Camera-IMU extrinsic calibration method based on intermittent sampling and RANSAC optimization
Xinfeng Liang,Xiaoqi Cheng,Haishu Tan,Yinrui Huang,Xiaosong Li
DOI: https://doi.org/10.1088/1361-6501/ad4dd2
IF: 2.398
2024-05-21
Measurement Science and Technology
Abstract:Accurately calibrate the extrinsic parameters between a camera and an Inertial Measurement Unit (IMU) is a prerequisite for achieving sensor data fusion in visual-inertial navigation systems. Due to delays introduced by triggering, transmission, and other factors, the sampled times of the camera and IMU do not align with the system timestamps, leading to a decrease in the accuracy of extrinsic calibration. Existing calibration methods are unable to avoid the temporal misalignment between the camera and IMU. This paper introduces a novel intermittent calibration sampling method to address temporal misalignment challenges. The approach entails detecting motion-stillness transition points in the IMU data as keyframe segmentation criteria, thereby decoupling data acquisition from system timestamps and aligning it with the sensor's motion process. Subsequently, the robot hand-eye calibration method is applied to the extrinsic calibration of the Camera-IMU, combining the Random Sample Consensus (RANSAC) algorithm to filter the poses of the camera and IMU. This process achieves precise calibration of extrinsic parameters. Experimental validation shows that the algorithm proposed in this paper enhances calibration accuracy when compared to the widely used Kalibr Camera-IMU calibration toolbox. The rotational extrinsic parameter accuracy increased by 181.25%, and the translational extrinsic parameter accuracy improved by 168.33%. These findings provide effective technical means for precise measurement of Camera-IMU extrinsic parameters.
engineering, multidisciplinary,instruments & instrumentation