Visual SLAM closed-loop detection method based on convolutional neural network and VLAD

Abstract

The invention discloses a visual SLAM closed-loop detection method based on a convolutional neural network and VLAD and the method comprises the following steps: cutting a VGG16 network, adding a pooling layer NetVLAD based on a VLAD idea to the last layer, and constructing a new network model VGG-NetVLAD; training parameters of a network model by using the large data set containing the triple; inputting the current query image into the VGG-NetVLAD, and extracting the output of the NetVLAD layer as the feature expression of the image; adopting a cosine distance as a standard for measuring thesimilarity between the images, and calculating similarity scores of the currently queried image and other images; counting scores between every two images to finally form a similarity matrix; and judging the generated closed-loop area according to the threshold value, and outputting an accuracy recall rate curve. The method considers that the local space characteristics of the image and the characteristics of the traditional artificial design are easily influenced by the environmental change, effectively improves the accuracy and recall rate of closed-loop detection, meets the real-time requirement, and has important significance for constructing a globally consistent map.

Description

technical field [0001] The invention relates to the fields of image processing, deep learning, simultaneous visual positioning and map construction, and in particular to a closed-loop detection method based on convolutional neural network and VLAD. Background technique [0002] In recent years, loop closure detection has become a key issue and research hotspot in the field of mobile robot navigation, especially on the issue of Visual Simultaneous Localization and Mapping (VSLAM). Visual SLAM is mainly composed of four parts: visual odometry, back-end graph optimization, loop closure detection, and mapping. Among them, closed-loop detection is also called position recognition, which means that the robot uses the image provided by the visual sensor to identify whether it has passed the previously reached position during the navigation process. Assuming that the camera captures two images at the current moment and an earlier moment, the task of closure detection is to judge wh...

AI Technical Summary

Problems solved by technology

However, these methods use artificially designed low-level features, which are sensitive to factors such as illumination and weather, and thus lack the necessary robustness.

Images