A registration method of point clouds with low coincidence degree

Abstract

The invention discloses a low coincidence degree point cloud registration method, which can complete the point cloud registration task in the low coincidence degree scene. Aiming at the problem that it is difficult to search for point pairs with the same name in low-coincidence scenarios, the self-attention mechanism is used to make the aggregation point realize the overall perception of the overall point cloud, and at the same time, the cross-attention mechanism is used to explicitly mine the overlapping area information to predict all points in the point cloud Confidence in overlapping regions, using probabilistic selection to sample point pairs in the matching stage within overlapping regions, improving the recall of the registration. At the same time, the receptive field of the convolution kernel is dynamically limited in the overlapping area, which avoids the extraction of invalid geometric neighborhood information and improves the precision and accuracy of point-by-point features.

Description

technical field [0001] The invention belongs to the fields of artificial intelligence, control science and engineering technology, and in particular relates to a point cloud registration method with low coincidence degree. Background technique [0002] Single-photon imaging is a new type of technology that uses pulsed laser light sources and nonlinear optical technologies such as wavelength division multiplexing to achieve ultra-high-precision and strong anti-noise imaging. Compared with traditional lidar systems, single-photon radar systems have significant advantages in long-distance and harsh environmental noise, and have great application potential in weak signal detection, long-distance imaging, and precision measurement. [0003] In 3D imaging, in order to obtain a complete 3D model of a real-world object or scene, it is usually necessary to use a single-photon radar system to collect multiple point cloud data of the target object point cloud at different spatial angle...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the shortcomings of the above-mentioned existing technologies, the present invention proposes a low-coincidence point cloud registration method, which is oriented to single-photon radar detection systems, and solves the problem of small number of point pairs with the same name and poor registration accuracy in low-coincidence scenarios. problem, explicitly mining overlapping area information, improving the feature matching accuracy and registration recall rate of point cloud registration in low-coincidence scenarios

Images