Video understanding method

• Main Authors: WANG JING, MIAO YANAN, HU WEIMING, TONG LINGLING, DU CUILAN, DENG ZHIFANG, ZHANG HONGYUAN, WANG BO, LI YANGXI
• Format: Patent
• Language: Chinese; English
• Published: 30.11.2018
• Subjects: CALCULATING; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention relates to the technical field of computer vision, in particular to a video understanding method which is aimed at solving the technical problems of how to extract the dense frame feature and long-term spatiotemporal feature of a video effectively. For the aim, the video understanding method provided by the invention includes the steps that first, a plurality of video frame groups ofa target video is obtained by residual network; then time sequence relation network is used to generate the time sequence relation feature of the video according to the video frame groups; and finally, the video behavior category of the target video is predicted according to the time sequence relation feature, wherein each video frame group comprises two ordered video frames, each ordered video frame comprises a plurality of video frames arranged in chronological order in sequence. Based on the above steps, the dense frame feature and the long-term spatiotemporal feature of the target video can be effectively obtaine