Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Automatic focusing method for human face region in high-magnification shallow depth-of-field state

A face area and automatic focus technology, applied in image data processing, television, biological neural network models, etc., can solve the problems of not being able to ensure the clarity of the background and characters at the same time, difficult to achieve real-time processing, single character details, etc., to achieve focus oscillation Unobtrusive, short search time, highly scalable effects

Active Publication Date: 2020-05-19
杭州晨安科技股份有限公司
View PDF5 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] (1) The depth of field will become shallow when the zoom lens is pulled to a large magnification, and it is impossible to ensure that the background and characters are clear at the same time
[0005] (2) The strategy of reducing noise and increasing the aperture will lead to a shallower depth of field, which cannot ensure that the background and characters are clear at the same time
[0006] (3) The background details are too rich, which is mistaken for the subject of the image in the autofocus algorithm and a false peak is calculated, resulting in a clear background and blurred characters
[0007] (4) The details of the characters are too single, and the weight in the auto-focus algorithm is too small, which is ignored by the algorithm when calculating FV (Focus Value), resulting in clear distances from other objects and blurred characters
However, in the face of well-designed, compact and cost-effective camera products, part of the computing power is often consumed in terms of imaging requirements, and then embedding face detection and area designation algorithms into the system will inevitably cause a shortage of computing power, making it difficult to achieve real-time processing.

Method used

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
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Automatic focusing method for human face region in high-magnification shallow depth-of-field state
  • Automatic focusing method for human face region in high-magnification shallow depth-of-field state
  • Automatic focusing method for human face region in high-magnification shallow depth-of-field state

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0058] The present invention will be further described in detail below with reference to the drawings and embodiments. The following embodiments are for explaining the present invention and the present invention is not limited to the following embodiments.

[0059] The present invention includes the following steps:

[0060] Step 1. Training the face detector based on YOLO V3 neural network, the steps are as follows:

[0061] (1) Build a neural network training server, and use GPU (Nvidia RTX 2080 Ti) to accelerate neural network training.

[0062] (2) Prepare training data: D={d 1 ,d 2 ,...,d n }, where D represents the face data set, d n Represents a single face image sample, n is the number of samples. In our training, n is approximately equal to 400,000, part of which comes from the open source face database, and the other part is the face data labeled by itself.

[0063] (3) Improve the YOLO V3 neural network, appropriately reduce the number of convolutional layers of YOLO V3, and ...

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

PUM

No PUM Login to View More

Abstract

The invention provides an automatic focusing method for a human face region in a high-magnification shallow depth-of-field state, which can perform human face detection, region optimization and automatic focusing in a camera product with insufficient computing power in a high-magnification shallow depth-of-field scene so as to find the clewest Focus motor position of a speaker region. The method comprises the following steps: 1, training a face detector based on a YOLO V3 neural network; 2, calculating a human face region set in the image by using a human face detector; 3, optimizing a human face region; 4, optimally selecting a face region to be matched with the region divided by the chip, and locking an interested focusing region; 5, calculating a region of interest; and 6, performing automatic focusing.

Description

Technical field [0001] The invention relates to a method for automatically focusing a face area in a state of large magnification and shallow depth of field, which is applied in the field of cameras for educational recording and broadcasting and video conferences. Background technique [0002] In recent years, in applications such as educational recording and broadcasting, video conferencing, zoom cameras such as 10x, 12x, or 20x are usually used to play real-time videos with characters as the subject in an indoor scene of 5-10 meters. The key point is to ensure that the character's face area is absolutely clear. In this context, it is proposed to use the face area as the key area of ​​autofocus to effectively solve this problem. [0003] Considering such a usage scenario, such as figure 1 As shown, in this three-dimensional scene, the background (blackboard, background wall, etc.), subject (speaker), and camera (10 times, 12 times, or 20 times) are in order from far to near. For...

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

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06T7/62G06N3/04H04N5/232
CPCG06T7/62G06V40/161H04N23/67H04N23/611G06N3/045
Inventor 王全强刘红艳毛海滨
Owner 杭州晨安科技股份有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products