Executive Summary : | Computer vision has made its impact on various fields and is primarily used for various applications such as Intelligent Transportation, Autonomous Driving, Depth Estimation, Image and Video Restoration, Data Generation, Moving Object Segmentation, Brain Tumour Segmentation, etc. Deep learning algorithms have been a major influence towards betterment of these computer vision applications through its generalizability. However, various environmental factors such as different weather conditions affect the performance of such computer vision applications as these applications expect clean data as the input. Weather degradations include rain streaks, rain drops, snow and haze or fog. These degradations may overlap on potentially useful information in an image or video frame and affect performance of computer vision applications. Removal of effects of weather degradations from images and videos is a challenging task, as each weather possesses certain physical properties. Existing literature consist of various methods for weather-specific application, e.g., rain removal only, haze removal only, etc. But, there exist very few algorithms which aim at restoration of different types of weather degradations in a unified manner (i.e., through a single network). Further, existing methods have large computational complexity, which limits their applicability in practical scenarios. Hence, there is a dire need of unified automatic technology for multi-weather video restoration. This project would provide novel practices of research in the area of multi-weather video restoration for the development of intelligent transportation systems. The combination of deep convolutional neural networks (CNN), natural (RGB) video and Thermal video data would be utilized to develop the new unified technology for multi-weather image/video restoration. |