Three-Dimensional Reconstruction and Analysis of Target Laser Point Cloud Data Under Simulated Real Water Environment

Objective Underwater target detection has important applications in underwater accident search and rescue, equipment maintenance, and resource exploration. Presently, underwater acoustic detection is the most common detection method, but it has limitations such as low resolution and blurred target edges, making it difficult to effectively identify the target. The underwater optical detection system perfectly compensates for the lack of underwater acoustic detection, However, physical characteristics such as salinity, temperature, and turbidity frequently affect the real ocean water environment. The underwater optical detection system's working environment is sometimes harsh, which has an impact on the system. In addition, conducting underwater detection experiments directly in the ocean is difficult. As a result, it is important to regulate changes in the physical characteristics of various water bodies in the laboratory and study their impact on detection. The findings have a substantial impact on the design and development of related equipment. In this paper, two typical sea water environments are simulated through experiments, and the effects of changes in the physical properties of salinity and turbidity and the detection distance on the three-dimensional (3D) reconstruction of the laser point cloud of underwater submarines, gliders, and anchor mines are being researched. This study can provide a reference for the development of obstacle avoidance and detection modules for small underwater unmanned gliders. Methods We used lidar to collect the original point cloud data of targets in different water environments in the experimental glass tank. The original point cloud data obtained from the experiment was then processed using threshold segmentation, refraction correction, and point cloud denoising algorithms ( Figs. 3-4) , and a three-dimensional point cloud image containing only the target object was obtained ( Fig. 5) . In the original point cloud data, the threshold segmentation separated the target point cloud and the backscattered noise point. Refraction correction corrected the influence of refraction when crossing the medium by correcting the parameters in the spherical coordinate system. To filter out the noise points in the point cloud data, we first calculated the average distance from all points to their neighbors by point cloud denoising method and then set the filtering threshold based on the obtained value. Finally, we mounted the 3Dmax-processed standard point cloud data to the k-d tree and query the minimum distance between all points in the reconstructed point cloud data and the generated k-d tree, and save the closest point of all points in the reconstructed point cloud. The output was used to show the error between the reconstructed point cloud and the standard point cloud. Results and Discussions A set of point cloud reconstruction algorithms is designed in this paper (Fig. 3) , and three-dimensional reconstruction point cloud images of submarines, underwater gliders, and anchor mines in various water environments are obtained. The influence of water body salinity change on the point cloud reconstruction is summarized and the data are displayed in Table 2. The results show that salinity changes have a minor impact on point cloud reconstruction. The reconstructed point cloud images of three targets exposed to turbidity changes are listed in Table 3. The mean square error and effective points between the reconstructed and standard point clouds vary with turbidity (Fig. 6). As the turbidity increases, the number of effective points shows a downward trend, and the mean error shows an upward trend. In a low turbidity environment, raising the filter threshold can help reduce mean error. The detection distance has an impact on the reconstruction of the point cloud as well. The lidar' s field of view is constant. As the detection distance increases, the proportion of the target's area occupied in the entire detection plane decreases, resulting in a decrease in the number of reflected beams received, as well as a decrease in the number of effective points for reconstructing the point cloud ( Fig. 7). The increase in the mean error is because the backscattering effect of the water body increases with the increase of the detection distance, which affects the detection accuracy. Conclusions The two real water environments of the Bohai Sea and the South China Sea are simulated in this study by numerically setting the two parameters of salinity and turbidity, and experiments are conducted using this as the medium to obtain the original point cloud data, using the methods of threshold segmentation, refraction correction, and point cloud filtering for processing and reconstruction. In the follow-up error analysis, the k-d tree algorithm is used to calculate the error between the reconstructed point cloud and the standard point cloud, and the cause and impact are analyzed. The results show that: 1) the detection of underwater targets is unaffected by changes in salinity. The average error of the target changes within 1 mm in water environments of 30 PSU and 35 PSU. 2) Changes in water turbidity primarily affect light backscattering in the underwater transmission process, resulting in more noise points in the target point cloud and lowering imaging quality. 3) When the target's detection distance is fixedly increased, the detection effect gradually deteriorates due to the large attenuation of the water body, and the target point contour of the cloud becomes blurred, The mean error effect increases as the number of effective point clouds decreases. In conclusion, the turbidity influence is dominant in point cloud reconstruction under different water environments, the regularity is not obvious, and the detection distance influence is controllable.