GPU-based unmanned aerial vehicle image feature point extraction method

Abstract

The embodiment of the invention provides a GPU-based unmanned aerial vehicle image feature point extraction method and device, equipment and a computer readable storage medium. The method comprises the steps: uploading an original unmanned aerial vehicle image and a Gaussian filtering function to a GPU through a CPU; preprocessing the original unmanned aerial vehicle image through a GPU by adopting the Gaussian filter function, and generating a Gaussian pyramid image; searching the Gaussian pyramid image by adopting a preset algorithm to obtain angular point feature points, and reading the Gaussian pyramid image back to the CPU; performing layer subtraction on the Gaussian pyramid image through a CPU to generate a Gaussian difference pyramid, and transmitting the Gaussian difference pyramid to a GPU; carrying out spot feature point extraction on the Gaussian difference pyramid through a GPU, and obtaining spot feature points; and completing feature point extraction of the unmanned aerial vehicle image. In this way, the speed of extracting the feature points is increased.

Description

technical field [0001] Embodiments of the present disclosure generally relate to the field of image processing, and more specifically, relate to a GPU-based method, device, device, and computer-readable storage medium for extracting feature points from UAV images. Background technique [0002] With the widespread application of UAVs in various industries, the technology of 3D information reconstruction based on UAV images is becoming more and more mature. UAVs acquire images with small image size, high image overlap rate, high resolution, and large amount of image information. , with a wide range of action, it is inevitable that there will be partial occlusion of ground objects, and when acquiring images, the number of images in a survey area is large, and when flying over a large area, multiple flights are required to obtain data, which will As a result, the lighting conditions of different sorties vary greatly. In this case, conventional feature point extraction methods f...

AI Technical Summary

Problems solved by technology

[0002] With the widespread application of UAVs in various industries, the technology of 3D information reconstruction based on UAV images is becoming more and more mature. UAVs acquire images with small image size, high image overlap rate, high resolution, and large amount of image information. , with a wide range of action, it is inevitable that there will be partial occlusion of ground objects, and when acquiring images, the number of images in a survey area is large, and when flying over a large area, multiple flights are required to obtain data, which will Leading to large differences in lighting conditions during different sorties

In this case, conventional feature point extraction methods are faced with the problems of large amount of calculation, slow speed, and limited point extraction. At the same time, these algorithms are mainly based on CPU for calculation, and the extraction speed is slower than that of GPU.

Images