The camera pose relative to the scene can be estimated in all six degrees of freedom (DOFs) by using a stereo-camera system or by incorporating some a priori knowledge of the scene when a monocular system is used. The information provided by finding and associating image points of interest through a monocular video stream (monocular visual tracking) can be used to estimate the camera orientation relative to an absolute reference frame. The concurrent estimation of environment structure and motion allows to recover the perception of depth, otherwise lost from a single perspective view, using multiple images taken from different viewpoints .
The main shortcoming of vision-based tracking systems is the slow acquisition rate, which is due to both the physics of the image acquisition process and the computational workload of the computer-vision algorithms, especially those used to detect the visual features in each image frame.
The consequence is that vision-based tracking systems lack robustness against fast motion dynamics, which may easily lead to loss of visual features. Another difficulty with vision-based tracking systems is that the line of sight between the camera and objects within its FOV must be preserved as much as possible, in other words vision-based tracking systems are severely prone to problems of occlusions.
Inertial-based tracking systems integrate Inertial Measurement Units (IMUs) that incorporate accelerometers and gyroscopes for measuring translational accelerations and angular velocities of the objects they are affixed to with high sampling rates; this feature Drug_discovery makes them ideally suited to capture fast motion dynamics.
Being internally referenced and immune to shadowing and occlusions, inertial sensors can track body motion, in principle, without restrictions in space. Unfortunately, measurements of linear accelerations and angular velocities Carfilzomib are affected by time-varying bias and wideband measurement noise of inertial sensors.
Accurate estimates of body orientation in the three-dimensional (3D) space can be produced using quite complex filtering algorithms, sometimes with the addition of magnetic sensors that sense the Earth’s magnetic field to help producing drift-free heading estimates ; conversely, the 3D body position can be accurately estimated in tracking systems operating in a single IMU configuration only within temporally limited intervals of time, unless specific motion constraints are known and exploited to mitigate the double-time integration errors of gravity-compensated measured accelerations. The latter approach has been successfully implemented in strap-down inertial navigation systems (INS) for applications of pedestrian navigation [11,12].