Research On Key Technologies Of Indoor Location Based On WLAN Hybrid Dual Radio Frequency Fingerprint

With the widespread popularization of mobile intelligent terminals and wireless networks,the application demands for location-based services have shown a rapid growth trend.As the basic technique of location services,indoor positioning is the key to addressing the failure of global satellite navigation system to provide users with accurate indoor location services effectively.Among many indoor positioning techniques,the WLAN RF fingerprint-based technique is considered one of the most potential techniques owing to its advantages in cost,complexity and universality.However,in the case of prolonged deployment in the indoor environment,the conventional WLAN RF fingerprint-based positioning systems still face many difficulties and challenges,such as fuzzy fingerprint representation,high fingerprint database building cost and low position estimation accuracy,making it imperative to carry out in-depth research.To this end,taking the actual positioning scenario as the point of departure,this study explores key issues around the fingerprint database building strategy and expansion model,as well as the positioning algorithm based on the current location services demands for positioning performance.The major research tasks are as follows:(1)Due to serious multipath effects and interferences from other co-frequency devices in the indoor environment,the conventional WLAN 2.4G RF fingerprint features are unable to represent the position information accurately,thereby affecting the positioning accuracy.Additionally,the conventional strategies for fingerprint database building are time-consuming and labor-intensive,which restrict the ubiquitous application of WLAN RF fingerprint-based positioning technique.To this end,this study proposes a strategy for building a hybrid dual RF fingerprint database.Initially,exploiting the concept that building such a fingerprint database can effectively improve the differentiation of fingerprint features at different reference points,the originally independent WLAN 2.4G and 5G RF fingerprints are fused,and the characterization of RF fingerprint features at different locations is improved by constructing highdimensional fingerprint features.Next,a set of automatic fingerprint acquisition measures is designed for reducing the cost of fingerprint database collection.Finally,considering the influence of indoor environmental noise on fingerprint acquisition,the automatically collected fingerprint data are preprocessed via the proposed iterative recursive weighted filter and high-dimensional spatial vector alignment mechanism,thereby improving the accuracy of the collected fingerprint data.The results show that the proposed hybrid dual RF fingerprint database building strategy is effective in improving the system positioning accuracy and reducing the cost of fingerprint database construction without requiring any manual involvement or manually annotated data.(2)During the prolonged indoor deployment,the positioning systems are easily affected by environmental factors,showing prominent amplitude fluctuations of RF fingerprint RSSI.This leads to deviation of the real-time acquired RSSI in the online phase from the initially built fingerprint database,so that the positioning systems cannot provide stable and reliable positioning services for a prolonged time.Addressing this problem,this study proposes a hybrid dual RF fingerprint database-based adaptive expansion model.Initially,through quantitative RSSI measurement in the real environment taking into account the significant and controllable indoor climatic factors,an important factor influencing the positioning accuracy — relative humidity — is clarified and demonstrated,which has not been concerned or even noticed previously.Next,based on the positioning system architecture proposed in Chapter 2,a relationship model between fingerprint databases under different relative humidity is designed,so that the expanded fingerprint database can dynamically adapt to the changes in indoor relative humidity.The designed model not only has the advantages of principal component analysis and multiple linear regression,but also effectively improves the local prediction performance through the constructed locally linear embedding.According to the experimental results after prolonged model deployment in the actual environment,the proposed fingerprint database expansion model is effective in expanding the fingerprint database under different relative humidity,which improves the positioning performance and stability of the positioning system without requiring additional hardware deployment or manual involvement.(3)Although abundant spatial and temporal features are hidden in the mobile trajectory data from the indoor positioning process,the existing positioning algorithms can't effectively utilize these implicit features to improve the positioning performance.In response,this study proposes an indoor positioning algorithm based on the spatiotemporal association of mobile trajectories,where the implicit spatially-local and long-term dependent features are captured by mining the deep spatiotemporal association features hidden in the mobile trajectory,thereby improving the algorithmic positioning performance.Initially,the mobile trajectory fingerprint data associated with position estimation are learned by the built long-term convolutional neural network,a model allowing effective acquisition of the spatially-local and long-term dependent features hidden in the mobile trajectory data.Next,although more training samples are required for the deep learning model,the sole reliance on manual sampling of walking markers in the indoor environment limits its performance due to excessively high costs.Addressing this problem,a mobile trajectory-based data augmentation algorithm is proposed.With this algorithm,the mobile trajectories are segmented into subtrajectories of different lengths through multi-scale sliding windows,mobile trajectory aggregation and track reversal,thereby increasing the sample size to improve the model performance.The results show that the positioning algorithm proposed herein allows effective acquisition of spatiotemporal association features from the mobile trajectory data for improving the positioning performance,which outperforms other compared methods.