Method For Fast Detection And Pose Estimation Of Underwater Target

With the continuous improvement of the requirements for the efficiency of underwater mine operation and salvage,more and more research is invested in the use of mechanical automation equipment to replace humans in completing underwater salvage tasks.Computer vision puts eyes on mechanical automation equipment,and its low cost and versatility make it more and more widely used in industrial production.Due to the complexity of the underwater environment,the color difference between the target and the background is small,the environment affects the color distribution of the target,and it is difficult to detect the color of underwater targets;the loss of features leads to unsatisfactory disparity maps,which affects the accuracy and efficiency of pose estimation.In response to the above problems,this article first studies the underwater image enhancement problem,and then studies the fast color tracking and pose estimation problems.The details are as follows:First of all,to solve the problem that the color feature of the target in the underwater image is weakened or even lost due to the absorption of light by water,a dual gamma transform is proposed to enhance the underwater image.The light attenuates exponentially in the medium.Based on the gamma transformation with exponential repair properties,the calculation of the gamma factor is combined with the idea of white balance to realize the restoration of underwater images;for the problem of feature loss,the use of pixels and he contrast relationship between the domain images widens the gap and realizes the feature enhancement in the image.Secondly,in view of the problem that the target color range is constantly changing due to environmental influences in underwater target detection,a strategy of self-correction of color extraction range is proposed.By continuously detecting the color of the identified point in the target area,comparing it with the target and the background,determining the color value of the new target point and then determining the new color extraction range,realizing the use of color to track the underwater target;combined with the depth of the subject,camera parameters and other information,to achieve the prediction of the underwater target object motion range,reduce the area of the target detection area,thereby reducing the target detection time.Then,in view of the time-consuming problem of using point clouds for underwater target detection,the feature point-based pose estimation method is used,after the disparity map is obtained by using the semi-global stereo matching algorithm,combined with the known identification position,the specific point field is extracted Parallax,calculate the three-dimensional information of a specific point;filter the three-dimensional points to form a vector,combine the parallax error and the depth relationship,and give different vectors different weights to obtain the target pose.Through on-land experiments and comparing point cloud data,the detection speed can be effectively improved under similar accuracy conditions.Finally,by comparing the color difference between the background and the target after the underwater enhancement algorithm,and the stereo matching before and after the image enhancement,the effect of the disparity map is compared to verify the superiority and feasibility of the underwater image enhancement algorithm,the overall experiment is compared with the required indicators to verify that the underwater target detection and pose estimation meet the time and accuracy requirements.