Target Acquisition Method and Equipment

Abstract

The present application provides a target acquisition method and device, which obtains the target estimated position of each scale in the video frame according to the global characteristics of each video frame, and performs clustering processing on the target estimated position in each video frame Obtaining the corresponding target candidate area, and according to all the target candidate areas in each video frame, combined with the confidence and corresponding scale processing of the target candidate area, determining the target actual area in the video frame can be quickly , Effective acquisition of one or more targets, especially the ability to accurately distinguish and acquire multiple targets.

Description

technical field [0001] The present application relates to the fields of communications and computers, and in particular to a target acquisition method and device. Background technique [0002] With the advent of the 4G era, the information transmission methods on the mobile terminal are not limited to text and pictures, but more from video. Internet companies have also launched relevant application interfaces one after another, and video information acquisition technology has become one of the recent research hotspots. [0003] Most of the existing multi-target tracking technologies are used in radar and aviation fields, including space tracking methods and time-space tracking methods. The features of the moving target are tracked; the time-space tracking method uses the features of the target in the space domain and the motion characteristics of the time domain at the same time, and it is divided into two types: contrast tracking and image correlation tracking. Others inc...

AI Technical Summary

Problems solved by technology

[0004] The applicable scenarios of existing multi-target tracking technology are relatively single, and the types of tracking targets are relatively single. Classification algorithm; on the other hand, multi-target tracking not only needs to face the problem of distinguishing the target and the background, but also needs to face the problem of distinguishing between targets

[0005] As far as the target tracking algorithm is concerned, the existing OpenTLD has achieved a relatively stable effect in terms of a simple single target, and the source code is open, but it is only for a single target

However, the existing multi-class target model is established by CNN, multi-target detection is performed at the beginning of the first frame of the video, the location of each target is given, and then the traditional target tracking method can be used for tracking. The process of target detection requires a lot of calculations, and it needs to train a huge model offline, which consumes a lot of computing and storage, and it is difficult to meet the requirements of real-time applications in video

Images