Visual localization technology of AGV based on global sparse map

In order to realize the high-precision localization of automated guided vehicle (AGV) in complex industrial environment and overcome the influence of environment change, a vision localization method based on a global sparse map was proposed. First, a large-capacity two-dimensional coded point was designed, which was set on the ground as an artificial landmark. Based on a quad recognition algorithm, the coded points were accurately segmented and identified in complex industrial environment. The feature points from different images were properly matched by using the coded information provided by coded points. Then, a block-optimization three-dimensional reconstruction algorithm was designed to build a map for a large-scale industrial environment, which provided a sparse electronic map for AGV visual localization. The visual localization of AGV was realized by matching the feature points from the visual sensor and sparse electronic maps. The repeated precision of AGV is less than 0.5 mm, the angle deviation is less than 0.5°, and the average displacement error of trajectory is less than 0.1%. The practical application shows that the method can realize the visual localization of AGV in complex industrial environment. The speed and precision of localization both meet the requirements of industrial application, which provides a new way for vision-based localization of AGV. © 2019, Editorial Board of JBUAA. All right reserved.