عنوان مقاله [English]
The accuracy of positioning depends on the quality of the technology used. Various technologies and techniques are used for positioning which are classified as absolute and dead-reckoning groups. Classified as absolute positioning technologies,GPS receiversface a variety of different errors in the real-time positioning of a moving object, which reduces the accuracy and precision of the position received from these receivers. On the other hand, dead-reckoning sensors such as gyroscopes and magnetometers which measure real-time state of a moving object also have cumulative errors.Therefore, observations made by all of these sensors are not free from the noise generated during the measurement process.The amount of this noise may vary depending on various factors, including the precision of the sensor and features of the measuring environment. Thus,due to thecorrelation between observations made by these two categories of sensors and the difference between their precision and the nature of their errors,ifnoise is reduced inobservations made by them, their complementary features can be used to reduce errors made by each of them.High-quality positioning technologies are expensive and require high expertise.As a result,lower quality and cheaper global navigation satellite systems (like GPS) widelyavailable in smartphones are more commonly used. One of the most important features of these inexpensive technologies is that they are highly susceptible to factors producing noise.
The present studyinvestigates the effect of gradual reduction of noise from data collected by sensors, accelerometers, magnetometers, gyroscopes, and GPS technology in smartphones on improvement of vehicle positioning. The proposed method is based on using acceleration, azimuth, latitude, longitude and roll angle parameters as an input for the Kalman algorithm and investigates the effect of reducing noise produced by these parameters using the least-squares method onimprovement of the resulting position calculated by the Kalman algorithm. To reach this aim, the roll angle parameter is extracted from the angular Velocity() in y-direction and the azimuth parameter is extracted from the magnetic field() in both x and y directions. These parameters along with the acceleration(a) parameter in x and y directions and the geographic coordinates are selected for the Kalman filtering algorithm. In the proposed method, data received from sensors share common sources of noise produceddue to drift, random movements and bias errors.To reduce this noise independently and systematically, method of averaging with the least-squares is usedfor data produced by each sensor. Thus, noise in the received data is considered as a random parameter and noise reduction is performed based on the percentage of changes in the corrected and observed data in the range of 1 to 10%. Kalman algorithm is implemented for 10 levels of noise reduction and the results areinvestigated and compared.The filter calculates and improves an estimate of position vector x, denoted by with minimum mean square error using a recursive model. The main objective is to derive an accurate estimate of for the state of the observed system at time of k. Implementing Kalman filter consists of a prediction step and an updating step. The result is compared todata received from a more accurate reference using RMSE.
Results and Discussions
The study area consists of lane no. 2 of the South-North (East-West) Azadegan Highway, Tehran, Iran with a total area of about 26km. Results show that compared to the reference data, using Kalman filter has decreased errorsin positioning the car from 0.8274 m to 0.6763 m with a 2%noise reduction. With a 10% noise reduction, the accuracy of this method has increases to 0.6771 m. This improved accuracy is due to noise reduction and consequently an increase in the correlation between the parameters. Accordingly, the threshold limit for noise reduction and improved positioning using Kalman filter is low and can be recognized by an investigation of a few lowlimits. According to the findings, although reducing the effect of noise can improve positioning with Kalman filter and smart phone sensors, irregular changes in the accuracy of noise reduction methods require determining an optimal percentage for noise reduction.