دانلود رایگان مقاله الگوریتم تعیین اندازه نمونه برای سیستم تعیین موقعیت داخلی مبتنی بر اثر انگشت

Abstract

Provision of accurate location information is an important task in the Internet of Things (IoT) applications and scenarios. This need has boosted the research and development of fingerprint based, indoor localization systems, since GPS information is not available in indoor environments. Performance evaluation of such systems and their related localization algorithms, is usually based on sampling collection in predetermined test environments. The sample size determination and sampling methodology can significantly affect the reliability of the outcome. This work proposes an algorithm that calculates the minimum sample size of positioning data required for objective performance evaluation of fingerprint based localization systems. The use of a correct, independent, unbiased and representative sample size can speed up the training, evaluation and calibration procedures of a fingerprint based localization system, while ensuring that the system’s true accuracy is achieved. The proposed Sample Size Determination Algorithm (SSDA) takes into consideration the desired confidence level, the resulting standard deviation of a small size preliminary sample as well as the error approximation with respect to the actual error of the system and proposes the final sample size for the evaluation and/or calibration and/or training of the utilized radio-maps. Additionally, the SSDA, assumes random sample allocation in the area of interest in order to avoid biased results. Risks arising from the selection of a sample of convenience are also investigated. Finally, the performance of the proposed algorithm is tested in both measured and simulated radio-maps.

6. Conclusion

In this paper an algorithm is presented (SSDA) that allows the calculation of a safe margin sample size to be used during the training, calibration and performance evaluation of fingerprint based localization systems. The proposed methodology suggests the utilization of an initial preliminary sample selection, the definition of an acceptable positioning error bound and a predetermined confidence interval. The suggested sample size is then extracted by converting the locations from infinite to discrete and by setting a minimum grid size for the area of interest. Additionally, the importance of selecting a simple random sample is highlighted and compared with a sample of convenience, demonstrating that in the latter case, the results can vary systematically, leading to unreliable conclusions. Finally, SSDA was tested in radio-maps that were generated through measurements and simulations. The outcome indicated that the estimated data sample size, objectively captures the actual system’s positioning accuracy performance. The presented work contributes towards the standardization of RTLS evaluation procedures.