• Ms. Kevadkar Pragati Ashok
• Prof. Jadhav Dattatraya A

Detecting shot boundary and gradual shot change happens to the prominent research problem in the field of video retrieval or indexing. Video shot boundary detection (SBD) is commonly an important and first step for indexing, and retrieval, video data, browsing, and content-based video analysis and many other such technologies. There have been great efforts put forward to enhance the precision of SBD algorithms. But many of these works are oriented towards signal and not towards interpretable features of frames. In our paper, we put forward a video shot boundary detection structure based on Convolutional Neural Networks (CNNs). Candidate segment selection is adopted which determines the locations of shot boundaries and removes most non-boundary frames. This kind of preprocessing method helps in improving both the speed and accuracy of the SBD algorithm. Later on, features of frames in a shot are synthesized and semantic labels for the shot are generated. This modular CNN network will be superior to the state-of-art methods on RAI Dataset with better than average real-time deduction speed even on just one mediocre GPU. Randomly generated transitions using selected shots from the TRECVID IACC.3 dataset will be employed under the training process. The proposed system achieved an average precision of 0.966, recall of 0.864 and F1 score of 0.912.

CNN, RAI Dataset, Shot Boundary Detection (SBD), Video Frame

"Deep Learning Approach for Video Shot Boundary Detection", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.7, Issue 6, page no.411-417, June-2020, Available :http://www.jetir.org/papers/JETIR2006058.pdf

"Deep Learning Approach for Video Shot Boundary Detection", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.7, Issue 6, page no. pp411-417, June-2020, Available at : http://www.jetir.org/papers/JETIR2006058.pdf