Video sparse detection method

Abstract

The invention discloses a video sparse detection method, which comprises the following steps of: pre-determining a corresponding relationship between a target characteristic and a working parameter for determining the working time of a tracker, which indicates that when the target has the characteristic, the tracker can replace a detector to perform target detection in a corresponding period; whenvideo detection is carried out, alternately starting the detector and the tracker to work; in each alternating period, starting the detector firstly, determining target features according to a detection result, and matching tracker working parameters corresponding to the target features through the corresponding relation; and initializing the tracker as a detector, and carrying out target detection by the tracker according to the working parameters. On the basis that the current detector works independently, the tracker is used for replacing detection tasks of part of video frames, so that onthe basis that the detection performance of a small part of video is not lost or is only lost, the calculation expenditure of video detection is greatly reduced, and the processing speed is increased.

Description

technical field [0001] The invention relates to the technical fields of target recognition and target extraction, in particular to a video sparse detection method. Background technique [0002] Among all kinds of big data, images and videos are "the largest big data". According to Cisco statistics, video content accounts for about 90% of the total Internet traffic; and in the rapidly developing mobile network, the proportion of video traffic is as high as 64%, and is growing at a compound annual growth rate of more than 130%. It can be seen that image and video data occupy a dominant position in big data, so the processing of image and video is the key to the application of big data. Moreover, compared with text, voice and other data, images and videos have larger data volumes and higher dimensions, and their expression, processing, transmission and utilization are more technically challenging. Therefore, integrating computer vision technology into the video data processin...

AI Technical Summary

Problems solved by technology

However, although the calculation resource consumption of the tracker is low, the calculation of the target features is relatively rough, and there is often a phenomenon of tracking and loss, which leads to errors such as local missed detection and overall missed detection of the target.

[0008] It can be seen that the existing technology cannot take into account the detection accuracy and detection speed

Images