1. An improved Visual Indoor Navigation Method based on Fully convolutional Neural Network
- Author
-
Li Chenning, Wen Chengqing, and Xu Haowei
- Subjects
Matching (graph theory) ,Position (vector) ,business.industry ,Computer science ,Template matching ,Cognitive neuroscience of visual object recognition ,Computer vision ,Artificial intelligence ,business ,Object (computer science) ,Fuzzy logic ,Convolutional neural network ,Image (mathematics) - Abstract
The indoor positioning algorithm based on visual object recognition is a new kind of indoor positioning solution, in which, user’s position can be determined by steps as identifying objects around, matching general positions by a digital map and precise positioning through nonlinear calculation. However, limited by low objects detection rate and low detection accuracy, the availability of the localization algorithm is greatly affected. Therefore, this paper presents a novel method that utilizes Fully Convolutional Networks (FCN), which brings a great improvement in contour recognition accuracy when identifying objects in an image, and thus improving the final positioning results. In additional, a pre-trained CNN (Convolution Neural Network) is also proposed in the paper, which is used to get a coarse location for solving the problem of fuzzy floor and location brought by similar object distribution in the existence algorithm. Verifying by experimental tests, The results show that compared with the previous TM-based (template matching) and the R-CNN based localization algorithm (Region-based Convolution Neural Network), the methods proposed in this paper can effectively improve the success rate from 85% to 93% and the positioning error drop from 2.6m to 1.4m of the indoor positioning algorithm.
- Published
- 2020