SLAM Loop Detection Algorithm Based on Deep Neural Network

Abstract

Loop detection is an important part of the visual SLAM system, which can reduce the cumulative error generated by the visual odometer during the positioning process and construct a globally consistent environmental map. The paper proposes a loop detection method based on deep learning neural network. The loop detection module is used to reduce the accumulation of errors between adjacent frames, overcome the shortcomings of the traditional visual SLAM loop detection based on the artificial mark feature point algorithm, and improve the detection accuracy of the system in a complex environment. Experiments show that the algorithm has good accuracy and speed, which can meet the requirements of the visual SLAM system. Deep neural networks and their learning algorithms, as successful big data analysis methods, have been well-known in academia and industry. Compared with traditional methods, deep learning methods are data-driven and can automatically extract features from data. They have significant advantages for analyzing unstructured, unclear and changeable patterns, and cross-domain big data.