An Indoor Localization Method Based on Depth Migration and Model Parameter Integration

Abstract

The invention belongs to the technical field of indoor positioning, and in particular relates to an indoor positioning method based on depth migration and model parameter integration. The present invention utilizes depth migration to implement mean distance minimization constraints and second-order statistical difference minimization constraints on the deep features of the source domain and the target domain, which can minimize domain differences and enable the model to effectively adapt to complex indoor environments. Using the idea of ​​parameter integration, let the model used for prediction use the exponential moving average mechanism to integrate the parameters of the network trained by the gradient descent method in each training step, which reduces the jitter of the neural network during the training process and ensures the stability of the prediction model. Output. The invention can effectively overcome the problem of error increase caused by environmental changes and measurement deviations of heterogeneous equipment in complex indoor environments and the problem of jitter in neural network training.

Description

technical field [0001] The invention belongs to the technical field of indoor positioning, and in particular relates to a method for positioning in complex indoor environments based on depth migration and model parameter integration. Background technique [0002] With the popularization of the mobile Internet and the widespread use of smart devices, location-based service requests based on indoor environments have become a huge traffic portal. It has many application scenarios, such as large public places such as shopping malls, office buildings, and airport terminals. Provide location navigation and path planning; assist crowd evacuation and firefighter rescue in case of emergencies such as fire; efficiently deploy materials in smart logistics, etc. Indoor positioning systems need to install transmitters at fixed locations to send positioning signals, and a large number of installations will consume manpower and financial resources. Therefore, WiFi, which has been widely de...

AI Technical Summary

Problems solved by technology

In the actual positioning environment, the distribution difference caused by environmental changes is mainly reflected in the variance change, and the distribution difference caused by the measurement values ​​of heterogeneous equipment is mainly reflected in the mean value change. Therefore, the above two methods can only be used for one of the influencing factors. Effective constraints, the reduction of distribution differences is obviously insufficient

In addition, the deep neural network will have obvious jitter during training, resulting in unstable output

Based on the above two reasons, it is difficult for this type of method to achieve stable and accurate positioning results in complex indoor positioning environments.

Images