You are currently viewing رسالة ماجستير زهراء يحيى / بعنوان: An Automated Stereo Vision System for Surveillance Dusty and Foggy Weather Conditions Enhancement using Deep Learning

رسالة ماجستير زهراء يحيى / بعنوان: An Automated Stereo Vision System for Surveillance Dusty and Foggy Weather Conditions Enhancement using Deep Learning

Abstract

Recording videos under dusty and foggy conditions presents numerous challenges due to frequently low light levels, reduced brightness and contrast, and increased noise, all of which significantly degrade the overall image quality. Such adverse weather conditions profoundly impact daily life by deteriorating air quality, reducing visibility, and causing health hazards like respiratory problems and asthma. Furthermore, these conditions disrupt travel by causing accidents, flight delays, and challenges for marine and rail transportation systems. They also negatively affect agriculture and hinder business and economic activities. Addressing this issue manually is both time-consuming and labor-intensive. Therefore, this thesis proposes an automated system for enhancing surveillance video visibility using advanced machine learning and deep learning techniques.

The proposed system is comprised of two distinct models. The first model is “Automated Surveillance Video Visibility Detection Based on a Deep Supervised Learning Approach,” which is designed to detect and classify the surveillance video visibility conditions. This model employs a highly sophisticated deep learning architecture that is trained on a comprehensive dataset of various weather conditions to achieve accurate detection and classification. Supervised learning approach allows the model to learn from labeled data, thereby improving its predictive performance and reliability in real-world scenarios. The second model is the “Automated Surveillance Video Visibility Enhancement Based on Transfer Learning using an Unsupervised Learning Approach,” which aims to enhance the clarity of video frames. This model utilizes many unsupervised learning methods to improve image sharpness, contrast, and brightness, successfully reducing the interference of dust and fog-induced noise and distortions. Through the use of transfer learning, the proposed model may adjust each video frame clarity based on selecting the best model for the detected bad surveillance video visibility only in which the processing resources and time are decreasing. This dual-model system provides a comprehensive answer to the issue of reduced video quality in unfavorable weather conditions.

          The proposed system is implemented using a combination of selfly capture datasets that include videos affected by dust and fog. The obtained results demonstrate that the proposed system achieves an accuracy of 99.87% for video classification and a significant improvement in visibility for video enhancement. Furthermore, this proposed system outperforms previous methods in terms of both classification accuracy and enhancement quality.

The experimental result of enhancement algorithm for dusty which is multispectral video enhancement that SNR rise by (0.0857), BRISQUE decrease by (5.3833) and NIQE decreased by (0.405), for foggy video applying fog rectification enhancement that SNR rise by (0.0745), BRISQUE decrease by (3.48) and NIQE decreased by (0.39)

اترك تعليقاً