CN112414431A - Robust vehicle-mounted multi-sensor external parameter calibration method - Google Patents

Abstract

The invention relates to a robust vehicle-mounted multi-sensor external parameter calibration method, and belongs to the technical field of vehicle-mounted sensor calibration methods. The method comprises the following steps: step 1: respectively acquiring pose sequences of sensor time alignment; step 2: determining a sliding window, and estimating sensor external parameters by using a pose sequence in the window; and step 3: moving a window, re-estimating external parameters, and fusing the external parameters with the external parameters estimated previously; and 4, step 4: and (5) repeating the step (3) to obtain stable calibration parameters corrected along with time. The method does not need specific fields, specific marks and operation of professionals, can continuously calibrate the multi-sensor external parameters in the continuous running process of the vehicle, ensures the accuracy of the multi-sensor external parameters, and has good robustness for the sensor data with errors.

Description

Robust vehicle-mounted multi-sensor external parameter calibration method

Technical Field

The invention relates to a robust vehicle-mounted multi-sensor external parameter calibration method, and belongs to the technical field of vehicle-mounted sensor calibration methods.

Background

Autonomous vehicles need to be able to sense the surrounding environment and locate themselves quickly, accurately, and robustly. In order to achieve the purpose, the automatic driving vehicle is generally equipped with various sensors such as a camera, an IMU (inertial measurement unit), a GPS, a laser radar and the like, and data are fused, and the calibration of the sensors is very important. Calibration of the internal parameters, which is often done at the time of sensor production, ensures that a single sensor can provide accurate data. The external parameter calibration is required to be carried out after the final arrangement of the sensors is completed, and is the basis of multi-sensor fusion.

Conventional external reference calibration methods require that some reference markers are known, usually in a laboratory, and need to be recalibrated when the layout of the sensor changes. Due to factors such as vibration of the vehicle in the long-time running process, the position of the sensor is difficult to avoid, and most consumers do not have a standard calibration place and calibration skill. Therefore, the consumer-grade automatic driving automobile needs to be provided with an external reference calibration method which can be automatically completed after the sensor is installed and does not need professional knowledge and a specific place.

Disclosure of Invention

In order to solve the problems that the original external parameter calibration fails due to the fact that the sensor deviates due to vibration and the like in the vehicle running process, a professional needs to recalibrate the sensor in a professional field, and the sensor is difficult to be used commercially, the invention provides a robust vehicle-mounted multi-sensor external parameter calibration method.

The invention adopts the following technical scheme for solving the technical problems:

a robust vehicle-mounted multi-sensor external reference calibration method comprises the following steps:

step 1: respectively acquiring pose sequences of sensor time alignment;

step 2: determining a sliding window, and estimating sensor external parameters by using a pose sequence in the window;

and step 3: moving a window, re-estimating external parameters, and fusing the external parameters with the external parameters estimated previously;

and 4, step 4: and (5) repeating the step (3) to obtain stable calibration parameters corrected along with time.

The specific process of step 1 is as follows:

for a monocular or binocular camera, acquiring a pose sequence by using any visual SLAM algorithm; for the laser radar, acquiring a pose sequence by using any laser SLAM algorithm; and for the IMU sensor, acquiring a pose sequence through integration.

The invention has the following beneficial effects:

the method does not need specific fields, specific marks and operation of professionals, can continuously calibrate the multi-sensor external parameters in the continuous running process of the vehicle, ensures the accuracy of the multi-sensor external parameters, and has good robustness for the sensor data with errors.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

FIG. 2 is a comparison graph of the effect of the method of the present invention and the effect of the conventional external reference calibration method.

Detailed Description

The invention is described in further detail below with reference to the accompanying drawings.

A robust vehicle-mounted multi-sensor external reference calibration method is shown in FIG. 1, and comprises the following steps:

step 1: respectively acquiring pose sequences of multi-sensor time alignment;

step 2: determining a sliding window, and estimating sensor external parameters by using a pose sequence in the window;

and step 3: moving a window, re-estimating external parameters, and fusing the external parameters with the external parameters estimated previously;

and 4, step 4: and (5) repeating the step (3) to obtain stable calibration parameters corrected along with time.

(1) Acquiring a pose sequence of multi-sensor time alignment:

for a monocular or binocular camera, a pose sequence can be acquired by using any visual SLAM algorithm; for the laser radar, a pose sequence can be obtained by using any laser SLAM algorithm; the IMU sensor can acquire a pose sequence through integration. Alignment of the timestamps is achieved using hardware triggering and linear interpolation.

(2) Determining a sliding window, and estimating sensor external parameters by using a pose sequence in the window:

different window sizes can be selected according to hardware computing power and actual use requirements. A. B respectively represents the poses of the two sensors, and the time alignment pose sequence is recorded as:

ζ_A＝*A₀ A₁ … A_n+

ζ_B＝*B₀ B₁ … B_n+

therein, ζ_AAs a pose sequence of sensor A, A₀For the first frame pose of sensor A in the sliding window, A₁For a second frame pose of sensor A in the sliding window, A_nFor the nth frame position, ζ, of sensor A in the sliding window_BAs a pose sequence of sensor B, B₀For the first frame pose of sensor B in the sliding window, B₁For the second frame pose of sensor B in the sliding window, B_nIs the nth frame pose of sensor B in the sliding window.

Wherein: r_AIs a rotation matrix of sensor A for a certain frame, t_AIs the translation vector, R, of a certain frame of sensor A_BIs a rotation matrix of sensor B for a certain frame, t_BIs the translation vector for sensor B for a certain frame.

The cost equation is:

wherein h () represents the cost equation, A_τFor the # th frame position of sensor a,

B_τis the position of the Tth frame of the sensor B, and is each frame traversed in the sliding window;

wherein: r is the rotation matrix of sensor A relative to B, and t is the translation vector of sensor A relative to B.

And (4) performing iterative optimization on the cost equation by using the unit matrix or the external parameter calibrated before as an initial value and using a Levenberg-Marquardt method. And respectively calculating errors for each pair of poses in the pose sequence by using the optimized external parameter estimation:

E_τ＝A_τX-XB_τ

wherein: e_τIs the error of the tau frame pose.

And eliminating the pose pairs with the errors larger than the threshold value, and optimizing the cost equation again. The process is repeated until all pose pairs meet the error requirement, or the maximum iteration number is reached, and the external parameters X of the A, B two sensors can be obtained. If more sensors exist, optimization can be carried out in the same way to obtain the external parameter estimation of the multiple sensors.

(3) Moving the window, re-estimating the outliers, and fusing with the previously estimated outliers:

and moving the window backwards for several frames according to the calculation capacity and the actual use requirement to obtain a new time-aligned pose sequence. The external parameters are re-estimated using the method of step 2. And calculating the mean value of the external parameters.

(4) And (5) repeating the step (3) to obtain stable calibration parameters corrected along with time.

As shown in fig. 2, a combination of a laser radar and a binocular camera is fixed on the vehicle, external parameters of the combination are calibrated, a laser range finder is used for monitoring the distance between two sensors, and the error between the monitoring result of the range finder and the external parameters of the sensors is calculated. When the vehicle runs through the deceleration strip and bumps, the sensor displaces, and the external parameter error of the sensor becomes large. The curve calibrated in driving is the result of continuously performing external reference calibration in the driving process according to the method of the invention, and the curve without calibration in driving is a comparison group. It can be seen that the method provided by the invention can effectively reduce the external parameter change caused by the displacement of the sensor and obtain the accurate external parameter calibration result of the multiple sensors.

Claims (2)

1. A robust vehicle-mounted multi-sensor external reference calibration method is characterized by comprising the following steps:

step 1: respectively acquiring pose sequences of sensor time alignment;

step 2: determining a sliding window, and estimating sensor external parameters by using a pose sequence in the window;

and step 3: moving a window, re-estimating external parameters, and fusing the external parameters with the external parameters estimated previously;

and 4, step 4: and (5) repeating the step (3) to obtain stable calibration parameters corrected along with time.

2. The robust vehicle-mounted multi-sensor external reference calibration method according to claim 1, wherein the specific process of step 1 is as follows:

for a monocular or binocular camera, acquiring a pose sequence by using any visual SLAM algorithm; for the laser radar, acquiring a pose sequence by using any laser SLAM algorithm; and for the IMU sensor, acquiring a pose sequence through integration.

Images