Research On Multiple Oil Drop Target Tracking Method Based On Machine Vision

With the development of computer hardware and deep learning algorithms,machine vision technology is gradually integrated with other fields.At present,the recovery of oil fields in China presents a high water content and low flow rate state,and the oil phase often exists in the form of oil bubbles.The current measurement methods mostly focus on obtaining the overall flow velocity and flow field of oil-water two-phase flow,and cannot obtain the motion law of oil droplets in the oil-water twophase flow.Therefore,it is necessary to use machine vision technology to achieve multi oil droplet target tracking.Firstly,detecting multiple oil droplet targets in oil-water two-phase flow is a prerequisite for tracking multiple oil droplet targets using machine vision technology.Due to the low grayscale differentiation of oil droplets in the background of oil-water two-phase flow and differences in image quality,existing object detection features cannot accurately detect multiple oil droplet targets in oil-water two-phase flow.This paper proposes the construction of integrated features(ACFHG),which achieves accurate characterization of multiple oil droplet targets in oil-water two-phase flow by integrating improved local gravity features with aggregated channel features,Utilizing the Adaboost algorithm to train and obtain integrated feature detectors and other feature detectors to compare and analyze the detection performance of multiple oil droplet targets in actual oil-water two-phase flow images.Secondly,the GM-PHD filter tracking algorithm based on clustering analysis is used to achieve multi oil droplet target tracking of oil-water two-phase flow.In response to the problem of poor tracking performance of the GM-PHD filter algorithm in noisy environments,this article uses an improved elbow method to obtain the optimal number of clusters in the measurement set,and analyzes each point in the measurement set based on the optimal number of clusters to determine and eliminate noise points.After removing noise points,the measurement set uses the GM-PHD filter algorithm to achieve multi oil droplet target tracking,Finally,compare the multi target tracking performance of the GM-PHD filter algorithm based on clustering analysis with the traditional GM-PHD filter algorithm under the same simulation conditions.Finally,repair the flow field obtained by traditional PIV/PTV.The PIV/PTV flow field can provide reference for tracking multiple oil droplet targets.When using the PIV/PTV algorithm for flow field acquisition,due to the inherent characteristics of oilwater two-phase flow,obtaining vector maps often has a moderate amount of blank space and errors.This paper uses inpainting technology to repair the blank vector and error vector in the flow field vector diagram obtained by PIV using the Criminisi algorithm.It improves the recognition ability of edges and textures in the repair process by improving the calculation formula of the defect edge priority of the Criminisi algorithm,and improves the repair accuracy of the Criminisi algorithm by improving the matching principle.Finally,the improved Criminisi algorithm will be simulated and validated with the traditional Criminisi algorithm to compare the repair effect on vector images.