How to Avoid Inaccurate Picking After Project Stabilization

1. When calibrating external parameters, avoid using the offset mechanism unless you fully understand how it accounts for inaccuracies

:white_check_mark: Correct Approach

  1. If you are using Mech-Vision 1.8.0, the software by default does not use the offset. In this case, the camera’s offset parameters are decoupled from the Mech-Vision software.

  2. For versions before Mech-Vision 1.8.0, Mech-Vision reads the camera’s offset parameters. Therefore, when deploying the project, update the camera offset to the identity matrix, which is equivalent to not using the offset. For specific instructions, refer to Quick Reset Tool of Hand-Eye System Compensation Parameters for Mech-Vision <1.8.0.

  3. In very rare cases, offset is only used to compensate for the errors of a robotic arm in situations where the accuracy of the robotic arm is extremely poor. In such cases, it is necessary to strictly follow the calibration method with offset. Moreover, the preferred solution to address the accuracy issues of the robotic arm is to recalibrate the zero point of the arm and reconfigure the robotic arm through various means.
    The compensatory effect of offset on robotic arm errors is far inferior to the methods mentioned above, and it comes with significant stability risks. It is considered a last resort when no other options are available. Therefore, using offset requires a thorough understanding of its principles and calibration methods to take full responsibility for potential picking accuracy issues.

:x: Incorrect Approach

  1. Avoid using the default offset parameters of the camera, as doing so will reduce the redundancy of picking error and lead to more frequent instances of inaccurate picking. Calibrating the offset requires accurate parameter settings and sufficient sampling of corresponding calibration board poses. If the sampling requirements cannot be met, calibrating the offset will increase system errors, reduce error redundancy, and increase the likelihood of future inaccuracies in picking.

Two Misconceptions

  1. Does successfully picking objects after calibrating the extrinsic parameters with offset mean that the offset is correct?
    No, it is not correct. Most picking tasks have significant error redundancy, so the ability to pick objects after the initial calibration of extrinsic parameters does not guarantee the correctness of the offset. The accuracy of the offset depends on the calibration method. Generally, the methods used for extrinsic parameter calibration do not meet the requirements for offset calibration, making them incorrect.

  2. Does selecting the option to recalculate compensation parameters and achieving smaller residuals during extrinsic parameter calibration mean that recalculating compensation parameters is a better choice?
    No, quite the opposite. Selecting the option to recalculate compensation parameters might decrease system stability, leading to frequent inaccuracies in picking. The residuals in extrinsic parameter calibration refer to residuals in calibration poses and do not represent actual picking errors. Smaller residuals might be a result of overfitting. In most cases, the calibration poses used during extrinsic parameter calibration are insufficient for calculating the offset accurately. Offset requires dense pose sampling. Therefore, recalculating compensation parameters is not a better choice in most situations, and avoiding the use of offset is a more robust approach.

Reference for the Principles, History, and Specifics of Offset

2. Camera Preheating

:white_check_mark: Correct Practices

  1. To minimize inaccuracies, the camera should be preheated for at least 30 minutes when calibrating extrinsic parameters, intrinsic parameters, or conducting rapid internal parameter correction, as well as calibrating other compensation parameters. For high-accuracy applications, it is recommended to preheat the camera for 45 minutes or more. The preheating process can be achieved by powering on the camera or continuously capturing images.

:x: Incorrect Practices

  1. Avoid immediately starting to use the camera when it is not powered on. In such cases, the camera has not yet reached a stable state, leading to insufficient stability. Even if stable captures are possible at that moment, it poses a risk to the long-term stability of the system.

3. After Stable Project Operation: Addressing Inaccuracies (To be updated)

:white_check_mark: Correct Practices

  1. Once inaccuracies are observed after a period of stable operation, start by examining the camera accuracy. If the scalar difference significantly exceeds the standards
    During the correction, ensure that the calibration board is positioned at the middle depth of field of the camera, with its distance from the camera being 2/3 of the nearest working distance and 1/3 of the farthest working distance. It is advisable to position it closer to the farthest working distance as longer distances are less stable than shorter ones.
  2. After rapid intrinsic parameter correction, during the external parameter calibration, pay attention to using the unit matrix for the offset. For specific details, refer to the first part of this document.

:x: Incorrect Practices

  1. When inaccuracies occur after a stable operation period, avoid solely recalibrating the camera’s extrinsic parameters without checking the camera accuracy. This is because extrinsic parameters represent rigid body transformations, capable of correcting linear errors only, leaving nonlinear errors unresolved. Nonlinear errors can be reduced by rapidly correcting the camera’s intrinsic parameters or by recalibrating the camera. It is important to note that the effectiveness of rapid intrinsic parameter correction is generally inferior to recalibrating the entire camera system.

  2. When the camera accuracy meets the factory standards, avoid quick correction of the camera’s intrinsic parameters. Rapid adjustment of intrinsic parameters uses calibration board of specifications that do not generally comply with the normal camera calibration process, potentially causing a decrease in the camera’s accuracy.