Cameras and Inertial/Magnetic Sensor Units Alignment Calibration

Abstract

Abstract-Due to the external acceleration interference/magnetic disturbance, the inertial/magnetic measurements are usually fused with visual data for drift-free orientation estimation, which plays an important role for a wide variety of applications, ranging from virtual reality, robot, computer vision, to bio-motion analysis and navigation. However, in order to perform data fusion, alignment calibration must be performed in advance to determine the difference between the sensor coordinate system and camera coordinate system. Since orientation estimation performance of the inertial/magnetic sensor unit is immune to the selection of the inertial/magnetic sensor frame original point, we therefore ignore the translational difference by assuming the sensor and camera coordinate systems sharing the same original point and focus on the rotational alignment difference only in this paper. By exploiting the intrinsic restrictions among the coordinate transformations, the rotational alignment calibration problem is formulated by a simplified hand-eye equation AX = XB (A, X and B are all rotation matrices). A two-step iterative algorithm is then proposed to solve such simplified hand-eye calibration task. Detailed laboratory validation has been performed and the good experimental results have illustrated the effectiveness of the proposed alignment calibration method.

Wearable Biofeedback Suit to Promote and Monitor Aquatic Exercises: A Feasibility Study

“…Joint angles estimated by the sensor network were compared to those estimated by an optical tracking system. The two independent systems are calibrated to be coherent in coordinate systems, as previously introduced in sensor fusion approaches [30], and their data is processed with the same mathematical steps. RMS angle error between the two systems ranged from 4.0° to 6.3°, with a significant correlation coefficient always higher than 0.990.…”

“…The output of each Kalman Filter is a quaternion, to feed angle reconstruction, as in [24], [26]. It was in fact decided to acquire and process data using quaternions, which allows us, on the one hand, to improve computational efficiency, something crucial for real-time applications and, on the other hand, to avoid singularities [27], [28], especially in body motion estimation [11], [30]. Inertial-based human joint angle tracking has been already investigated in the literature (e.g., [23], [31]).…”

Minimal solutions for the rotational alignment of IMU-camera systems using homography constraints

“…The existing methods can be divided into three groups. The first group of methods solves the rotational and translational components separately [6,8,9] or only solves the rotational component [20]. The second group of methods solves the rotational and translational components simultaneously [5,6,7,21].…”

Quantifying Variation in Gait Features from Wearable Inertial Sensors Using Mixed Effects Models

“…In general, these methods are limited to providing accurate one-time calibration. Even with proper pre- and post-calibration to detect mounting location shifts [18], these shifts can still be unnoticed during continuous data collection in day-to-day use, despite proper precautions to prevent sensor shift.…”

“…Since inertial sensors are sensitive to orientation changes, mounting location shifts can also cause significant variations in sensor data. While these issues can be mitigated in a controlled laboratory environment by adopting elaborate calibration procedures [18] before and after data collection, it is very challenging to control or correct these factors in the free-living environment during continuous data collection. Therefore, these factors must be characterized by statistical modeling before the gait analysis results can be applied to assist clinical inference.…”