Target site association method and device, terminal and computer readable storage medium
CN116052210BActive Publication Date: 2026-08-28ZHEJIANG DAHUA TECH CO LTD
Patent Information
- Application Number
- CN202211689106.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-27
- Publication Date
- 2026-08-28
- Estimated Expiration
- 2042-12-27
AI Technical Summary
Technical Problem
[0003]本发明主要解决的技术问题是提供一种目标部位关联方法、装置、终端及计算机可读存储介质,解决现有技术中关联准确率比较低的问题
Benefits of technology
[0035]本发明的有益效果是:区别于现有技术的情况,提供的一种目标部位关联方法、装置、终端及计算机可读存储介质,目标部位关联方法包括:对视频帧进行多目标检测,得到多个检测部位数据集;各检测部位数据集包括视频帧中具有相同检测类别的部位信息;响应于一检测部位数据集对应的检测类别为第一类别,另一检测部位数据集对应的检测类别为第二类别,则基于第一类别对应的各部位信息与第二类别对应的各部位信息组建候选关联对;其中,第二类别为第一类别的子类别;通过孪生网络模型对候选关联对进行识别,确定候选关联对对应的部位信息之间是否进行关联。本申请通过孪生网络模型检测第一类别的部位信息和第二类别的部位信息是否属于同一目标对象,进而实现对不同检测类别的部位信息之间的关联,大大提高了关联的准确性。
✦ Generated by Eureka AI based on patent content.
Smart Images

Figure CN116052210B_ABST
Abstract
The application provides a target part association method, device, terminal and computer readable storage medium, the target part association method comprises: performing multi-target detection on a video frame to obtain a plurality of detection part data sets; each detection part data set comprises part information with the same detection category in the video frame; in response to a detection category corresponding to one detection part data set being a first category and a detection category corresponding to another detection part data set being a second category, a candidate association pair is formed based on each part information corresponding to the first category and each part information corresponding to the second category; whether the part information corresponding to the candidate association pair is associated is determined by identifying the candidate association pair through a twin network model. The application detects whether the part information of the first category and the part information of the second category belong to the same target object through the twin network model, thereby realizing the association between the part information of different detection categories and greatly improving the accuracy of the association.
Need to check novelty before this filing date? Find Prior Art
Citation Information
Patent Citations
Associated object detection method and device
CN113544701A
Target tracking method, computer equipment and storage device
CN114548214A