MFNN: Position and Attitude Measurement Neural Network Based on Multi-Feature Fusion

Abstract

Accurate measurement of position and attitude information has become the basis of application engineering in related fields. Traditional position and attitude measurement schemes, including lasers, infrared light, etc., often place high demands on the measurement environment. The visual image-based measurement scheme needs to be combined with the actual situation to establish a complex pose calculation model. Therefore, exploring a measurement scheme for efficient solution has become an urgent need in related fields. In this paper, we propose a multi-feature fusion position and attitude measurement neural network (MFNN). We effectively introduce the HOG operator as an image feature and combine it with the original image for a specific ratio. At the same time, we draw on the utilization strategy of feature points in traditional measurement methods. By combining the extracted image feature points with the network, the measurement error of the six-dimensional information is effectively reduced. Based on this, we propose a new type of position and attitude measurement network architecture. MFNN has further improved the accuracy of related measurements.