3D technologies are used when it is necessary to analyze the geometric characteristics of the object. The pixels of the images acquired with these technologies are three-dimensional because they also include depth information (z) in addition to the spatial coordinates (x and y).

Thanks to 3D technologies it is possible to estimate the real 3D shape and position of the object in space.

All artificial vision applications  can benefit from analysing the third dimension of the object.

3D technologies are differentiated into:

A laser projects a bright line onto the scene framed by the camera, so that the acquired images contain the lines given by the intersection between the laser plane and the surfaces of the objects.

Knowing the equation of the laser plane and the position of the camera, through a triangulation process, we obtain the information necessary to establish the spatial coordinates (x,y) and the altitude (z) of the points P belonging to the line projected on the surface of the object.

By scanning the object, then taking multiple images of the laser line moving across the object’s surface, it is possible to reconstruct the 3D structure of the scanned object.

There are several possible configurations for performing laser triangulation.

It is possible to scan the moving object with the laser stationary in a fixed position or move the laser r on the surface of the stationary object.

Depending on the surface characteristics of the objects, it is possible to position the camera-laser pair according to 3 different configurations:

Our robot-guided vision system for picking bulk pieces in boxes performs three-dimensional reconstruction of objects using a 3D laser scanning technique.

Structured light is the projection of a light pattern (plane, grid, etc.) onto a surface, rather than a uniform light. One or more cameras observe the deformations of the pattern on the surface and, using the geometric information of the light, extract geometric information from the illuminated scene.

Our company has developed an automated inspection system for weld seam detection that uses structured light technology for 3D reconstruction.

Advantages:

Stereo vision is a technology that uses two or more cameras from different viewpoints to capture the same scene, simulating human binocular vision.

Images of the same scene are acquired at the same time and the correspondence between the points of interest in the different images is found. By calculating the disparity between the two images (i.e., the difference in position between the points), the depth of each point can be found.

A 3D point cloud of the scene framed by the cameras is then constructed.

The industrial vision system for the automatic greasing of hams uses stereo vision technology for the complete three-dimensional reconstruction of the product surface, even in the presence of irregular shapes and complex textures.

Time-of-Flight sensors acquire 3D information by measuring the time it takes for emitted light (usually modulated LEDs or lasers) to reach the object and return to the sensor. Each pixel on the sensor measures a phase delay or return time, generating a real-time depth map.
ToF systems are particularly suited to dynamic environments, thanks to their ability to reconstruct complete 3D scenes in a single shot, allowing for immediate volumetric perception.

The photometric stereo technique reconstructs the three-dimensional structure of an object by acquiring multiple images with the same camera but under different illumination angles. By analyzing how the surface reflects light under different lighting conditions, it is possible to reconstruct the object’s 3D structure.

This technique is particularly suitable for detecting small defects such as scratches, small incisions or bubbles that are difficult to see with other techniques.

It was not born as an absolute metrological technique to obtain precise measurements, but as a technique to reveal shape and defects with extreme sensitivity.

This technique can be applied with both matrix vision sensors and linear vision sensors.

Fields of application:

Deep learning in machine vision is the set of techniques based on neural networks that allow vision systems to recognize defects, objects, patterns and complex features automatically, adaptively and highly accurately, overcoming the limitations of traditional algorithms.

In the case of 3D vision systems, neural networks are used to estimate depth or a point cloud starting from RGB or IR images, without dedicated 3D hardware.

Unlike traditional techniques, it does not require lasers, pattern projections, TOF sensors and it can work even using a single image acquired by a single camera.

Neural networks mimic the behavior of the human eye, which reconstructs depth using visual cues such as shadows, textures, edges, and reflections, thus learning the relationship between 2D images and 3D depth.