Vision enhancement using artificial intelligence techniques

This dissertation uses the concept of artificial intelligence and deep learning, with a focus on removing rainwater from heavy rain videos based on a former research published on CVPR in 2018 carried out by my supervisor. Clear vision is important for vision based system. To provide clear vision, i...

全面介紹

Saved in:
書目詳細資料
主要作者: Yu, Xiaohan
其他作者: Chau Lap Pui
格式: Theses and Dissertations
語言:English
出版: 2019
主題:
在線閱讀:http://hdl.handle.net/10356/78838
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
機構: Nanyang Technological University
語言: English
實物特徵
總結:This dissertation uses the concept of artificial intelligence and deep learning, with a focus on removing rainwater from heavy rain videos based on a former research published on CVPR in 2018 carried out by my supervisor. Clear vision is important for vision based system. To provide clear vision, it is important to ensure there is no rain occlusion in rainy weather. Proposed papers show limitations dealing with both fast moving cameras and pretty heavy rain. In the former paper, their main idea was to create a method to achieve perfect frame alignment. So they proposed an algorithm using super pixel segmentation to increase the robust of towards rain occlusion and fast camera motion. This dissertation is to improve the rain mask generated by the Optimal Temporal Match Tensor T0 through the former method. It made good use of both former generated rain mask and T0 using 3D convolution along with 2D convolution to form a more accurate rain mask, which will be used to get the better performance of final derain sequences together with Sorted Spatial-Temporal Match Tensor T1 in the future work. Instead of using simple decision rule to decide the location of the raindrops, this dissertation came up with a method making use of the time coherence between adjacent frames of each sequence.