TOF-assisted Method For Improving The Detection Accuracy Of Underwater Fish Size With Binocular Vision

In aquaculture fisheries,growth situation information such as fish size is of great significance to the management of aquatic products.The size of the fish and its change information not only reflect its own growth status,but also reflect the problems existing in the aquaculture process.The traditional fish size measurement method is to manually measure the fish with a ruler and other tools after taking it out.This method is not only time-consuming and labor-intensive,but also causes certain damage to the fish itself.With the continuous expansion of intensive production and aquaculture,the drawbacks of traditional size measurement methods are increasingly exposed.How to obtain growth information such as the size of aquaculture fish efficiently,accurately and in situ has become an urgent problem to be solved.This paper takes the green-finned pufferfish as the experimental object,based on the binocular vision theory,and adopts the binocular vision detection algorithm assisted by TOF(Time of Flight)to realize the underwater fish size based on computer vision in aquaculture environment/detection.The main research contents are as follows:(1)Introduce the basic principle and basic process of binocular vision technology,the imaging principle of TOF depth camera,analyze the advantages and defects of single binocular vision technology or TOF technology in 3D measurement,find out the texture area and occlusion area The difference between the two measurement accuracy.(2)The structure of the underwater vision system is designed,mainly including the selection of the pressure-resistant housing,the optical lens,and the design of the sealing scheme.The experiment proves that the structure can meet the requirements of sealing and compression resistance.By analyzing the axis-parallel binocular ranging system,its measurement error transfer function is deduced.The value range of each parameter of the optical system finally determines the parameters of the underwater binocular vision system.(3)Aiming at the shortcomings of a single binocular vision or TOF depth camera in 3D measurement of dynamic fish bodies,a TOF-assisted binocular vision depth acquisition algorithm is used to complement the advantages of the two and obtain high-quality depth.picture.The algorithm guides the binocular stereo matching process by designing the TOF depth energy function,and improves the matching accuracy of the binocular matching in the low texture area and the repeated texture area.In other areas of the picture.TOF and binocular stereo matching confidence weight functions are designed respectively to achieve pixel-level fusion of the two,so as to obtain a higher-precision depth map of the underwater scene.From the results of boxplots and scatterplots,the performance of TOF-assisted binocular vision is better.In order to obtain a high-quality depth map,this paper adopts the TOF-assisted binocular stereo matching algorithm to realize the in-situ,accurate and efficient detection of underwater dynamic fish.By designing the underwater optical structure,the in-situ collection of underwater fish photos is realized.