CN111582166B - Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection - Google Patents

Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection Download PDF

Info

Publication number
CN111582166B
CN111582166B CN202010379602.8A CN202010379602A CN111582166B CN 111582166 B CN111582166 B CN 111582166B CN 202010379602 A CN202010379602 A CN 202010379602A CN 111582166 B CN111582166 B CN 111582166B
Authority
CN
China
Prior art keywords
video image
image
rectangular frame
modeling
steps
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010379602.8A
Other languages
Chinese (zh)
Other versions
CN111582166A (en
Inventor
黄远
刘毅
林鹏程
梁刚
郭昊
林涛睿
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Vclusters Information Technology Co ltd
Original Assignee
Shenzhen Vclusters Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Vclusters Information Technology Co ltd filed Critical Shenzhen Vclusters Information Technology Co ltd
Priority to CN202010379602.8A priority Critical patent/CN111582166B/en
Publication of CN111582166A publication Critical patent/CN111582166A/en
Application granted granted Critical
Publication of CN111582166B publication Critical patent/CN111582166B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for detecting remnants based on Gaussian modeling and YoLo V3 detection, which covers the fields of computer vision, artificial intelligence, object detection, intelligent monitoring and the like. The detection method comprises the following steps: performing frame extraction processing according to video image data obtained by monitoring video, then cutting operation, and performing graying pretreatment; carrying out Gaussian mixture modeling operation on the preprocessed video image, and separating background video images to obtain foreground video images; performing Gaussian mixture modeling operation on the obtained foreground video image, and separating the foreground video image to obtain a new background video image; acquiring a left object in a background video image to judge whether the left object is a pedestrian or not; tracking the track and timing after the centroid is acquired; outputting the video image to an original video image for alarming; the method has high adaptability to the environment, high anti-interference performance and high robustness.

Description

Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection
Technical Field
The invention relates to application of a computer vision technology in the field of intelligent monitoring, in particular to a method for detecting a legacy based on Gaussian modeling and YoLo V3 target detection.
Background
Nowadays, with the popularization of video monitoring equipment, a monitoring camera of the mildly hemp is installed in public places such as airports, subways, trains and the like, and in places such as shops and stores. However, since the previous video monitoring can only store and view the monitored scene, some critical places also need to wait for staff day and night, which wastes manpower and material resources greatly, and can not give out warning when something happens, thus the improvement and upgrading of the traditional video monitoring is an urgent matter.
A better method is to install a 'brain' for traditional video monitoring, and can automatically identify the abnormality in the monitoring scene, such as the detection of the left-over object, and realize real-time alarm.
There are several existing methods for detecting carryover:
(1) The method comprises the following steps: the method has the advantages that the double-background modeling is adopted, and then the difference taking operation is carried out, so that the retention target is obtained, but the calculation amount involved in the method is huge, the larger the image is, the clearer the image is, the more time is consumed, the accuracy is reduced due to the fact that the definition is reduced, and the method is not suitable for monitoring scenes needing real-time detection.
(2) The second method is as follows: firstly, modeling a monitoring video image, setting a threshold value, marking a foreground object with overlong existence time, screening an object mask from the foreground object, carrying out characteristic analysis on the object image, carrying out correlation degree assessment on the foreground object marked by a source image and the screened object, and judging as a carry-over object if the foreground object marked by the source image is larger than the threshold value. However, this method has a disadvantage that the influence of noise is relatively large, and if a noise is marked, the subsequent operation is not affected, and the false recognition rate is increased.
(3) And a third method: the method for updating the double-mask background adopts double-background modeling: firstly, a long-short double-background Gaussian model is established, then, subtraction operation is carried out on the foreground obtained by the short background model and the long background model, the obtained binary image is a left-behind object, and the defect is that the short background model is very fast in updating time, and the foreground image of the short background Gaussian model can be quickly fused into the background, so that the detection accuracy of the left-behind object is low.
Accordingly, the prior art has problems and further improvements are needed.
Disclosure of Invention
In order to make up for the defects of the prior art, the invention provides a method for detecting the carryover based on Gaussian modeling and YoLo V3 target detection, which avoids the problems of low detection speed, high complexity and the like caused by using a double-background model, can cope with target shielding to a certain extent so as to reduce the detection omission of the carryover, reduces false detection caused by shadow, illumination and other problems, removes false detection caused by moving objects in a scene, solves the problem of false alarm and height omission caused by the detection of the carryover, and improves the accuracy of the detection of the carryover.
In order to achieve the above purpose, the specific technical scheme of the invention is as follows:
a method for detecting a legacy based on Gaussian modeling and YoLo V3 target detection comprises the following steps:
f1, collecting an original image;
f2, modeling by using Gaussian mixture to obtain a foreground video image;
f3, performing secondary modeling to obtain a background video image;
f4, obtaining a coordinate point of the retained object and an external rectangular frame;
f5, excluding pedestrians to avoid false detection;
and F6, tracking an external rectangular frame, setting a time threshold and giving an alarm in real time.
The step F1 is implemented by collecting an original image and using the following steps:
(1) Performing frame extraction processing on an original monitoring video image, taking 1 frame from 10 frames, and then adjusting the Resize of the synthesized monitoring video image to be between 100 x 100 and 1280 x 720, wherein generally 352x288 can meet the requirements of real-time and accuracy simultaneously;
(2) The image is subjected to graying treatment to obtain a gray image, so that the Gaussian modeling operand can be greatly reduced, and the memory pressure can be relieved;
and F2, carrying out Gaussian mixture modeling to obtain a foreground video image, wherein the foreground video image is realized by the following steps:
(1) For video images with frame rate more than 500 frames, adopting a normal modeling sensitivity method; for video images with frame rate less than 500 frames, a rapid modeling method is adopted, so as to ensure that images with low frame rate are not limited by modeling sensitivity;
(2) Thresholding the separated foreground video image;
and F3, secondarily modeling to obtain a background video image, wherein the background video image is realized by the following steps:
(1) Modeling the foreground video image obtained in the step (2) in the step F1 again, so that noise caused by micro-motion objects such as leaves, fans and the like can be reduced, and a clearer video image of a detained object can be obtained;
(2) Thresholding the separated background video image;
and F4, obtaining a coordinate point of a detained object and an external rectangular frame, and realizing the method by using the following steps:
(1) Acquiring coordinate points of objects in the background video image in the step (2) in the F3 and externally connecting rectangular frames;
(2) Feeding back the coordinate points and the circumscribed rectangle frame to the original image in the step (1) in the F1;
according to the step F5, the false detection caused by pedestrian avoidance is eliminated, and the method is realized by the following steps:
(1) Selecting all pedestrian frames in the original image through a YoLo v3 target detection algorithm, outputting rectangular frame coordinate values, and assuming that the subscript starts from 1;
(2) F, calculating the superposition area of all the pedestrian frames and the circumscribed rectangular frame of the object obtained in the step F4, setting a proportion threshold value of 0.85, and if the proportion threshold value is larger than the proportion threshold value, considering the object in the circumscribed rectangular frame as a pedestrian to be abandoned;
step F6 is to track an external rectangular frame, set a time threshold and give an alarm in real time, and is realized by the following steps:
(1) Tracking the target track through the mass center;
(2) A time threshold is set for the rectangular frame, and if the existing time exceeds 20s, the additional rectangular frame mark is output to the original video image.
Compared with the traditional method for detecting the residues, the invention has the following advantages:
(1) The method adopts a secondary modeling mode, has low pixel requirement, can detect the resolution of 100 x 100 at the lowest, has little operand, can meet the requirement of real-time detection, and improves the real-time efficiency of an algorithm;
(2) The invention has low sensitivity to noise, and a pedestrian detection and filtration mode is adopted to screen out a batch of pedestrians causing false detection, so that the false detection rate of object retention detection is reduced;
(3) The invention adopts a mode of setting the threshold value of time, can prevent the object from being taken away suddenly after being detained for a period of time, and can improve the accuracy of detection;
drawings
FIG. 1 is a workflow diagram of an overall implementation of the present invention;
FIG. 2 is a flow chart of the present invention for avoiding false detection by pedestrians.
Detailed Description
In order that those of ordinary skill in the art will readily understand and practice the invention, embodiments of the invention will be further described with reference to the drawings.
In order to reduce the shortcoming of huge operand brought by double-background modeling, the monitoring video image can be detected in real time, and in order to provide the accuracy of detecting the carryover, the invention provides a method for detecting the carryover based on Gaussian modeling and YoLo V3 target detection, and the specific steps are shown in the accompanying figure 1, and the following description is given in the following steps of the accompanying figure 1:
f1 collecting original image
The video monitoring image is obtained through video streaming, and because the image motion difference between frames is not changed greatly, and in order to reduce the modeling operation amount, one image is extracted every 10 frames, and the image resize is adjusted to a certain range interval, and the resolution of 352x288 is calculated according to an empirical value, so that the requirements of high accuracy and real-time can be met. And the RGB image is converted into a gray image, so that the calculation amount of the modeling work in the next step is reduced.
F2, modeling the Gaussian mixture to obtain a foreground video image;
in the last step we have obtained the video image that was produced and done some pre-processing work. In this step, the video image will be modeled for the first time, and the foreground and background images will be separated by the mixture gaussian modeling.
F3, modeling for the second time to obtain a background video image;
for the foreground image obtained in the last step, noise phenomena caused by a plurality of micro-motion objects such as clouds, wind leaves and fan lamps are unfavorable for later analysis, so that the foreground image is modeled again, and the interference of noise on the analysis of the retentate is removed.
F4, obtaining a coordinate point of the retained object and an external rectangular frame;
the background image after the secondary modeling is used, because the noise is in continuous variation, the background image always keeps a foreground in the process of the secondary modeling, and a stationary object is converted into a background video image;
traversing the video image through an 8-connected-area algorithm, setting the initial area of the connected area to be 5, removing part of noise to obtain all the connected areas, and obtaining 4 coordinate points, an external rectangular frame and a centroid according to the positions of the connected areas in the image.
F5, excluding pedestrians to avoid false detection;
and obtaining an external rectangular frame of the suspected left object through the background video image, outputting the rectangular frame to the original image through the step F4, cutting the partial image, and adjusting the size of the cut image to 212x 212.
Inputting the cut image into a YoLo V3 network model to detect whether a pedestrian target exists, referring to fig. 2, the specific implementation process is described as follows:
1) If no pedestrian exists, sending the suspected left-over object circumscribed rectangular frame into the step F6;
2) If the pedestrian exists, outputting a pedestrian rectangular frame area A, and performing intersection calculation on the pedestrian frame and the cut image, wherein the calculation formula is as follows: c=a/(212 x 212);
3) If C >0.85, the left-over object is judged to be a pedestrian, and the rectangular box is discarded;
4) If C <0.85, the left-over object is judged to be the left-over object, and the left-over object is sent to step F6 for the next processing;
f6, tracking an external rectangular frame, setting a time threshold and giving an alarm in real time;
and F5, judging all the circumscribed rectangular frames to obtain a circumscribed rectangular frame of a non-pedestrian, tracking the mass center of the circumscribed rectangular frame obtained in the F4, and if the existence time of the circumscribed rectangular frame is longer than 20s, sending the circumscribed rectangular frame into an original image for marking, displaying the rectangular frame on the original video image and alarming.
The foregoing examples illustrate only a few embodiments of the invention and are described in detail herein without thereby limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.

Claims (2)

1. A method for detecting a legacy based on gaussian modeling and YoLo V3 target detection, comprising the steps of:
(1) Collecting an original image;
(2) Modeling by using a Gaussian mixture to obtain a foreground video image;
(3) Secondarily modeling to obtain a background video image;
(4) Obtaining a coordinate point of a detained object and an external rectangular frame and a centroid;
(5) The pedestrian is eliminated, so that false detection is avoided;
(6) Tracking an external rectangular frame, setting a time threshold value and giving an alarm in real time;
the method for acquiring the original image in the step (1) comprises the following steps:
performing frame extraction processing according to video image data obtained by monitoring video, then cutting operation, and performing graying pretreatment;
the step (2) specifically comprises the following steps:
carrying out Gaussian mixture modeling operation on the preprocessed video image, and separating background video images to obtain foreground video images;
the step (3) specifically comprises the following steps:
performing Gaussian mixture modeling operation on the acquired foreground video image, and separating the foreground video image to obtain a new background video image;
the step (4) specifically comprises the following steps:
detecting all the background targets in the communication area to obtain coordinate points of the object and positions of the circumscribed rectangular frame and the mass center;
the step (5) comprises the following steps:
a) Obtaining a coordinate point of an object and an external rectangular frame through the step (4), and then matting the position of the rectangular frame on the original video image and cutting the position of the rectangular frame into 212x212 resolution;
b) Sending the image into a YoLo V3 model, and identifying whether pedestrians exist;
c) If not, sending the circumscribed rectangle frame corresponding to the image into the step (6) for tracking;
d) If the pedestrian exists, carrying out intersection operation on the rectangular pedestrian frame and the image to obtain an area threshold D;
e) If the area threshold D is more than 0.85, judging that the circumscribed rectangular frame corresponding to the image is a pedestrian, and abandoning the operation of the step (6);
f) If the area threshold D is less than 0.85, determining that the image corresponds to an object in the circumscribed rectangular frame, and sending the object to the step (6) for tracking.
2. The method for detecting the carryover based on Gaussian modeling and YoLo V3 target detection according to claim 1, wherein the step (6) is characterized in that all circumscribed rectangular frames are subjected to target tracking through the centroid of the circumscribed rectangular frames, if any circumscribed rectangular frame exists for more than 20s, the circumscribed rectangular frame is sent into an original image for marking, the rectangular frame is displayed on the original video image, and an alarm is given.
CN202010379602.8A 2020-05-07 2020-05-07 Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection Active CN111582166B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010379602.8A CN111582166B (en) 2020-05-07 2020-05-07 Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010379602.8A CN111582166B (en) 2020-05-07 2020-05-07 Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection

Publications (2)

Publication Number Publication Date
CN111582166A CN111582166A (en) 2020-08-25
CN111582166B true CN111582166B (en) 2023-05-05

Family

ID=72113277

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010379602.8A Active CN111582166B (en) 2020-05-07 2020-05-07 Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection

Country Status (1)

Country Link
CN (1) CN111582166B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114394100B (en) * 2022-01-12 2024-04-05 深圳力维智联技术有限公司 Unmanned patrol car control system and unmanned car
CN114205583B (en) * 2022-01-20 2024-01-16 深圳市瑞驰信息技术有限公司 Video coding method and system based on H265 and electronic equipment
CN114694092A (en) * 2022-03-15 2022-07-01 华南理工大学 Expressway monitoring video object-throwing detection method based on mixed background model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016024743A (en) * 2014-07-23 2016-02-08 西谷 隆夫 Moving image processor, moving image information processing program and moving image processing method
CN107507221A (en) * 2017-07-28 2017-12-22 天津大学 With reference to frame difference method and the moving object detection and tracking method of mixed Gauss model
WO2018130016A1 (en) * 2017-01-10 2018-07-19 哈尔滨工业大学深圳研究生院 Parking detection method and device based on monitoring video
CN109558790A (en) * 2018-10-09 2019-04-02 中国电子科技集团公司电子科学研究院 A kind of pedestrian target detection method, apparatus and system

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104156942B (en) * 2014-07-02 2017-02-15 华南理工大学 Detection method for remnants in complex environment
CN107527009B (en) * 2017-07-11 2020-09-04 浙江汉凡软件科技有限公司 Remnant detection method based on YOLO target detection
CN107871315B (en) * 2017-10-09 2020-08-14 中国电子科技集团公司第二十八研究所 Video image motion detection method and device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2016024743A (en) * 2014-07-23 2016-02-08 西谷 隆夫 Moving image processor, moving image information processing program and moving image processing method
WO2018130016A1 (en) * 2017-01-10 2018-07-19 哈尔滨工业大学深圳研究生院 Parking detection method and device based on monitoring video
CN107507221A (en) * 2017-07-28 2017-12-22 天津大学 With reference to frame difference method and the moving object detection and tracking method of mixed Gauss model
CN109558790A (en) * 2018-10-09 2019-04-02 中国电子科技集团公司电子科学研究院 A kind of pedestrian target detection method, apparatus and system

Also Published As

Publication number Publication date
CN111582166A (en) 2020-08-25

Similar Documents

Publication Publication Date Title
CN111582166B (en) Method for detecting remnants based on Gaussian modeling and YoLo V3 target detection
EP1805715B1 (en) A method and system for processing video data
JP4668921B2 (en) Object detection in images
US7778445B2 (en) Method and system for the detection of removed objects in video images
US7916944B2 (en) System and method for feature level foreground segmentation
US20140198257A1 (en) System and method for motion detection in a surveillance video
CN110659546A (en) Illegal booth detection method and device
CN103793921B (en) Moving object extraction method and moving object extraction device
JP7125843B2 (en) Fault detection system
US20130027550A1 (en) Method and device for video surveillance
CN112866654B (en) Intelligent video monitoring system
CN113920585A (en) Behavior recognition method and device, equipment and storage medium
CN114821978B (en) Method, device and medium for eliminating false alarm
Chandrasekhar et al. A survey of techniques for background subtraction and traffic analysis on surveillance video
JP3736836B2 (en) Object detection method, object detection apparatus, and program
CN111145219B (en) Efficient video moving target detection method based on Codebook principle
CN113947744A (en) Fire image detection method, system, equipment and storage medium based on video
JP3536913B2 (en) Falling object detection system, falling object detection method, and recording medium
JPH05183901A (en) Moving body monitor device
KR20150055481A (en) Background-based method for removing shadow pixels in an image
CN115719471A (en) Detection method and device for remnants
JP3490196B2 (en) Image processing apparatus and method
Kryjak Foreground object segmentation in dynamic background scenarios
CN117292296A (en) Falling object detection method and system based on laser shooting and self-encoder
CN115482253A (en) Moving object detection method and device and perimeter intrusion detection system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant