1 e sin LLa
Chapter 5 Integration of GPS/INS/GIS
5.4 Navigation System Design
5.4.2 Programming Procedure
After the hardware and software were represented, the next work would be devoted to design the navigation programming procedure. The procedure could be divided into two major parts, one is the data receive in two microprocessors; the other one is the data process in main processor. In two microprocessors, one is master and the other one is slave, the slave would handle the GPS message decode and data transmitting between the master and slave. The master would manage the IMU data, the sampling rate in the navigation system, and the interface between the master and the main processor.
The slave would receive the GPS message that is the word form, and then decode these words to the number of latitude and longitude per second. As the latitude and longitude have been decoded, the data would be transmitted to the master.
Furthermore, the IMU outputs from sensors are analog signal, thus it is necessary to convert them to digital signal that could be processed in microprocessor. After using an A/D convert to completing this work, the master would handle the IMU digital
Figure 5.8 Shapefile in ArcView
output, and then sends a request to the slave to get the GPS data. If GPS data were available, then the master would send GPS data and IMU data to the main processor with RS232 in 57600bps. However, if the GPS data were not available, the master would send the last GPS data. Figure 5.9 shows the data flowchart and how the master and slave work.
As the raw data has been transmitted to the main processor with RS232, the program developed with Visual Basic would start to process these data. The most important work of data process is to set the initial condition of the IMU data, since the
Start
8051 (M) 8051 (S)
Main Processor
Initialize Initialize
Lock S/H
A/D converter
Get GPS Message
Decode
Buffer
RS-232
Data ok?
No Request 1 Hz
GPS Available?
Yes
Send the last GPS data
No Yes
Figure 5.9 The data flowchart in two microprocessors
initial error would affect the further estimation and system performance far. The initial condition consists of position, velocity, and heading direction, which need to be estimated accurately.
The GPS positioning deviation is almost equal to 5 m and would not sometimes stable. Therefore, the initial heading direction measured from two nearby points would not be reliable and unstable, and isn’t sometimes equal to the practical heading direction shown as figure 5.10 (a). As the INS integrated with GPS, the heading direction measured from GPS could not be used to correct the heading direction in INS. In order to solve this problem, a method that measures the heading direction from 1stand 5thpoints and obtains the continuous stable heading direction is proposed.
Measuring heading direction from two nearby points may not suitable due to the unstable characteristic in short-term time. The proposed method shown as figure 5.10 (b) is to measure two points whose time interval is longer. As there are five GPS points, the 1stand 5thpoints are chosen to measure the heading direction,and is measured as the same method. While the,,,and 4 are measured, they may not, however, practical due to the fact that one of themmay be the error shown as figure 5.11 (c). Thus, in order to determine the practical heading direction, the proposed method obtains a stable heading direction from several continuous past heading directions shown as figure 5.11 (d). If these past angles were almost the same, then the last angle 4 would be the initial heading direction. If they had large difference as figure 5.11 (c), the method would keep finding the stable heading direction as initial heading direction for INS.
After the heading direction initialization procedure, the initial position could be also obtained as the last position in heading direction initialization procedure. Since the position is considered as the stable position from GPS. Furthermore, as the initial position is given, the initial velocity would be accumulated from acceleration while
the system is started. The initial condition would be summarized as figure 5.11.
As the initial condition has been set, the next work would be devoted to estimate the position in the map. While the navigation works as the initial condition is given,
Practical trajectory
Heading direction measured from two nearby points
Figure 5.10 Heading direction estimation
(a)
(d) (b)
(c)
Figure 5.11 Initialize heading direction for INS
Start
Determine heading direction from 1stand 5th
If the past heading directions are stable?
Get the initial heading direction
Yes
No
Set the last position as initial position Accumulate the
initial velocity from acceleration
Initial condition complete
INS would be used to estimate the navigation information before the next GPS signal coming. As a new GPS signal is coming, the checking procedure would check that if this GPS signal were reliable. This checking procedure may employ the heading direction estimation to check the reliability of GPS data. If there were a strange GPS point P1 shown as figure 5.12 (a), i.e., the heading direction'is not similar to , the checking procedure would not treat P1 as reliable and keep determining the INS until GPS data gives a stable heading direction. Even though P1 locates near the turn and the vehicle would turn shown as figure 5.12 (b), the procedure would also treat P1 as unreliable and keep determining the INS. Since if the vehicle will turn, INS would detect and calculate the turn with less distortion. As a reliable GPS data is coming, the errors would be determined from subtracting the INS information from the GPS information. Then put these errors into the Kalman filter to calculate the optimal errors that would be subtracting from INS information shown as in figure 5.1. All the procedures described above just employ the GPS/INS, and further the GIS would be adopted to achieve the optimal navigation information.
The GPS/INS/GIS integration has been introduced in 5.2, and the GIS would be adopted as the method in 5.2. Except for direction correcting purpose, the GIS also could detect if the vehicle pass through the new road that doesn’t exist in GIS
Figure 5.12 Check position reliability
(a)
P
'
P
(b)
'
Practical trajectory GPS point
Heading direction measured from nearby two points Heading direction measured from 1st and 5th points
database. In the map-matching algorithm, the searching mode would detect if there were a road near the vehicle. If not, the origin procedure would locate the vehicle in the previous road. However, if the vehicle keeps going on this new road, the matching point would locate the almost same point due to the fact that matching road is the same. In order to improve this phenomenon, as the procedure enters the searching mode, if there were no road in GIS database, GPS/INS position would replace matching position and show on the map. Consequently, the driver would know the location of vehicle continuously and reliably.
5.5 Experiments
In order to test the performance of this navigation system, there are two experiment paths. Path 1 shown as figure 5.13 would pass through the underground passage, urban area, and wooded area, and these areas would test the GPS/INS/GIS performance while GPS signal is available and examine the INS/GIS performance when GPS signal is unavailable. Path 2 shown as figure 5.14 would be planed in National Chiao Tung University whose some roads doesn’t exist and update in GIS database. Therefore, this path is useful to test GPS/INS performance without GIS correction.
Figure 5.15 shows the result of GPS/INS without the GIS correction, but the map-matching with GIS. The dash-line circle A and E in figure 5.15 shows the strange phenomenon of GPS that the longitude is static but the latitude varies, so that the trajectory would keep straight and vertical to parallel. However, the trajectory isn’t towards the real path and would cause the error matching shown in dash-line circle A.
Therefore, the most convenient and simple solution is to judge if the longitude is static, then the GPS signal would be classified to be unavailable.
Since this phenomenon occurred at short-term time, it is reliable to depend on the INS/GIS in section A and E. In the section B, the vehicle pass through an underground passage, and then the vehicle turn right and enter an urban area. From the figure 5.15, there is no available GPS signal for GPS/INS and the heading direction strayed from the real path, so that the location was matched to the error road until the GPS
Underground passage
Urban area
Wooded area
Figure 5.13 Path 1
Experiment Path
Figure 5.14 Path 2
Figure 5.15 Path 1 result without GIS correction
GPS position
GPS/INS position
Matching position on the road A
B
C D
E
signal was available in section C. However, the signals seemed to be untidy and the trajectory was rough, since there were more high buildings around the vehicle. In order to solve the phenomenon, the heading direction estimation shown in figure 5.10 and 5.12 would be adopted to smooth the trajectory for driver. There is a wooded area in Section D; the GPS signal, however, seemed to be available and was sometimes untidy in the middle of section D, so that the error matching would happen. This problem would be solved with the same method, heading direction estimation.
The path1 was experimenting with the modified procedure and the result is shown as figure 5.16. In the section A and E, the problem of strange GPS signal has been solved, so that the trajectory was smoother and there was no error matching. As the vehicle pass through the underground passage, the GIS aided the INS to correction the heading direction until the vehicle turned right. While the INS and GIS detected the node and turn, the INS would work without the heading direction aid from GIS.
Therefore, the trajectory was smoother and more correct, and furthermore the matching was more correct.
While the GPS/INS/GIS performance and INS/GIS has been tested, the path 2 would be tested the performance of GPS/INS. Some roads in National Chiao Tung University are not updated in GIS database. Therefore, as the vehicle ran in the road shown as figure 5.17, the GIS would match the position to the other road where vehicle didn’t run. While the matching positions were almost on the same location and the vehicle was moving, the GPS/INS position would be substituted for matching position shown as section B. Section C is a wooded area, so that the GPS signal was unavailable and INS should work alone. Even the INS worked without the aid of GPS or GIS, the trajectory was reliable and smooth due to the integration of acceleration and angular rate. The equipment employed in this experiment is shown as figure 5.18, the IMU set consists of one accelerometer and one gyro.
Figure 5.16 Path 1 result with GIS correction
GPS position
GPS/INS position
Matching position on the road
A B E
GPS position
GPS/INS position
Matching position on the road A
C D
Figure 5.17 Path 2 result
B
Figure 5.18 Experiment equipment
GPS antenna
IMU set
12 V battery
Main processor
Chapter 6 Conclusions
This thesis mainly contributes to adopting Kalman filter to integrate GPS with INS and map-matching algorithm to integrate GPS/INS with GIS. In chapter 2, the pseudo range would be determined from the positioning equation. With the pseudo range and carrier phase, the absolute position of vehicle would be calculated.
However, GPS is not always available for navigation. INS built up by a set of inertial measurement units (IMU) can provide continuous position and would not be affected by external interference. Therefore, in order to get continuous and smooth navigation information, INS introduced in chapter 3 is integrated with GPS. The coordinates of GPS and INS are different; thus, the position, velocity, and attitude of them need to integrate. Therefore, the required coordinate transformation is presented in section 3.2.
In order to determine the navigation information in INS, the dynamic equation introduced in section 3.3 would be adopted to calculate this information. As determining this information from dynamic equation, the error caused from integration would accumulate and affect the performance of INS. Thus, the INS error model is derived to estimate the error. Then the Kalman filter presented in section 5.3 would be employed to integrate GPS/INS and estimate the error based on INS error model and observation model from GPS.
Even the GPS/INS is more accurate and reliable than GPS or INS alone, the absolute position from GPS/INS can’t provide sufficient information for driver. GIS presented in chapter 4 would provide more accurate database to match the path. In this thesis, the position would be matched on the map in Taiwan coordinate. A map-matching algorithm, which includes initial mode, node nearby detection, searching mode, and tracking mode, would be used to match the position of GPS/INS to the correct road and point on map. This algorithm would provide the relative
position and the GIS provides the geographic information for driver.
As the system starts to navigate, the hardware constructed based on two microprocessors would receive the data that consists of acceleration and angular rate from INS and signal from GPS. Then these data would be transmitted to the main processor via RS232 interface. In the main processor, these GPS and INS data would be integrated with the conventional Kalman filter that is commonly adopted in navigation system to get optimal solution. As the map-matching algorithm is employed, there are still some matching errors due to the wrong computational heading direction. Therefore, a proposed heading direction correction method would compare and fit the trajectory of GPS/INS to the vehicle’s real path. This proposed method would fail as the vehicle makes a turn. Therefore, a method that detects the node from GIS database and measures the variation of direction from angular velocity of vehicle is proposed to adjust the heading direction correction. Furthermore, as the vehicle runs in the road that doesn’t exist and isn’t updated in GIS database, a proposed strategy would detect whether there is a road near the vehicle or not. If yes, the GPS/INS position would replace the matching position. With the strategy, the navigation system would provide the reliable geographic information for driver.
Consequently, with the experiments in section 5.5, the navigation system is flexible in different environments, even the underground passage, wooded area, urban area, and new built road. Furthermore, the results show these methods are feasible, and the trajectory is continuous, smooth, and reliable.
A flexible GPS/INS/GIS real-time navigation system has been proposed in this thesis. However, there are some drawbacks in this navigation system. One is that the road of GIS database is constructed based on the centerline of road, and the road in the map doesn’thave the width. Therefore, the error of GIS database would depend on the width of roads. The other one drawback is that even the experiment tested the
performance of system in several different environments. However, there may be some strange environments, such as long tunnels, and the system may be fail in these environments due to the fact that the error in INS would accumulate, as GPS was unavailable for a long time. Another drawback is the small sampling rate due to the ability of main processor. As the navigation system would be adopted in autonomous vehicle control system, the sampling rate is too small to observe the dynamics.
Finally, the measurement error is poor to know and the GPS noise is not white Gaussian in practice. However, Kalman filter is implemented in the linear system and the noises of process and measurement are white Gaussian. Therefore, the estimated information from Kalman filter is not optimal. In order to solve this problem, some researchers adopt the adaptive Kalman filter to improve the poor knowledge of measurement error. [8] The adaptive Kalman filter would update the measurement error and the error estimation would be more accurate. However, the adaptive Kalman filter would take more computation time to estimate and update errors. In this thesis, the processor isn’t able to fulfill the computation in time. Thus, in order to employ the adaptive Kalman filter, it is necessary to modify the entire algorithm used in this thesis. The future work will be devoted to modify the Kalman filter and map-matching algorithm to reduce the computation time and obtain more accurate navigation information.
Reference:
[1] J. Farrell and M. Barth, “The Global Positioning System and Inertial Navigation”,McGraw-Hill, New York, 1998.
[2] R. C. Brown and Y. C. Hwang, “Introduction to Random Signalsand Applied Kalman Filter”,John Wiley & Sons, New York, 1997.
[3] Christopher Jekeli, “Inertial Navigation Systems With Geodetic Applications”, Walter de Gruyter, Berlin, New York, 2001.
[4] I. Y. Baritzhack and N. Berman, “Control Theoretic Approach to Inertial Navigation System,”Journal of Guidance, Control, and Dynamic, Vol.11, No.3, pp.237-245, 1988.
[5] Nebot Eduardo, Sukkarieh Salah, and Durrant-whyte Hugh, “Inertial navigation aided with GPS information”, Mechatronics and Machine Vision in Practice, 1997 Fourth Annual Conference on Proceedings, pp.169 –174, 23-25 Sept. 1997.
[6] Gordon, G.S, “Navigation systems integration”, Airborne Navigation Systems Workshop (Ref. No. 1998/275), IEE, pp.411 –416,10 Feb 1998.
[7] Liu Lichuan, Tian Zengshan, and Huang Shun-ji, “An algorithm for integrating GPS/INS attitude determination system”, Radar, 2001 CIE International Conference on Proceedings, pp. 167 –170, 15-18 Oct. 2001.
[8] Christopher Hide, Terry Moore, and Martin Smith, “Adaptive kalman filtering for low-cost INS/GPS”,The Journal of Navigation, Vol. 56, pp.143-152, 2003.
[9] 謝銘峰, “GPS/DGPS 與慣性導航系統之整合研究”, 國立交通大學控制工
程研究所碩士論文, 1995.
[10]李忠隆, “應用於車輛導航之 GPS/INS 整合系統設計”, 國立交通大學電機
與控制工程研究所碩士論文, 2000.
[11]紀佳宏, “應用於即時車輛導航系統之 GPS/GIS 整合設計”, 國立交通大學
電機與控制工程研究所碩士論文, 2002.
[12]莊智清, 黃國興, “電子導航”, 全華科技圖書股份有限公司, 台北, 2001.
[13]黃國與, “慣性導航系統原理與應用”, 全華科技圖書股份有限公司, 台北, 1991.
[14]長天科技股份有限公司, “HOLUX GM80/81 衛星接收模組使用手冊”.
[15]Analog Devices ADXL202 datasheet.
[16]Murata ENV-05A datasheet.
[16]Murata ENV-05A datasheet.