3D machine vision is a system that recognizes the positions and postures of machines and parts in three dimensions. One of the challenges of conventional machine vision is that in order for robots to work efficiently, it is necessary for humans to correct processes that robots are not good at.

3D machine vision solves this problem, which has been squeezing the time and work of the process for some time, by changing the image processing method from 2D processing to 3D processing.

Usage of 3D machine vision

Here, we will introduce how to use 3D machine vision, especially 3D machine vision, which enables 3D image processing, from machine vision, which was mainly used for 2D image processing.

Bulk parts that are picked up one by one from the parts that are randomly stacked in a specific area such as a parts box is difficult to work with robots alone with conventional machine vision. It was necessary to manually change the arrangement of parts, but 3D image processing has made it possible to automate this process and improve efficiency.

In addition to picking and loading and unloading robots related to production, it is also used for autonomous driving technology, monitoring in the medical industry, etc., and the range of applications is expanding as a technology that is expected to develop further in the future.

Principle of 3D machine vision

Next, we will explain the principle of 3D machine vision. 3D machine vision can be broadly divided into pattern projection, distance measurement, and component recognition steps. First, in the process of pattern projection and distance measurement, multiple patterns are projected onto the object and the distance of the object is measured.

Next, in the parts recognition process, the position of the parts is recognized by the dictionary data registered in advance and the 3D CAD model, and it is judged whether or not the hand of the robot arm can operate without touching anything other than the object.

Finally, the judgment result is transferred to the robot controller, and the robot actually performs the operation.

Traditional robot vision and 3D machine vision

Most of the conventional robot visions have been used for systems such as acquiring flatly placed parts with a robot arm. This is a system in which parts that are aligned in a two-dimensional plane with a certain degree of order are imaged and image-processed with a vision, and the position shift and phase shift of the parts are canceled and picked up by the robot arm.

However, in recent years, there has been a demand for a system that picks up parts piled up in bulk with a robot arm. However, the conventional robot vision system built on the premise of a two-dimensional plane cannot handle this. Therefore, the 3D vision system attracted attention.

Challenges for 3D machine vision

The challenge with 3D machine vision is to be “vulnerable to disturbances and small errors.”

For example, consider the task of picking up parts that are stacked separately in a parts box. The inside of the parts box is photographed with 3D machine vision, and the parts that can be acquired by the robot by image processing are identified. At this time, if the position of the lighting that illuminates the parts is changed as a disturbance and the amount of light is changed, a problem may occur in which the parts that were previously visible in 3D machine vision cannot be seen. Lighting is such an important factor for 3D machine vision, and if the conditions are broken by some kind of disturbance, the implemented algorithms cannot handle it, and the system may not be viable.

Strictly speaking, all the parts in the parts box have different shapes. This is because each part has small scratches and slight dimensional errors. If the parts are aligned neatly, there is often no effect even if there is such a difference, but it may not be possible to judge that the parts are correct depending on the angle and position of the parts in bulk. It is. In that case, parts that should be treated as normal products will be treated as abnormal products, and optimum production will not be possible.

In recent years, it has been considered to utilize AI technology for 3D machine vision in order to overcome such problems.