Error Correction Method with Precise Map Data for GPS/DR Based on Vision/Vehicle Speed Sensor

Location is one of the most important information in the industrial world and there are many systems provide location information. GPS is the most universal positioning method. Especially, RTK(Real-Time Kinematic) algorithm is commonly used for ITS(Intelligent Transport System) precise positioning. Moreover, for the ITS, the fundamental requirement is that the positioning system can provide users information about in which lane or not. In the open sky environment, RTK can be a good practical solution. In the previous research, RTK floating solution can provide sub-meter level accuracy even though integer ambiguity was not solved. However, in bad condition receiving satellite signals such as urban area, there are positioning errors because of multipath, cycle slips, signal block and so on. And there are tall buildings on both sides of the street, which is pretty obvious most part of ITS environment, thus satellite based positioning system use not the both sides but the front and the rear satellite signals. It occurs lateral positioning error on the street, and it cause critical problems for land transportation environment. In order to overcome this problems, there were various attempts using infrastructure and another sensors. But, infrastructure and additional sensors become pecuniary burden. Construction of infrastructure is a national issue, so this cannot be realized easily and pretty much expensive. And additional sensors such as IMU(Inertial Measurement Unit) and especially LRF(Laser Range Finder) is very expensive are big burden to users. The most users of ITS are civilians, therefore the main target system can be car such as sedans and saloons. And in these days, most cars have ECU(Electronic Control Unit) which can provide vehicle speed and black box for land vehicle is not only easy to find but also getting cheaper. Thus the pecuniary burden is not a big problem. That’s why many vision based navigation algorithm are studied. DR(Dead Reckoning) navigation is one of the most used method for fusion GPS and sensors. In the DR navigation, there are many method such as loosely coupling, tightly coupling and so on. In this paper, the loosely coupled GPS/DR navigation was used. However the weakest point of DR is an accumulated errors. Because of this errors, GPS is used for boundary of errors. But if the accuracy of GPS is bad, positioning error of DR navigation occurs. In this paper, ready-made speed sensor from ECU in the vehicle and equipped vision sensor were used. Using only this two sensors, the development cost can be reduced. The vehicle obviously drives through the lane, speed sensor provides longitudinal and vision sensor provides lateral information. If lane is detected, the distance and angle between vehicle and lane can be measured. Lane detection is performed by edge detection of ROI(Region of Interest) which can reduce computation time and extraction of line parameter by Hough transform. The transformation of line parameter is required from image frame to navigation frame in order to use to DR navigation, and it can be easily performed by IPM(Inverse Perspective Mapping). In this paper, the vehicle and vision sensor assumed perfectly aligned. It means that the slope of lane in IPM image is equal to the angle between vehicle and lane. Using these information, DR navigation is performed and errors are accumulated unfortunately. But in the GPS, the longitudinal error is the smaller part, and in order to distinguish lane, longitudinal error is not critical problem. However the lateral error is critical. Thus correction of lateral error accumulation should be processed. In this paper the precise map data is used for the correction. Provision of precise position information to user means that there are precise map data already. Because user don’t want to know couple of digits of location, but display the location information on the map. Thus this assumption is acceptable. The lateral error of vehicle is accumulated by change of heading of vehicle motion and lane. However surveyed lane information is in the precise map data, thus this can correct the heading of lane with zero uncertainty. It means that the angle between vehicle and lane extracted from vision sensor make heading angle calculation of vehicle possible. In this paper, DR navigation was performed and analyzed using the heading and speed information measured from speed meter in vehicle, vision sensor and precise map data by actual experiment. As a result, precise map data can correct lateral error accumulation of DR navigation. And by correcting the lateral error of GPS using proposed DR navigation, the positioning result can provide lane distinguishable accuracy.