The improvement of indoor localization precision through partial least square(PLS) and swarm(PSO) methods

Indoor positioning accuracy affects the usability of many IOT applications such as personnel and merchandise tracking. Among the many indoor positioning methods, the signal pattern matching method is also known as the feature fingerprint method. The deployment is simple and low cost. Its working principle is through the collection of wireless signal samples to establish a fingerprint database. When receiving the localization request, the input data is put into the comparison with fingerprint database to determine the exact location of the target emitting the same wireless signal patterns. Many studies are based on this principle to develop different algorithms to improve the positioning accuracy. The shortcomings of this approach is that the positioning precision is easily perturbed by the noisy environment conditions. These noisy signal samples recorded in the fingerprint database will lose the reference value, when low efficiency indexing algorithm or wrong models are used. This paper proposes several novel machine learning based algorithms and indexing methods for indoor positioning accuracy enhancements. They are a modified swarm algorithm, partial least square(PLS) algorithm and genetic algorithms. Different from regular fingerprint localization methods which make use of statistics based models to characterize the signal patterns, our methods make use of several A.I. based algorithms to classify the signal patterns for localization. Compared to other statistics based indoor fingerprint localization methods, our method can reach a precision of 95 % with low development cost and a resolution of 16 cm in a complex computer laboratory environment.