High-accuracy intelligent identification target tracking system and method for security camera

A technology of intelligent recognition and target tracking, applied in character and pattern recognition, instruments, biological neural network models, etc., can solve problems such as imbalance
CN112613558APending Publication Date: 2021-04-06WUHAN INSTITUTE OF TECHNOLOGY
2 Cites 2 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
WUHAN INSTITUTE OF TECHNOLOGY
Publication Date
2021-04-06

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention provides a high-accuracy intelligent identification target tracking system and method for a security camera, and the system and method solve a problem of class imbalance through the cascading of a series of RPN (Region Proposal Network), achieve the full mining of cross-layer features, and achieve a stable visual tracking function. According to the invention, a new multi-stage tracking framework, namely the twin cascaded RPN, is proposed for the first time, a tracker based on the twin cascaded RPN is achieved, the recognition capability of the twin cascaded RPN for utilizing advanced semantic information and low-level spatial information is further improved, and through multi-step regression, the positioning is more accurate, and the distribution sequence of training samples is more balanced; the classifier of the RPN is more discriminative in sequence when more difficult interferents are distinguished, and target positioning is more accurate and has real-time performance under a complex background.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of security cameras, and in particular relates to a high-precision intelligent identification target tracking system and method for security cameras. Background technique

[0002] In terms of object tracking, Siamese-RPN has achieved good results, but may drift to the background, especially in the presence of similar semantic interference. There are two main reasons:

[0003] First, the distribution of training samples is unbalanced: (1) the positive samples are much smaller than the negative samples, resulting in invalid Siamese network training; (2) most of the negative samples are simple negative samples (non-similar non-semantic backgrounds), which are important in learning to distinguish There is little useful information when using a classifier. Therefore, the classifier is dominated by easily classified background samples, and the performance of the classifier degrades when encountering difficult si...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More