Research On Indoor Lightweight Fingerprint Localization Method Based On Spatial Visual Feature Descriptor

With the development of society and the progress of science and technology,indoor service robots have deeply entered people's daily life.Accurate,stable and efficient indoor localization technology is a powerful guarantee for them to provide services indoors.Indoor environment is complex and wireless signal is influenced by environment.Visual information is an ideal location information source because of its richness and not limited by hardware network.How to ensure the accuracy,stability and real-time in localization has become a hot topic in the indoor localization technology based on visual information.The low dimension feature descriptor extraction algorithm and lightweight fingerprint database construction algorithm is studied to avoid the reduction of positioning stability,real-time performance and accuracy from uneven distribution and high dimension of visual features,large scale of fingerprint database and information redundancy.A feature descriptor extraction algorithm with even distribution and low dimension and a lightweight visual feature fingerprint database construction algorithm are proposed.A fingerprint localization algorithm based on lightweight visual feature fingerprint database is proposed,which effectively improve the positioning speed on the premise of ensuring the positioning accuracy and stability.The specific research work is as follows:1.Aiming at the problems of the stability and speed reduction in feature fingerprint matching caused by the centralized distribution of visual features and the high dimensions of descriptors,the Dual-screening Adaptive Low-dimensional Binary Descriptor(DALBID)with even distribution and low dimension is proposed.In feature extraction,the adaptive threshold in local pixel neighborhood is used instead of manually setting threshold.After feature extraction,weak feature points are filtered and removed as well as unstable noise points located at the edge.In feature description,binary pixel information in the neighborhood around feature points is binarized.The DALBID reduces the number of features while avoiding feature concentration distribution and reducing the number of features.This algorithm ensures the stability of the positioning results from the feature distribution,and improves the matching speed of the fingerprint from the two aspects of the number of features and the dimension of descriptor.2.In order to solve the problem of positioning stability and accuracy reduction caused by unreasonable distribution of fingerprint points and redundancy of fingerprint information when constructing the visual information fingerprint database,a lightweight visual feature fingerprint database construction algorithm is proposed.The algorithm starts with the extraction process of visual features,calculates the scale invariance effective range of the feature descriptor according to the scale pyramid used.Then the reasonable interval of fingerprint point distribution in indoor space is calculated from the scale invariance.The Adaptive Iterative K-means Clustering Algorithm(AIK-means)is proposed after analyzing the sources of redundant information in fingerprint database.Redundant information is removed by feature clustering,and a lightweight visual feature fingerprint database is constructed.This algorithm reduces the scale of the fingerprint library from the two aspects of fingerprint point distribution and fingerprint database information.3.To solve the low positioning efficiency which affected by the efficiency of fingerprint matching,a fast retrieval strategy based on lightweight visual feature fingerprint database is proposed.Based on the lightweight fingerprint database,the mapping between the fingerprint and the coordinate axis of the fingerprint point is established.The location of the point to be located is quickly obtained by coordinate retrieval.In this paper,the fingerprint location algorithm based on lightweight visual feature fingerprint library is also completed.The matching number of feature fingerprint on coordinate axis is weighted as the confidence to improve the positioning speed effectively.Through experimental test and evaluation,the DALBID visual feature descriptor proposed in this paper is distributed sparsly in different positions in indoor space,and is good in scale invariance,robust to illumination and image noise.The features are stable and specific,few in quantity and low in dimension,which effectively improves the location stability and location speed.The lightweight fingerprint database construction algorithm reduces the number of fingerprint points.The size of fingerprint database is reduced,and the fingerprint data is more concise.The fast retrieval strategy of fingerprint database further improves the positioning speed.In the scene of comprehensive office,the average positioning error is 0.76 m and the standard deviation is 0.41 m.What's more,because of the low dimension of fingerprint and small scale of fingerprint database,the speed of fingerprint location method is increased by 60%compared with the location method using sift visual feature.Compared with the location method using orb visual feature as location fingerprint,the speed is increased by about 40%.