Map and Localization error prediction dataset
This dataset is derived from a research paper proposing a wireless localization correction methodology based on Convolutional Neural Networks (CNN). The approach involves feature extraction from maps that depict both line of sight (LOS) and non-line of sight (NLOS) effects. The research includes four prediction tasks, categorizing CNN models based on building distribution and propagation mode, resulting in models with low prediction loss. Additionally, an error compensation scheme is designed using CNN-predicted localization errors. Comprehensive comparisons are made between the accuracy of the Time Difference of Arrival (TDOA) wireless localization algorithm and the TDOA results after error compensation. Overall, the CNN prediction method demonstrates significant performance in correcting localization errors. This dataset highlights the importance of preprocessing environmental maps to extract features related to localization error distribution using CNN, especially in complex scenarios involving multipath propagation.
This dataset contains relevant data for wireless localization, including map data and localization error labels. Below are explanations for two significant folders in the dataset:
This folder stores a portion of the map data used for training. Map data plays a crucial role in wireless localization tasks, encompassing the topological structure of the environment, terrain information, and other geographical data relevant to location. These map data are utilized for training and performance evaluation of localization algorithms. The map data includes various types of geographic information, with specific contents dependent on the dataset's purpose and application scenarios.
The localization error labels stored in this folder are vital for optimizing the accuracy of traditional TDOA-based localization methods. In wireless localization tasks, precise location estimation is crucial. These error labels record the associated error information for each location estimate, typically in the form of distance and angle errors, used as a label for deep learning model training.