1 State quantity of target positioning
Use the lower left corner as the origin to establish a rectangular coordinate system,
x: the abscissa of the trolley;
y: the longitudinal coordinate of the car;
2 Multi-sensor fusion method-Kalman filter
Here we first look at Baidu's description in the Apollo course:
"In Apollo, the inertial navigation solution (inertial navigation solution) is used for the prediction step of the Kalman Filter. GNSS (GPS) and LiDAR localization are used for the measurement update step of the Kalman Filter. "
3 Type of sensor used
3.1 Sensors used
Speedometer (category IMU): it will display the moving speed of the car in real time (if it moves, the moving speed will be displayed; if it does not move, the speed will be 0);
SLAM front end (using feature sub-algorithm): will give the relative displacement of the car;
GPS: global coordinate information, (not very accurate, can be further precise through template matching);
3 Writing the equation of state
When there is no movement, the transfer matrix is the identity matrix;
SLAM-front is used as the control variable
；
4 Writing of observation equations
4.1 High-precision positioning based on GPS coordinates
Input: GPS coordinate data;
use GPS coordinates to obtain the coverage of high-precision maps, and then use template matching for precise positioning,
so as to obtain accurate coordinates after querying based on high-precision maps;
4.1.1 Optimization of template matching efficiency
Teacher Chen said this: "Can only change the matching algorithm, or use the image pyramid, or dimension reduction"
Don't know if you can change the step size?
4.2 Model of the observation equation
If the value of the observation sensor is determined to be unreliable at this time, the output of the observation equation can be set to 0 at this time;
5 Remarks
5.1 I heard that SIFT implementation of VLFeat is better than OpenCV
Try it next time
5.2 In actual project Will it really converge?
