CN109478328A

CN109478328A - Method for tracking target, device and image processing equipment

Info

Publication number: CN109478328A
Application number: CN201680087596.0A
Authority: CN
Inventors: 白向晖; 伍健荣
Original assignee: Fujitsu Ltd
Current assignee: Fujitsu Ltd
Priority date: 2016-09-30
Filing date: 2016-09-30
Publication date: 2019-03-15
Also published as: WO2018058531A1

Abstract

A target tracking method, device and image processing device. The target tracking method includes: determining a plurality of different sizes of the detection target according to a predetermined scaling factor; for each current size, calculating a reference block of the detection target and one or more of the detection targets in the current frame. feature vector distance between candidate blocks, and determine the matching value and candidate block of the current size according to the feature vector distance; Tracking results in the current frame. Thus, even if the detection target is significantly enlarged or reduced in the image frame, the accuracy of the target tracking will not be reduced.

Description

PCT国内申请，说明书已公开。PCT domestic application, the description has been published.

Claims

A kind of method for tracking target, tracks the detection target in video, and the method for tracking target includes:

Multiple and different sizes of the detection target are determined according to scheduled zoom factor；

For each current size, the reference block and the detection target feature vector distance between one or more candidate blocks in the current frame of the detection target are calculated, and determines the matching value and candidate blocks of the current size according to described eigenvector distance；

According to the multiple various sizes of matching value and candidate blocks, tracking result of the detection target in the present frame is determined.
Method for tracking target according to claim 1, wherein the multiple difference size includes: normal size, minification and expansion size.
Method for tracking target according to claim 1, wherein the method for tracking target further include:

The reference block of the detection target is determined for the current size；

Level sampling interval and Vertical Sampling interval are determined according to the size of the reference block.
Method for tracking target according to claim 3, wherein calculate the reference block and the detection target feature vector distance between one or more candidate blocks in the current frame of the detection target, comprising:

According to the position of the reference block, the level sampling interval and the Vertical Sampling interval, the search window in the present frame is set；

According to the size of the reference block, the level sampling interval and the Vertical Sampling interval, one or more candidate blocks are extracted from described search window；

Feature vector is generated for each candidate blocks and the reference block；And

Calculate the vector distance between the feature vector of the reference block and the feature vector of each candidate blocks.
Method for tracking target according to claim 4, wherein the matching value and candidate blocks of the current size are determined according to described eigenvector distance, comprising:

The smallest candidate blocks of vector distance between the feature vector of the reference block are determined as to the tracking block of the current size；And

Using the smallest vector distance as the matching value of the current size.
Method for tracking target according to claim 4, wherein generate feature vector for each candidate blocks and the reference block, comprising:

According to the level sampling interval and the Vertical Sampling interval, the pixel in the candidate blocks or the reference block is sampled；

The characteristic value of the multiple pixels obtained using sampling constructs the feature vector of the candidate blocks or the reference block.
Method for tracking target according to claim 6, wherein the characteristic value of the multiple pixels obtained using sampling constructs the feature vector of the candidate blocks or the reference block, comprising:

The gray value for multiple pixels that sampling obtains is configured to gray feature vector；

The histograms of oriented gradients value for multiple pixels that sampling obtains is configured to histograms of oriented gradients feature vector；

Histograms of oriented gradients feature vector described in the gray feature vector sum is merged, the feature vector of the candidate blocks or the reference block is obtained.
Method for tracking target according to claim 7, wherein the method for tracking target further include:

Weight coefficient is assigned respectively for histograms of oriented gradients feature vector described in the gray feature vector sum；

And histograms of oriented gradients feature vector described in the gray feature vector sum after imparting weight coefficient is merged.
Method for tracking target according to claim 1, wherein tracking result of the detection target in the present frame is determined according to the multiple various sizes of matching value and candidate blocks, comprising:

The matching value under more the multiple difference size；And

Candidate blocks corresponding to smallest match value are determined as tracking block of the detection target in the present frame.
Method for tracking target according to claim 9, wherein the method for tracking target further include:

The reference block is updated using the information of the tracking block.
A kind of target tracker, tracks the detection target in video, and the target tracker includes:

Size determination unit determines multiple and different sizes of the detection target according to scheduled zoom factor；

Candidate determination unit, it is for each current size, the reference block and the detection target feature vector distance between one or more candidate blocks in the current frame of the detection target are calculated, and determines the matching value and candidate blocks of the current size according to described eigenvector distance；

As a result determination unit determines tracking result of the detection target in the present frame according to the multiple various sizes of matching value and candidate blocks.
Target tracker according to claim 11, wherein the multiple difference size includes: normal size, minification and expansion size.
Target tracker according to claim 11, wherein the target tracker further include:

Reference block determination unit determines the current size reference block of the detection target；

It is spaced determination unit, level sampling interval and Vertical Sampling interval are determined according to the size of the reference block.
Target tracker according to claim 13, wherein it is described candidate determination unit include:

The search window in the present frame is arranged according to the position of the reference block, the level sampling interval and the Vertical Sampling interval in window setting unit；

Candidate blocks extracting unit extracts one or more candidate blocks according to the size of the reference block, the level sampling interval and the Vertical Sampling interval from described search window；

Vector generation unit is that each candidate blocks and the reference block generate feature vector；And

Metrics calculation unit calculates the vector distance between the feature vector of the reference block and the feature vector of each candidate blocks.
Target tracker according to claim 14, wherein candidate's determination unit further include:

The smallest candidate blocks of vector distance between the feature vector of the reference block are determined as the tracking block of the current size by candidate blocks determination unit；And

Matching value determination unit, using the smallest vector distance as the matching value of the current size.
Target tracker according to claim 14, wherein the vector generation unit includes:

Pixel sampling unit samples the pixel in the candidate blocks or the reference block according to the level sampling interval and the Vertical Sampling interval；

The characteristic value of vector structural unit, the multiple pixels obtained using sampling constructs the feature vector of the candidate blocks or the reference block.
Target tracker according to claim 16, wherein the vector structural unit includes:

The gray value for multiple pixels that sampling obtains is configured to gray feature vector by primary vector structural unit；

The histograms of oriented gradients value for multiple pixels that sampling obtains is configured to histograms of oriented gradients feature vector by secondary vector structural unit；And

Vector combining unit merges histograms of oriented gradients feature vector described in the gray feature vector sum, obtains the feature vector of the candidate blocks or the reference block.
Target tracker according to claim 11, wherein the result determination unit includes:

Matching value comparing unit, it is more the multiple difference sizes under the matching value；And

Block determination unit is tracked, candidate blocks corresponding to smallest match value are determined as tracking block of the detection target in the present frame.
Target tracker according to claim 18, wherein the target tracker further include:

Reference block updating unit is updated the reference block using the information of the tracking block.
A kind of image processing equipment, wherein described image processing equipment includes target tracker as claimed in claim 11.