Matching matrix image matching method in unmanned vehicle monocular vision positioning

Abstract

The invention relates to a matching matrix image matching method in unmanned vehicle monocular vision positioning. The method comprises the following steps: 1) through a deep convolutional neural network (DCNN), all inputted test images and positioning map images are subjected to global feature description, and a third convolutional layer is extracted as a feature vector; 2) a principal component analysis method is adopted to carry out dimension reduction on all feature vectors; 3) according to the feature vectors after dimension reduction of the test images and the positioning map images, a matching matrix, that is, a matching image, is constructed; 4) the matching image is subjected to OTSU binaryzation processing, and a binaryzation image after processing is obtained; 5) after the binaryzation image is subjected to morphological processing, linear fitting is carried out on the image; and 6) the corresponding test image and the positioning map image on the fit line are the corresponding matched images. Compared with the prior art, the method has the advantages that the algorithm is designed easily, the visualization effects are good, the calculation speed is quicker, hardware requirements are reduced and the like.

Description

technical field [0001] The invention relates to the field of unmanned vehicle positioning, in particular to an image matching method for matching matrices in unmanned vehicle monocular vision positioning. Background technique [0002] Most of the existing technologies have the following two problems: [0003] 1. Most technologies do not perform data compression processing on image features, but directly use a certain layer of deep network features (such as the third convolutional layer), which requires a huge amount of data calculation, takes a long time and requires high hardware requirements. [0004] 2. Some technologies use the method of direct image compression to reduce the amount of computation. For example, the SeqSLAM algorithm does not use the deep learning network to extract features but directly compresses the image by 64*32 or 32*24. There is no screening of information in this process , but to intercept a certain area, it is easy to miss important information....

AI Technical Summary

Problems solved by technology

[0003] 1. Most technologies do not perform data compression processing on image features, but directly use a certain layer of deep network features (such as the third convolutional layer), which has a huge amount of data calculation, takes a long time and requires high hardware requirements

[0004] 2. Some technologies use the method of direct image compression to reduce the amount of computation. For example, the SeqSLAM algorithm does not use the deep learning network to extract features but directly compresses the image by 64*32 or 32*24. There is no screening of information in this process , but to intercept a certain area, it is easy to miss important information

Images