Research Article
Implementation of Video Abstract Algorithm Based on CUDA
@INPROCEEDINGS{10.1007/978-3-319-73447-7_43, author={Hui Li and Zhigang Gai and Enxiao Liu and Shousheng Liu and Yingying Gai and Lin Cao and Heng Li}, title={Implementation of Video Abstract Algorithm Based on CUDA}, proceedings={Machine Learning and Intelligent Communications. Second International Conference, MLICOM 2017, Weihai, China, August 5-6, 2017, Proceedings, Part II}, proceedings_a={MLICOM}, year={2018}, month={2}, keywords={Video abstract Gaussian mixture model Particle filter GPU CUDA Parallel computing}, doi={10.1007/978-3-319-73447-7_43} }
- Hui Li
Zhigang Gai
Enxiao Liu
Shousheng Liu
Yingying Gai
Lin Cao
Heng Li
Year: 2018
Implementation of Video Abstract Algorithm Based on CUDA
MLICOM
Springer
DOI: 10.1007/978-3-319-73447-7_43
Abstract
The dynamic video abstract is an important part of video content analysis. Firstly, the objective of motion is analyzed, and the objective of the movement is extracted. Then, the moving trajectory of each target is analyzed, and different targets are spliced into a common background scene, and they are combined in some way. The algorithm uses Gaussian mixture model and particle filter to do a large number of calculations to achieve the background modeling and the detection of moving object. With the increase of image resolution, the computing increased significantly. To improve the real-time performance of the algorithm, a video abstract algorithm based on CUDA is proposed in this paper. Through the data analysis and parallel mining of the algorithm, time-consuming modules of the calculation, such as Histogram equalization, Gaussian mixture model, particle filter, were implemented in GPU by using massively parallel processing threads to improve the efficiency. The experimental results show that the algorithm can improve the calculation speed significantly in NVIDIA Tesla K20 and CUDA7.5.