In the vision-guided robot operation, it is necessary to acquire the 3D image of the target in the scene dynamically when there is relative motion between the target and the vision system. Binocular stereo vision can obtain images of the left and right cameras synchronously and quickly and has better dynamic adaptability compared with the 3D imaging methods characterized by scanning, such as laser radar and linear scanning structured light technology. The binocular stereo vision technology assisted by active speckle projection illumination enhances the texture information of the target surface and improves the matching accuracy of corresponding points of the left and right images. Thus, it is a simple and effective approach. At present, the research on the binocular stereo vision system with active speckle projection applied in the air is mature relatively, and many commercial products have been developed, such as the Kinect system of Microsoft, the Vic-3D measurement system of Correlation Solutions, and the Q-400 system of Dantec in Germany. However, when this technology is applied in the water, there are problems such as the pinhole model failure, unsatisfied matching conditions of polar constraints, and the image degradation caused by the underwater environment absorption and scattering of left and right images of the projected speckle. This will affect the matching accuracy of the corresponding points and the underwater 3D imaging effect.

We rebuild an underwater binocular vision imaging model that actively projects speckle patterns based on the 4D parameter representation of light. The influence of the speckle pattern on the matching accuracy of underwater binocular corresponding points is analyzed based on MATLAB 2015b. The experimental device of underwater binocular vision dynamic 3D imaging system with active speckle projection is mainly composed of a speckle pattern projector and two cameras. The employed speckle pattern projector is a projector. The speckle pattern generated by the computer is projected on the underwater target by the projector, and the left and right cameras synchronously and quickly shoot the underwater moving object with the speckle pattern on the surface. According to the principle of binocular stereo vision, the 3D image of the underwater target is calculated.

The simulation results of the relationship between speckle pattern and matching accuracy of underwater binocular imaging are shown in Fig. 6. With the rising speckle size, the maximum matching error first decreases and then increases. When the speckle size is between 3 pixel and 15 pixel, the maximum matching error is less than 0.7 pixel, and when the speckle size is 9 pixel, the matching accuracy is the highest. This is because when the speckle density is constant, too large or too small speckle is not conducive to matching, and the appropriate size of the speckle can enhance the matching clues of the corresponding points of the left and right images. Fig. 7 shows the influence of speckle density on the matching accuracy when the speckle size is 9 pixel. With the increase in speckle density, the maximum matching error first decreases and then increases. When the speckle density is 1.5% to 3.5%, the matching error is less than 0.6 pixel, and when the speckle density is 2%, the matching accuracy is the highest. Fig. 8 presents that the maximum matching error gradually increases with the decreasing object distance (from 2400 mm to 3600 mm), but the maximum matching error is still less than 1.1 pixel. This method has high matching accuracy. In addition, the purpose of the underwater experiment is to investigate the underwater dynamic 3D imaging error of the established experimental device. The experimental scenario is shown in Fig. 10. The projector is connected to the computer, the speckle density generated by the computer is 2%, and the speckle size is 9 pixel. The projector projects the speckle pattern diagonally downward on the underwater target surface. The underwater target is a standard ball that is suspended by a string at about 3 m in front of the experimental device. As shown in Fig. 11, the standard ball swings on the parallel plane of two cameras, the starting angle is about 50°, and the ball can swing freely when it is released at zero initial speed. It has different instantaneous linear speeds at different positions and has the maximum instantaneous speed at the lowest point. The maximum instantaneous speed is about 1.2 m/s. The 3D point cloud can be calculated based on the captured left and right images. Fig. 13 shows the 3D point cloud images at the corresponding positions in Fig. 12. The PolyWorks software is adopted to fit the obtained 3D point cloud into a sphere (picture-in-picture in Fig. 13). The diameter of the sphere and coordinates of the center of the sphere are obtained (data in the upper right corner of picture-in-picture), and the dynamic measurement error is obtained by comparing the diameter of the standard sphere. The dynamic imaging experiment has been operated on many times. The diameter error of the standard ball at the lowest point and the standard deviation of the measured result are shown in Table 3. The experimental results show that the standard deviation of dynamic measurement error at the maximum instantaneous speed of the standard ball is 2.4 mm with a sound dynamic 3D imaging effect.

We conduct a study on the underwater binocular stereo vision dynamic imaging technology based on the active speckle projection, analyze the influence of active speckle pattern projection on the matching accuracy of corresponding points of underwater binocular stereo vision, and establish the experimental device of underwater binocular vision dynamic 3D imaging system based on active speckle projection. The experimental results indicate that the underwater binocular stereo vision technology with the active speckle projection has a sound dynamic 3D imaging effect, and the dynamic measurement error is within the static error determined by the structure and system parameters of the binocular stereo vision experimental device.