CN115063448A - Multi-target tracking method for oilfield operation scene - Google Patents

Multi-target tracking method for oilfield operation scene Download PDF

Info

Publication number
CN115063448A
CN115063448A CN202210739370.1A CN202210739370A CN115063448A CN 115063448 A CN115063448 A CN 115063448A CN 202210739370 A CN202210739370 A CN 202210739370A CN 115063448 A CN115063448 A CN 115063448A
Authority
CN
China
Prior art keywords
target
operation scene
detection
target tracking
image
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.)
Pending
Application number
CN202210739370.1A
Other languages
Chinese (zh)
Inventor
吴婷
梁鸿
张千
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.)
China University of Petroleum East China
Original Assignee
China University of Petroleum East China
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 China University of Petroleum East China filed Critical China University of Petroleum East China
Priority to CN202210739370.1A priority Critical patent/CN115063448A/en
Publication of CN115063448A publication Critical patent/CN115063448A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a multi-target tracking method and system oriented to an oilfield operation scene. The invention provides a multi-target tracking method oriented to an oilfield operation scene by adopting a mode of firstly detecting and then tracking and combining an advanced target detection model YOLOX. Firstly, extracting a detection frame of a target by using YOLOX, and converting the detection frame into a gray image; performing motion prediction on the target by using a Kalman filtering method, and extracting features of the predicted target through a pedestrian re-identification network; the invention changes the number of pedestrian re-identification network channels from three channels to a single channel, and removes the interference of color information on the identification result; and carrying out similarity comparison on the extracted features and the classification features to obtain similarity scores, and carrying out ID matching through a distribution algorithm to obtain tracking results. The method is oriented to the oil field operation scene, the model detection precision is high, the speed is high, and multiple targets can be tracked in real time.

Description

Multi-target tracking method for oilfield operation scene
Technical Field
The invention relates to the field of computer vision, in particular to a multi-target tracking method for an oil field operation scene.
Background
The multi-target tracking is a deep learning task for distinguishing a plurality of targets appearing in a video by using different IDs, and is widely applied to the fields of intelligent monitoring and unmanned driving. The current mainstream multi-target tracking method is based on a mode of detecting first and then tracking, the tracking effect depends on the quality of the target detection method to a great extent, and detectors can be fixed in some multi-target tracking competitions. The video is input into a target detection network frame by frame to generate a detection frame, and then the same target of the two frames before and after the frame is associated to obtain the track of the target.
Object detection is one of the most important and challenging branches of the computer vision field. Most of the most advanced object detectors use the deep learning network as their backbone and detection network to extract features from the input image (or video), respectively, for classification and localization. Object detection is a computer technology related to computer vision and image processing for detecting instances of a certain class of semantic objects (e.g., people, buildings, or cars) in digital images and videos. The research field of target detection comprises multi-class detection, edge detection, salient target detection, posture detection, scene text detection, face detection, pedestrian detection and the like. Target detection is an important component of scene understanding, and is widely applied to many fields of modern life, such as the safety field, the military field, the traffic field, the medical field and the life field.
Aiming at special scenes of oilfield operation, workers wear work clothes and safety helmets with the same color and style due to the regulation of relevant regulations of the oilfield, so that the appearances of targets are similar, and the difficulty in tracking and distinguishing personnel ID is caused; in addition, oil field operation scenes are mostly worked in the field, weather and light influence the collected images of the camera, and a plurality of large-scale devices are arranged, so that the environment is complex, and the problem that personnel are blocked by a machine and tracking fails easily occurs.
The development direction of oil extraction in the 21 st century will turn to the important direction of intelligent oil extraction, and the application of intelligent technology in the field of oil extraction development becomes the development trend of the petroleum industry. The multi-target tracking is applied to the oil field operation field, so that the walking track of workers can be identified, and whether the workers enter an illegal area, do not travel according to a specified route, shuttle the workers and the like is judged. The existing target detection can only meet the requirements of tracking large human-member appearance difference and simple environment in a common scene, and cannot ensure real-time and accurate tracking in an oil field scene. Therefore, the method for tracking multiple targets in the oil field scene is significant.
Disclosure of Invention
The invention aims to provide a multi-target tracking method for an oilfield operation scene, which aims to solve the problems in the prior art and can accurately detect and track workers in the oilfield operation scene in real time.
In order to achieve the above purpose, the invention provides the following scheme:
the invention provides a multi-target tracking method for an oilfield operation scene, which comprises the following steps:
s1, acquiring image data in an oilfield operation scene, performing data annotation through the data after image preprocessing by using data set annotation software, and dividing a training set and a test set;
s2, training a YOLOX target detection model based on the data set, and verifying the detection precision of the trained target detection model through a test set;
s3, training the multi-target tracking model based on the data set, and verifying the tracking precision of the trained multi-target tracking model through a test set;
and S4, inputting the oilfield operation scene video to track the personnel and verify the personnel in real time based on the multi-target tracking model facing the oilfield operation scene.
Preferably, in the oil field operation scene, workers wear the work clothes, the appearances are very similar, and the workers are difficult to distinguish by naked eyes.
Preferably, the image data acquisition sources in S1 include but are not limited to: image data of a well team camera and image data of a handheld camera; selecting images of workers wearing industrial clothes, and removing the images which are small in time interval, small in scene change, low in image quality and unclear;
preferably, the specific method of data labeling in S1 is to label the target coordinate frame by using a data labeling tool, label img, to generate a target detection data set; labeling the target ID based on the target detection data set to generate a multi-target tracking data set;
preferably, the method for obtaining the multi-target tracking model oriented to the similar feature scene in S4 includes the following steps:
s4.1, obtaining the coordinate position of a target frame in an input image through a YOLOX target detection model, constructing a Kalman filtering algorithm to calculate the current predicted position of the target, and determining a candidate frame by using the position relation between the predicted position and a detection frame;
s4.2, converting the candidate frame image into a gray image from a color image, inputting the gray image into a single-channel pedestrian re-recognition network, and extracting features to obtain a feature vector;
s4.3, constructing a distance calculation algorithm, calculating the distance score between the extracted candidate target features and the track features, and distributing the ID to the target through a Hungarian matching algorithm;
preferably, the specific method for determining the candidate frame by using the position relationship between the predicted position and the detection frame in S4.1 is as follows: calculating the intersection ratio of the predicted position and the detection frame, and taking the detection frame as a candidate frame if the intersection ratio is greater than a set intersection ratio threshold;
preferably, the method for converting the color image into the grayscale image in S4.2 is as follows: gray ═ R0.299 + G0.587 + B0.114, where Gray represents the resulting pixel Gray value, and R, G, B represents the values of the three red, green, and blue channels of the pixel, respectively;
the invention discloses the following technical effects:
the invention provides a multiple target tracking method based on a YOLOX target detection algorithm, which is based on an improved pedestrian re-identification network, changes three channels into a single channel, removes the interference of color information on an identification result, trains a YOLOX network model and a pedestrian re-identification model based on a specific oil field operation scene data set, and can realize the real-time tracking of workers on an oil field operation site;
the invention can realize multi-target tracking of a plurality of scenes (oil extraction operation, drilling operation and offshore drilling platform operation) of oilfield operation, adopts a mode of tracking after detection, firstly uses YOLOX to extract a detection frame of a target, and converts the detection frame into a gray image; performing motion prediction on the target by using a Kalman filtering method, and extracting features of the predicted target through a pedestrian re-identification network;
the invention changes the number of pedestrian re-identification network channels from three channels to a single channel, and removes the interference of color information on the identification result; and carrying out similarity comparison on the extracted features and the classification features to obtain similarity scores, and carrying out ID matching through a Hungarian distribution algorithm to obtain tracking results. The method is oriented to the oil field operation scene, the model detection precision is high, the speed is high, and multiple targets can be tracked in real time.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is an overall network flow diagram;
fig. 3 is a diagram of a backbone network structure.
Detailed Description
The present invention is further described below in conjunction with the following drawings, the structure and principle of which will be apparent to those skilled in the art. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Compared with the traditional general multi-target tracking method, the multi-target tracking method for the oilfield operation scene is more beneficial to distinguishing targets with similar characteristics such as appearance colors, and the like, has smaller parameters and calculated amount of a model, and is more suitable for application in the oilfield operation scene.
A multi-target tracking method for oilfield operation scenes comprises the following steps:
(1) acquiring image data in an oilfield operation scene, performing data annotation through data after image preprocessing by using data set annotation software, and dividing a training set and a test set;
(2) training a YOLOX target detection model based on the data set in the step (1), and verifying the detection precision of the trained target detection model through a test set;
(3) training the multi-target tracking model based on the data set in the step (1), and verifying the tracking precision of the trained multi-target tracking model through a test set;
(4) obtaining the coordinate position of a target frame in the input image through the YOLOX target detection model trained in the step (2), constructing a Kalman filtering algorithm to calculate the current predicted position of the target, and determining a candidate frame by using the position relation between the predicted position and the detection frame;
(5) converting the candidate frame image into a gray image from a color image, inputting the gray image into a single-channel pedestrian re-recognition network, and extracting features to obtain a feature vector;
(6) constructing a distance calculation algorithm, calculating the distance score between the extracted candidate target features and the track features, and distributing the ID to the target through a Hungarian matching algorithm;
(7) and (4) inputting an oilfield operation scene video to track the personnel and verify the personnel in real time based on the multi-target tracking model facing the oilfield operation scene obtained in the step (4-6).
Preferably, in the step (3), the performance of the method of the present invention is calculated by using the verification set data, and the main performance indexes include: MOTA (multi-target tracking accuracy), IDF1(IDF1 score), MT (number of tracks tracked), ML (number of tracks lost), FP (false positive), FN (false negative), IDs (number of ID transforms), FPs (number of frames processed per second), and the like;
preferably, the strategy used for matching the candidate frame with the trajectory in the step (6) is secondary matching, and the detailed steps of the secondary matching strategy are as follows
Table 1:
Figure RE-GDA0003796977400000041
TABLE 1 matching strategy flow of candidate frame and trajectory in the invention
The algorithm can effectively reduce the influence of the detection box with low score on the whole matching result in the matching process. As described in Table 1, the Track indices T and Detection indices D are input into the matching algorithm, and the matching result is obtained by the first matching and the second matching. Costmetrixc. Conf is calculated according to features extracted by ReiD network i Is the confidence of the ith Detection, e is the minimum confidence, and the Detection frame with the confidence greater than e is taken as D high And taking the detection box with the confidence coefficient less than the epsilon as D low MinCostMatching is a linear assignment function that assigns a target box to a trace track with a minimum total cost of assignment. Line 6 to Line 8 is a primary matching, and a linear distribution method is used for detecting high resolution D high Track indices T, gate matrix G, to perform minimum cost matching to obtain matching result Match a Unmatched trajectory T unmatched And unmatched detection box D unmatched Line 11 to Line 13 is a secondary match, for the unmatched detection frame D in the primary detection unmatched And low score detection D low Gate matrix G, unmatched track T unmatched Carrying out minimum cost matching to obtain a matching result Match b Match result of the second matching algorithm a +Match b The result of the tracking for that frame.
And after training is finished, performing performance test on the oil field multi-target tracking data set. The MOTA, IDF1, MT, ML, FP, FN, IDs, FPs of the tracking result are calculated. The method is compared with various performance indexes of other mainstream multi-target tracking models and recorded in a table 2. These models include MOTR, CorrTracker, FairMOT, CSTRock + +, TrackFormer, TransMOT, MPNTrack.
The above-described embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solutions of the present invention can be made by those skilled in the art without departing from the spirit of the present invention, and the technical solutions of the present invention are within the scope of the present invention defined by the claims.
Figure RE-GDA0003796977400000051
Table 2 comparison of the multi-target tracking model for oilfield operation scenarios with other mainstream models.

Claims (7)

1. A multi-target tracking method oriented to oilfield operation scenes is characterized in that: the method comprises the following steps:
s1, acquiring image data in an oilfield operation scene, performing data annotation through the data after image preprocessing by using data set annotation software, and dividing a training set and a test set;
s2, training a YOLOX target detection model based on the data set, and verifying the detection precision of the trained target detection model through a test set;
s3, training the multi-target tracking model based on the data set, and verifying the tracking precision of the trained multi-target tracking model through a test set;
and S4, inputting the video of the oilfield operation scene to track the personnel and verify the personnel in real time based on the multi-target tracking model facing the oilfield operation scene.
2. The multi-target tracking method oriented to the oilfield operation scene according to claim 1, wherein the method comprises the following steps: in the oil field operation scene, workers wear the work clothes, the appearances are very similar, and the workers are difficult to distinguish by naked eyes.
3. The multi-target tracking method oriented to the oilfield operation scene according to claim 2, wherein the method comprises the following steps: the image data acquisition sources in S1 include, but are not limited to: image data of a well team camera and image data of a handheld camera; the specific method of image preprocessing is to select the images of workers wearing the industrial clothes, and remove the images with small time interval, little scene change, low image quality and blurry images.
4. The multi-target tracking method oriented to the oilfield operation scene according to claim 3, wherein the method comprises the following steps: the specific method of data annotation in the step S1 is to use a data annotation tool LabelImg to annotate the target coordinate frame to generate a target detection data set; and labeling the target ID based on the target detection data set to generate a multi-target tracking data set.
5. The multi-target tracking method oriented to the oilfield operation scene according to claim 1, wherein the method comprises the following steps: the method for obtaining the multi-target tracking model oriented to the similar characteristic scene in the S4 comprises the following steps:
s4.1, obtaining the coordinate position of a target frame in an input image through a YOLOX target detection model, constructing a Kalman filtering algorithm to calculate the current predicted position of the target, and determining a candidate frame by using the position relation between the predicted position and a detection frame;
s4.2, converting the candidate frame image into a gray image from a color image, inputting the gray image into a single-channel pedestrian re-recognition network, and extracting features to obtain a feature vector;
and S4.3, constructing a distance calculation algorithm, calculating the distance score between the extracted candidate target features and the track features, and distributing the ID to the target through a Hungarian matching algorithm.
6. The multi-target tracking method oriented to the oilfield operation scene according to claim 5, wherein the method comprises the following steps: the specific method for determining the candidate frame by using the position relationship between the predicted position and the detection frame in S4.1 is as follows: and calculating the intersection ratio of the predicted position and the detection frame, and taking the detection frame as a candidate frame if the intersection ratio is greater than a set intersection ratio threshold.
7. The multi-target tracking method oriented to the oilfield operation scene according to claim 5, wherein the method comprises the following steps: the method for converting the color image into the gray image in S4.2 is as follows: where Gray represents the resulting pixel Gray value, and R, G, B represents the values of the three channels of red, green, and blue, respectively, of the pixel, R × 0.299+ G × 0.587+ B × 0.114.
CN202210739370.1A 2022-06-28 2022-06-28 Multi-target tracking method for oilfield operation scene Pending CN115063448A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210739370.1A CN115063448A (en) 2022-06-28 2022-06-28 Multi-target tracking method for oilfield operation scene

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210739370.1A CN115063448A (en) 2022-06-28 2022-06-28 Multi-target tracking method for oilfield operation scene

Publications (1)

Publication Number Publication Date
CN115063448A true CN115063448A (en) 2022-09-16

Family

ID=83201538

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210739370.1A Pending CN115063448A (en) 2022-06-28 2022-06-28 Multi-target tracking method for oilfield operation scene

Country Status (1)

Country Link
CN (1) CN115063448A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115937498A (en) * 2023-03-14 2023-04-07 天津所托瑞安汽车科技有限公司 Target detection method and device and electronic equipment

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115937498A (en) * 2023-03-14 2023-04-07 天津所托瑞安汽车科技有限公司 Target detection method and device and electronic equipment

Similar Documents

Publication Publication Date Title
CN105224912B (en) Video pedestrian's detect and track method based on movable information and Track association
CN104134222B (en) Traffic flow monitoring image detecting and tracking system and method based on multi-feature fusion
US7787656B2 (en) Method for counting people passing through a gate
CN102521565B (en) Garment identification method and system for low-resolution video
CN104915655A (en) Multi-path monitor video management method and device
Li et al. Robust people counting in video surveillance: Dataset and system
CN105023278A (en) Movable target tracking method and system based on optical flow approach
CN107145862A (en) A kind of multiple features matching multi-object tracking method based on Hough forest
CN104298964B (en) A kind of human body behavior act method for quickly identifying and device
CN106682573B (en) A kind of pedestrian tracting method of single camera
CN112669350A (en) Adaptive feature fusion intelligent substation human body target tracking method
CN105654139A (en) Real-time online multi-target tracking method adopting temporal dynamic appearance model
CN111968152B (en) Dynamic identity recognition method and device
CN104992453A (en) Target tracking method under complicated background based on extreme learning machine
CN110991315A (en) Method for detecting wearing state of safety helmet in real time based on deep learning
CN102568003A (en) Multi-camera target tracking method based on video structural description
CN108830204B (en) Method for detecting abnormality in target-oriented surveillance video
De Langis et al. Realtime multi-diver tracking and re-identification for underwater human-robot collaboration
CN104301585A (en) Method for detecting specific kind objective in movement scene in real time
CN112132873A (en) Multi-lens pedestrian recognition and tracking based on computer vision
Yamamoto et al. Multiple players tracking and identification using group detection and player number recognition in sports video
CN115063448A (en) Multi-target tracking method for oilfield operation scene
CN111402289A (en) Crowd performance error detection method based on deep learning
CN102004795A (en) Hand language searching method
Alagarsamy et al. Identifying the Missing People using Deep Learning Method

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