期刊名称：Indian Journal of Computer Science and Engineering
印刷版ISSN：2231-3850
电子版ISSN：0976-5166
出版年度：2021
卷号：12
期号：4
页码：1149-1162
DOI：10.21817/indjcse/2021/v12i4/211204211
语种：English
出版社：Engg Journals Publications
摘要：The majority of extant in recognition of action research is focused on high-quality videos with clearly apparent activity. The actions in videos are collected at a variety of resolutions in real-world surveillance situations. Most actions take place at a distance with a low resolution, and identifying such events is a difficult task. In this paper, we take a look at the effect of low video quality on human activity location from two points: recordings that are weakly gathered topographically (low level resolution) and transiently (lower frame rate), and packed recordings with movement obscuring and artifacts. To recognize the action four main steps are considered. First one is background subtraction using LPB operator. Second, the features are extracted using histogram of gradients (HOG) and Histogram of optical flow (HOF) algorithm which is used to estimate the motion of a person and eigen value algorithm which is used to recognize the person. These features are fused together and perform Firefly optimization (FO) technique to obtain optimized features. Thirdly, generate a code book for feature encoding. In feature encoding, the process is performed using bag of words. Finally action recognition is done using Support vector machine (SVM) classifier. The experimental results are performed on low resolution datasets like VIRAT dataset, KTH dataset and Soccer dataset. The result obtained using VIRAT is 91.46%, KTH is 92.40 % and Soccers is 90.51% in terms of accuracy.
