Mind Map

Content

Instructions for Creating Workpiece Point Cloud Templates

I. Preconditions

Precondition	Definition	Notes
Workpiece Pose Variation	Finite poses (countable): only fixed surfaces of the workpiece can be seen (such as the front side), and the valid Point Cloud region is the area on that surface that remains stably visible. Arbitrary poses: the workpiece may appear at any angle, and the valid Point Cloud region is the area that can be stably imaged in more than 80% of poses (excluding portions where Point Cloud is easily missing, such as deep holes, overexposed areas, grooves, etc.).	Ordered scenes: The template should rely only on the Point Cloud of fixed visible surfaces to avoid matching failure caused by invisible surfaces. Unordered scenes: Stable visible regions must be determined through multi-view scanning or CAD simulation. Exclude regions where Point Cloud is easily missing (such as reflective surfaces and deep concave structures) to avoid relying on unreliable Point Cloud. Confirmation method: Capture Point Cloud under different poses, count the visibility frequency of each region, and screen out the valid Point Cloud regions.
Camera Point Cloud Imaging Quality	The quality and completeness of the Point Cloud data obtained after recognizing the target workpiece with a 3D camera (such as structured light, stereo vision, etc.).	Point Cloud Missing Common causes: Workpiece properties (high reflectivity or translucent materials prevent laser return). Ambient light interference (strong illumination causes sensor saturation). View occlusion (self-occlusion of the workpiece or occlusion by external objects). Optimization strategies: Adjust lighting: use diffuse light or polarized filtering to reduce reflective interference. Local template: use Point Cloud from visible regions with better quality as the template. Discontinuous Point Cloud depth and high noise Common causes: Workpiece properties (multi-layer structures, transparent/semi-transparent materials). Camera exceeds the disparity range (Point Cloud distortion caused by measurement distance being too close/far). Occlusion from stacked objects (for example, stacked workpieces causing mixed Point Cloud). Optimization strategies: Filtering and denoising: use filtering and outlier removal functions to optimize Point Cloud quality. Depth calibration: ensure that the camera acquires data within the optimal working distance. Layered analysis: extract Point Cloud by regions for stacked workpieces to avoid mismatching.

II. Workpiece characteristics

Attribute (higher priority means higher importance)	Definition	Notes
Size	The physical size of the workpiece (volume, area) affects the Point Cloud acquisition density and the computational complexity of the template.	Large workpieces: It is recommended that the number of points in the Point Cloud Template not exceed 50000 (whether to downsample or cut at the original density and use it separately as a fine registration template should be determined based on accuracy requirements), so as to avoid wasting computational resources. Prioritize regions with obvious features (such as edges and hole positions) for dense Point Cloud sampling. The Point Cloud density can be reduced in purely planar areas to improve registration efficiency. Small workpieces: Ensure camera imaging quality and ensure that Point Cloud details are complete. If the visible Point Cloud is too sparse (fewer than 200 points), reduce the camera height appropriately to increase Point Cloud density.
Self-occlusion	The structure of the workpiece itself causes some regions to fail to form Point Cloud under certain views (such as multi-layer nesting, stacked workpieces, complex curved surfaces, etc.).	Point Cloud breaks or noise are likely to occur at occlusion edges. It is recommended that the Point Cloud Template directly use the Point Cloud processed by the PickWiz system Stable features that are not sensitive to occlusion (such as outer contours and fixed hole positions) should be selected as the main part of the template, and time-varying occluded regions should be ignored. If severe occlusion exists in a single view, local Point Cloud should be acquired from multiple views to create the template.
Three-dimensionality	The degree of depth variation of the workpiece in 3D space (such as height differences, stepped structures, curved-surface undulations, etc.).	Regions that can present Point Cloud should be used as the standard for screening valid template regions, so as to avoid interference from back-side Point Cloud. The Point Cloud density can be increased in edges and corners, the camera Point Cloud density can be retained in curved regions, and sparse sampling can be used in planar regions to balance accuracy and efficiency. Use no more than 50000 points as the baseline, and adjust template Point Cloud density on this basis.
Symmetric	The workpiece has a symmetric structure (such as bilateral symmetry or rotational symmetry).	Keypoint calibration: if a keypoint model is used, the Point Cloud Template and the CAD center point must be strictly aligned to avoid registration offset caused by mirror errors.
Concave	The surface of the workpiece has recessed structures (such as inner holes and grooves), which may cause Point Cloud missing or unstable imaging.	Outer contour first: prioritize extracting complete and visible outer contours as the main body of the template. Stability considerations: Side Point Cloud: stability under different views must be evaluated, and the most stable visible regions should be selected. Bottom Point Cloud: visibility is usually poor, so reliance should be reduced or only key regions should be used. Simplified template: if the Point Cloud quality in groove regions is poor, the template can be created using only edge contours plus a small number of feature points.
Convex	The surface of the workpiece has convex curved structures, which may lead to unstable Point Cloud imaging.	Reflection issue: convex curved surfaces easily cause edge reflections, making Point Cloud imaging quality unstable. The template should prioritize complete Point Cloud regions. Prioritize strong features: if the overall Point Cloud quality is poor, only highly distinctive features (such as marked points and edges) may be retained for registration.
Planar	The surface of the workpiece is approximately planar and lacks obvious internal features.	Edge enhancement: distribute Point Cloud more densely along the edges to enhance the guidance capability of registration. Compensation for missing tilted Point Cloud: If workpiece tilting causes blurred imaging of edge Point Cloud, a keypoint model can be used to assist registration instead. Or Point Cloud quality can be improved by optimizing lighting from multiple angles.
Dynamic update status	/	For workpieces that are prone to wear or deformation, changes in occlusion and three-dimensionality must be periodically re-evaluated and the template must be updated.

III. Basic principles

When using modeling for matching in industrial robot 3D vision, feature selection should follow the following principles
Characteristic	Description
Saliency	Select features that are prominent and easy to recognize on the object surface, such as edges, corners, textures, and concave/convex surfaces. These features should have strong distinguishability so that they can be quickly detected and matched in the scene.
Stability	Features should maintain a certain degree of stability against changes in lighting and viewpoint. Even if the object is rotated, scaled, or deformed, these features should remain relatively unchanged
Uniqueness	Features should be sufficiently distinctive and not easily confused with features of other objects or the environment. This helps improve the accuracy and reliability of matching.
Density	Distribute feature points as densely as possible on the object surface to increase matching redundancy. In this way, even if some feature points are occluded or lost, matching can still be completed by relying on other feature points.
Repeatability	Select features that can be stably detected under different viewpoints and lighting conditions. This ensures that the features can be accurately recognized in various scenarios.
Computational efficiency	The extraction and description process of features should be as efficient as possible to meet the real-time response requirements of industrial robots. Some features with relatively low computational cost can be selected.

IV. Notes for targeted scenarios

The general considerations for Point Cloud Template creation under different industrial scenarios and technical-path backgrounds are as follows

Scenario	Scene Image	Point Cloud Image Scene screenshot Point Cloud file	Template Direct capture image (file + image) Final template (file + image)	Point Cloud Template notes: Preconditions x Workpiece Characteristics - Basic Principles - Object Selection - Specific Form	Mitigation plan for notes
Generic Ordered				Preconditions: Workpiece characteristics: large elongated part (may deform), planar workpiece. Basic principles:	Preconditions: Workpiece characteristics:
Generic Unordered			Direct capture image	Preconditions: Workpiece characteristics: self-occlusion, concave shape (with round holes). Basic principles:	Preconditions: Workpiece characteristics:
Surface-Type Ordered				Preconditions: Workpiece characteristics: three-dimensionality (height difference, curved-surface undulation). Basic principles:	Preconditions: Workpiece characteristics:
Surface-Type Ordered	Process tolerances cause the spatial position of the small can to fluctuate up and down			Preconditions: Workpiece characteristics: three-dimensionality (height difference, curved-surface undulation). Basic principles:	Preconditions: Workpiece characteristics:
Surface-Type Unordered				Preconditions: Workpiece characteristics: symmetric (rotational symmetry), self-occlusion. Basic principles:	Preconditions: Workpiece characteristics:
Generic Unordered				Preconditions: Based on the scene placement and workpiece characteristics, front and back templates should be used to create the front Point Cloud and back Point Cloud of the workpiece separately, and keypoints should be taken from the overall workpiece. Workpiece characteristics: The workpiece is convex, and convex curved surfaces easily cause reflections, resulting in unstable Point Cloud imaging. Basic principles:	Preconditions: Use PickWiz software to separately generate the front and back Point Cloud templates (current view only), and generate keypoints for the overall workpiece. Workpiece characteristics: Use dual templates based on scene Point Cloud, and the templates must fit the front and back sides of the CAD
Surface-Type Positioning and Assembly (Matching Only)				Preconditions: Workpiece characteristics: Basic principles:	Preconditions: Workpiece characteristics:
Generic Positioning and Assembly				Preconditions: Workpiece characteristics: Basic principles:	Preconditions: Workpiece characteristics:

Previous pagePoint Cloud Template Creation Guide

Next pageScene Object Usage Guide