Single Camera Pick (Eye-in-Hand)

## Single Camera Pick (Eye-in-Hand)

Calculates and outputs the robot arm pick coordinates based on the workpiece position captured by the camera. (Camera position is not fixed, requires input of the camera position for each shot)
## Module Principle
This module is the same as the Single Camera Pick. It uses the captured image and optimized feature matching to locate the image pick point, then converts it to a mechanical point via calibration for the equipment to pick.
This module is for the **Eye-in-Hand** mode, meaning the camera's shooting position for each pick is not fixed and requires real-time input of the camera's shooting pose coordinates. Compared to the ordinary Single Camera Pick, this module has an additional **Robot Arm Shooting Pose** port for inputting the real-time changing camera shooting coordinates.

![](/media/202603/2026-03-12_171032_2634980.7497986021500556.png)

Mechanical models are divided into XYR, XR, YR (with angle) and XY, X, Y (without angle) types. The calculation methods for the two types of picking are different. Picking with an angle requires the input calibration data to also be of the XYR, XR, YR type with an angle. Picking without an angle has no requirement for calibration.
When the pick mode is multiple targets, both input and output are in array form. For the compensation amount in a mechanical model with angle R, the compensated R will affect the XY values. (That is, if only R is compensated, the XY position where the robot arm picks the workpiece will not change, but the actual mechanical XY coordinates will change slightly to compensate for the rotation.)

#### Teaching
Teaching is the process where an operator or engineer teaches the system how to identify target objects, understand the scene, or execute specific visual tasks before or during the picking process. The operation steps are: place a material on the picking plane, get its workpiece pixel point and workpiece pixel angle on the image, and record it by clicking **Learn Current Value**. Then manually align the robot arm with the teaching workpiece and record the picking pose to **Teach Pick Pose** and the current camera shooting pose to **Teach Shooting Pose**.

![](/media/202603/2026-03-12_171046_5693570.15972853144900956.png)
**Teaching** must be performed again every time a new pick module is created, calibration changes, or the robot arm's **pick position changes**.

## Usage Scenarios
Before performing picking or pick-and-place, the robot arm and camera calibration must be completed and introduced in the module.
Single camera picking is the most basic and commonly used technical solution in machine vision. It obtains 2D image information of the target object through a single industrial camera and combines image processing algorithms to achieve object identification, positioning, and picking.

## Usage Method

### Pre/Post-Sequence Modules

**Camera Image Acquisition**, **Optimized Feature Matching** to get the **workpiece image point coordinates** for single camera picking, **Calibration Reference** to introduce calibration for converting image point coordinates. Thread variable creation creates the camera shooting pose, or gets the real-time shooting position from the PLC via global variable reading.
After completing the port connections, run once first to learn the input image point coordinates, then manually align the mechanism to get the teach mechanical point input, and fill in the teach camera shooting pose at the same time. Run subsequently to get the pick coordinate output.

![](/media/202603/2026-03-12_171056_3930740.8569225152634734.png)

## Parameter Configuration Description
Mechanical Model:
【XYR】 Mechanical structure XY axes can translate and rotate.
【XY】 Mechanical structure XY axes can translate, but cannot rotate or do not need to rotate.
【XR (Y Separated)】 Mechanical structure only has X-axis translation and can rotate, Y-axis structure is separated.
【YR (X Separated)】 Mechanical structure only has Y-axis translation and can rotate, X-axis structure is separated.
【X (Y Separated)】 Mechanical structure only has X-axis translation, Y-axis structure is separated.
【Y (X Separated)】 Mechanical structure only has Y-axis translation, X-axis structure is separated.
【Pick Mode】 Single Target: Pick only once; Multiple Targets: Pick multiple targets in the field of view sequentially.
【Mechanical Travel】 Allowed travel range of the robot arm; exceeding the range will report an error.
【Compensation Amount】 Mechanical coordinates, compensated in the mechanical coordinate system. When clicking subscribe, the compensation amount can be introduced via the port.
【Reference Image】 The image taken by the camera during mechanical picking.

![](/media/202603/2026-03-12_171108_2866320.6266619496068543.png)

Mechanical travel represents the movable range of the mechanism. When the output absolute coordinates are greater than the mechanical travel, an error will be reported, but it will not affect the result output.
## Output Description

![](/media/202603/2026-03-12_171121_6458500.17881931810069318.png)
The output has two sets: Pick Pose and Offset Pose. Pick Pose is the actual picking pose of the robot arm, which is also the judgment range for [Mechanical Travel]. Offset Pose is the offset of the robot arm relative to the teaching position and does not judge mechanical travel.

| Feature | Pick Pose | Offset Pose |
|---|---|---|
| Coordinate Type | Absolute Coordinate (GPS style) | Relative Coordinate (Reference object style) |
| Safety Check | Full check of mechanical travel | No range check |
| Main Function | Final execution position | Position fine-tuning |
| Analogy | Detailed delivery address | "Front left of you" |
| Dependency | Absolute Coordinate (GPS style) | Contains teach coordinates + offset based only on teach coordinates |
## Precautions

1. When using a bottom camera, the X-axis needs to be mirrored because the coordinate system of the mechanism captured by the bottom camera is opposite to the actual one.
2. When using a pick type with an angle, the input calibration data is also required to be of the XYR, XR, YR type with an angle.
3. Every time a new pick module is created, calibration changes, or the robot arm's pick position changes, teaching must be performed again.
4. Learning the teach image point requires running once first to get the input before learning.
   
## Error Troubleshooting
1. When picking is inaccurate, prioritize reteaching and testing again.
2. Pay attention to whether the feature matching module acquires an accurate image.
3. If still unable to pick accurately, test the accuracy of the calibration (the calibration module introduces testing).
4. If there is a large offset angle leading to inaccurate XY, it may be caused by an error in the calculation of the calibration rotation center. Try to recalibrate, modify the calibration method, and improve the calculation of the rotation center (add calibration verification or use the long radius calibration module).
5. If the center is accurate but there is XY deviation around the periphery, consider image distortion (use the Image Distortion Correction module), or calibration scale calculation error (use the Calibration Correction module).

---

Click here to jump to the corresponding Chinese manual page: [Single Camera Pick (Eye-in-Hand)](/doc/143)