Image Based Localization Algorithm Using Similarity Measurements and Back-Propagation Neural Network

Abstract

In this paper, an image based localization technique is proposed by similarity measurements and back-propagation neural network (BPNN). First, the bilinear interpolation method is used for size normalization of the obtained images. Then, the relative distance is calculated as the label of each collection position. The image similarity measurements, such as cosine, Hist and SSIM similarity are chosen as the fingerprint of the collection position. At last, the BPNN is utilized for regression learning and obtain the distance based regression function. Field tests show that the proposed algorithm can obtain more accurate position estimation than other existing image based localization approaches.