الفهرس 
Chapter 1:Only 14 pages are availabe for public view
 |  | from | 115 |  |  |
| | | from | 115 | | |
Abstract
Crowd behaviour analysis is essential for systems of smart transportation to deliver efficient flow control and dynamic route planning for various transportation scenarios. This study adopts the challenge of crowd counting in adverse conditions of weather, such as heavy rain, which hampers traditional methods relying on visible spectrum perception.
A proposed approach is presented in this thesis to integrate advanced technologies that enhance crowd counting accuracy in complicated scenarios. The proposed approach comprises a two-step process. Firstly, a high-resolution rain removal (de-rain) model based on the U-Net-GAN architecture is implemented. This model efficiently eliminates heavy rain effect from crowd images, ensuring more clear visibility of scenes. Through Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), the de-raining model significantly enhances the quality of crowd images captured in inclement weather.
After rain removal, one of three crowd-counting models is carried out: Scale-adaptive selecting Network (SAS-Net), Segmentation-Guided Attention Network (SGA-Net), and Multi-CNN network. These models are designed to adapt to various crowd scales and complexities, capturing accurate crowd counts even in challenging conditions. The proposed framework, integrating the U-Net-GAN de-raining method with SAS-Net, demonstrates an exceptional performance. It yields highly-precise crowd-counting predictions with significantly-reduced Mean Squared Error (MSE). To confirm whether the suggested strategy is successful, different evaluation metrics are utilized, such as the Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR). The higher image quality is confirmed by these evaluations. Moreover, the robustness of the proposed framework is guaranteed in adverse weather conditions, establishing its effectiveness and reliability in real-world crowd management scenarios.