Elements, principles and cases of Point Cloud template production

About 2811 wordsAbout 9 min

mind map

content

TargetPoint Cloud template creation instructions

1. Preconditions

Precondition	define	Precautions
Target posture change	Limited postures (countable): only the fixed surface of the Target can be seen, and the effective Point Cloud area is the area where this surface is stably visible. Any posture: Target may appear at any angle. The effective Point Cloud area is the area that can stabilize imaging in more than 80% of postures (excluding deep holes, overexposed areas, grooves and other parts where Point Cloud is easily missing).	Ordered Scene: Disordered Scene: The stable visible area needs to be determined through multi-view scanning or CAD simulation. Exclude areas that are easily missing (such as reflective surfaces, deep concave structures) and avoid relying on unreliable Point Cloud. Confirmation method :
Camera imaging Point Cloud effect	The quality and integrity of the Point Cloud data obtained after identifying the target through 3DCamera (such as structured light, binocular vision, etc.).	Point Cloud is missing Ambient light interference (strong light exposure causes sensor saturation). Perspective occlusion (Target self-occlusion or external object occlusion). Optimization strategy : Partial template: Use Point Cloud with better visual effect as a template. Point Cloud depth is discontinuous and noisy Common reasons: Camera exceeds the parallax range (the measurement distance is too close/far resulting in Point Cloud distortion). Object stacking and occlusion (such as stacking Targets causing Point Cloud mixing). Optimization strategy : Depth calibration: Ensure that the Camera collects data within the optimal working distance. Hierarchical analysis: Extract Point Cloud from laminated Target regions to avoid mismatching.

The physical size (volume, area) of The degree of depth change of There are concave structures (such as inner holes and grooves) on the surface of There are concave structures (such as inner holes and grooves) on the surface of The surface of

Attribute (The higher the priority, the higher the priority)	definition	Notes
size	Target affects the Point Cloud collection density and template calculation complexity.	Large size Target: It is recommended that the number of Point Clouds in the Point Cloud template should not exceed 50000 (you need to decide whether to downsample or cut the original density and use it as a separate precision registration template according to the accuracy requirements). avoids waste of computing resources. preferentially selects areas with obvious features (such as edges and holes) for dense Point Cloud sampling. Pure planar areas can reduce Point Cloud density and improve registration efficiency. Small size Target: ensures Camera imaging quality and complete Point Cloud details. If there are too few visible Point Clouds (less than 200), the Camera height needs to be appropriately reduced to increase the Point Cloud density.
Self-shielding	The visualization result area displays 3D vision running results in real time and provides functions such as run, vision computing configuration, and historical data.	The occlusion edge is prone to Point Cloud breakage or noise. It is recommended that the Point Cloud template directly uses the Point Cloud processed by the pickwiz system should select stable features that are insensitive to occlusion (such as outer contours, fixed hole positions) as the template body, and ignore time-varying occlusion areas. If there is severe occlusion in a single viewing angle, it is necessary to collect partial Point Cloud from different viewing angles to create templates.
Stereometry	Target in three-dimensional space (such as height difference, step structure, surface relief, etc.).	should use the renderable Point Cloud area as the criterion for screening valid template areas to avoid interference from the Point Cloud on the back. Edge and corner areas can increase Point Cloud density, curved surface areas can retain CameraPoint Cloud density, and flat areas can be sparsely sampled to balance accuracy and efficiency. The Point Cloud volume is based on not exceeding 50,000, and the template Point Cloud density is adjusted under this baseline
Symmetrical type	Target has a symmetrical structure (such as left-right symmetry, rotational symmetry).	key point calibration: If you use a key point model, you need to strictly align the Point Cloud template and the CAD center point to avoid registration deviation caused by mirroring errors.
concave shape	Target, which may result in missing Point Cloud or unstable imaging.	Outer Contour Priority: Give priority to extracting the complete visible outer contour as the template body. stability considerations: Side Point Cloud: It is necessary to evaluate the stability under different viewing angles and select the most stable visible area. Bottom Point Cloud: Usually less visible, reduce dependencies or use critical areas. simplified template: If the quality of the Point Cloud in the groove area is poor, the template can be made with only the edge outline + a small number of feature points.
convex hull	Target, which may result in missing Point Cloud or unstable imaging.	reflection problem: convex curved surfaces can easily lead to edge reflections, and the Point Cloud imaging quality is unstable. The template should give priority to the complete Point Cloud area. Strong features are given priority: If the overall Point Cloud quality is poor, only high-discrimination features (such as marker points and edges) can be retained for registration.
Flat type	Target is approximately flat and lacks significant internal features.	edge enhancement: Centrally distribute Point Cloud at the edge to enhance registration and guidance capabilities. Tilt Point Cloud Missing Compensation: If the tilt of the Target causes blurred edge Point Cloud imaging, the key point model can be used to assist registration. or use multi-angle lighting to optimize Point Cloud quality.
dynamic update	/	For Targets that are prone to wear or deformation, occlusion and three-dimensional changes need to be re-evaluated regularly and the template updated.

3. Basic Principles

The extraction and description process of

When using modeling for matching in industrial Robot 3D vision, the following principles need to be followed when selecting features
Features	Description
Significance	selects those features that are prominent and easy to identify on the surface of the object, such as edges, corners, textures, concave and convex surfaces, etc. These features should be highly identifiable and easy to detect and match quickly in the scene.
Stability	features should have certain stability to changes in lighting, viewing angle, etc. Even if the object undergoes rotation, scaling and other deformations, these features can remain relatively unchanged
Uniqueness	features should be unique enough not to be easily confused with other objects or features in the environment. This is helpful to improve the accuracy and reliability of matching.
Intensity	distributes feature points as densely as possible on the surface of the object to increase the redundancy of matching. In this way, even if some feature points are occluded or lost, the matching can still be completed by relying on other feature points.
Repeatability	selects features that can be stably detected under different viewing angles and lighting conditions. This ensures that features can be accurately recognized in various Scenes.
Computational efficiency	features should be as efficient as possible to meet the needs of industrial robots for real-time response. Some features can be selected that require relatively little computational effort.

4. Precautions for Targeted Scene

Under the background of different industrial scenes and different technical paths, the general precautions for Point Cloud template production are as follows:

Scene	Scene diagram	Template direct image (file + picture) final template (file + picture)	Point Cloud template pays attention to the real preconditions x Target characteristics - basic principles - object selection - specific form	Precautions and solutions
Universal ordered			Prerequisites: Target characteristics: Large strips (possible deformation), flat Target Basic principles:	Prerequisite: Target characteristics:
Universal disorder		Straight image	Prerequisites: Target characteristics: self-occlusion, concave shape (with round hole) Basic principles:	Prerequisite: Target characteristics:
surface shape is ordered			Prerequisites: Target characteristics: three-dimensionality (with height differences, undulating surfaces) Basic principles:	Prerequisite: Target characteristics:
surface shape is ordered	Process error causes the spatial position of the small tank to fluctuate up and down		Prerequisites: Target characteristics: three-dimensionality (with height differences, undulating surfaces) Basic principles:	Prerequisite: Target characteristics:
surface disorder			Prerequisites: Target characteristics: Symmetry (rotational symmetry), self-occlusion Basic principles:	Prerequisite: Target characteristics:
Universal disorder			Prerequisites: According to the Scene placement and Target characteristics, front and back templates should be used to create the front Point Cloud and the back Point Cloud of the Target respectively. The key points are the overall key points of the Target. Target characteristics: Target convex hull type: convex surfaces can easily lead to reflections and the Point Cloud imaging quality is unstable. Basic principles:	Prerequisites: Use pickwiz software to generate front and back point cloud templates (current perspective), and generate key points of the entire workpiece Target characteristics: Use ScenePoint Cloud dual templates, and the templates must fit the front and back sides of the CAD
surface positioning assembly (matching only)			Prerequisites: Target Features: Basic Principles:	Prerequisite: Target characteristics:
Universal positioning assembly			Prerequisites: Target Features: Basic Principles:	Prerequisite: Target characteristics: