Article Title

Parazit Yankılı Ortamda Eşzamanlı Konum Belirleme ve Harita Oluşturma Problemi için Veri İlişkilendirme


Robot or vehicle has been tried to build up a map and simultaneously localize its position within an unknown environment or to update its position and map which was called problem in the early of 1990. Smith et al. (1990) called his problem simultaneous localization and mapping (SLAM). They shows SLAM is general problem that while mapping the environments measurement noises statistically dependent on before their values and monotonically growing with map building, so robot/vehicle incorrectly localize their position and obtain environment mapping. There are several theoretical and applicable studies available in the literature. There are some specific problems available with this method. For example, minimization of observation and measurement noises, increasing the number of object in the building environment and robot/vehicle remembering the occurrence of its position which is known as data association(correspondence) in the literature. Recent studies considering these problems and have tried to propose a solution with using different statistical methods based on Bayes theorem. While some of these studies trying to minimize the effects of observation noises on the mapping and position errors, some of them have developed algorithms to reduce the processing times cause of increasing the number of objects related environments causing problems in real-time applications (Montemerlo, M. et al 2001, Kim C. et al 2008). However, some studies have focused on the data association problem, which is known as remembering the occurrence of the robot/vehicle, in other words trying to solve the problem of uncertainty of the object location (Neira J. and Tardos J.D, Rex H. Wong et al 2010). These studies can be listed as Kalman- based estimators, sequential monte carlo approaches, known as particle filters and their derivatives, and expectation maximization based estimators. When observation noise statistically increases over time, measurement data can be obtained in complex and leads to formation of measurement uncertainty. SLAM method takes advantage of the hallmarks of an autonomous robot/vehicle location information while the surrounding the objects. If the landmarks are obtained the correct information, position of autonomous robot/vehicle is used to obtain the correct measurement. In some cases, the number of landmarks and to be close each other leads to the interference which landmark is arrival of the measurement landmarks. In this situations, declining the performance of estimators, leading to increase mean square error of mapping and positions of autonomous robot/vehicle, so SLAM problem causing the results in a number of uncertainties with improper obtaining building of the map and the vehicle’s position and heading angle. In this case, there is a need for data association. Successful data association is provided by observed measurement results from itself correctly association. In the SLAM problem, the estimator is able to forecast the new landmarks, recognize the false alarms (incorrect measurements) and follow the measurements correctly. The most basic algorithm for data association is nearest neighbor method. This method uses Mahalanobis distance during processing. Mahalanobis distance calculates the distance between measured and predicted observation. Algorithm accepts predicted target position closest to the measured position as valid measurement. According to observation measurement creating the acceptance region for next renewal of landmarks, acceptance region is referred to as gate. However, the measurement may not be associated with the nearest landmark, in NN filter not interested in this situation. Therefore, updated state vector may lead to divergence. It is also observed in dynamic environments, are not performing well (Rex H. Wong et al 2010). If number of landmarks is very high and landmarks to be close to each other in noisy observations, NN algorithms do not give better results addressed in the previous studies. Because of this, predictive value of actual measurement from the nearest measurement is known as a reference to valid measurement in the gate. NN algorithm shortens the processing time and not used all the measurements to reach conclusions quickly, so the situation away from the optimal structure. Probabilistic data association (PDA) algorithm calculates the probability of being the target measurement all measurement in the gate. PDA calculates a combined value of innovation, as it tries to solve stochastic problem of uncertainty. Using the relational likelihood of hypothesis provides a unified innovation, and with it creates a unified valid gate. During the operation, algorithm accepts the independent of target and not interfering neighboring landmarks. However, in SLAM problem landmarks are correlated with each other, make mutual interference cannot be ignored. Therefore, it is not suitable multiple object or target tracking and dynamic situations, or algorithm should be run for each target. This is a retreat for PDA and emphasizes joint PDA (JPDA) for multiple target tracking. JPDA has been developed for following all targets in a loop. There are some studies available using JPDA for solving data association problem of SLAM in the literature. (Rex H. Wong et al 2010)’ study used JPDA for solving of wireless sensor networks in SLAM problem and 3-scan JPDA algorithm was used. They said that on the basis of noisy sensor information and possible false repercussions that the signal has white noise and this leads to uncertainty in the position and angle of incoming signals. In other JPDA based approach proposed by Zhou et al. Their proposed method is measurement-oriented, using Depth-First-Search (DFS) algorithm for generation of hypothesis. These studies have been generally used to solve the problem of data association in SLAM problem. However, performance of the algorithms use the filters is not covered. A number of comparisons were only made during the uncertainties. Proposed study taking into account previous studies have focused on two new approaches on a more appropriate for SLAM problem. First, there is an alternative to be presented the problem of data association problem of SLAM in high noisy and uncertainty in feature based environment. Developed algorithm for the problem of SLAM application is proposed for the first time used related environment and scenario. Another improvement is the used filter. Extended Kalman filter (EKF) is generally preferred in previous studies (Rex H. Wong et al 2010, Montemerlo, M. et al 2002). In this study, unscented Kalman filter (UKF) is considered to be more successful in minimizing observation noise problem in SLAM. UKF tries to estimate posteriori probability distribution of state by selecting a certain number of sigma points on probability distribution with a non-linear function. This method allows filter less time to make accuracy and processing speed than FastSLAM based particle filters. JPDA based UKF is for the first time used for SLAM problem in this study. The correct filter used with solution of the uncertainty of situation is thought to achieve the desired result is more favorable.