algorithm effectively provided high-quality reconstructed images and demonstrated reliable performance in mitigating motion blur, establishing itself as a practical tool for enhancing the safety and precision of computer vision systems in the railway industry.

سيستم‌هاي بينايي كامپيوتر , لرزش دوربين قطار , رفع اغتشاشات لرزشي تصاوير دوربين‌هاي قطار , شبكه‌هاي عصبي كانولوشنال , شبكه DeblurGAN-v2

Computer vision systems , train camera vibration , motion blur reduction in train camera images , convolutional neural networks , DeblurGAN-v2 network

Hamidreza Heydarzadeh Moghadam

Dr. Bahman Ghorbani Vaghei