It is clear that increasing automation and robotization are making an important contribution to relieving employees of physically strenuous and repetitive tasks. Instead, people are taking over the monitoring, guidance and maintenance of robots and machines. However, these more demanding and appealing tasks are not the only consequence - companies also benefit from the additional time for innovation and further development, in addition to lower production costs, less downtime and increased quality.

Unfortunately, reality often does not yet meet these expectations. In order for robots to work reliably, they not only need a precisely defined task, the right size, power or a suitable gripper, but above all they must also recognize the respective parts - a core task of industrial image processing. In practice, however, the correct identification of the situation regularly presents these systems with immense challenges. Complex lighting and camera setups are therefore often used, individually customized for a specific task. In the past, this always required a great deal of experience and specialist knowledge in this area. And even then, success was not always immediate.

Anyone who has already implemented a few image processing projects knows that sensor and software limitations are difficult to avoid, especially with complex workpieces. As a result, the task is divided into manageable sub-areas for several camera/illumination pairs. This often results in extensive setups. There is no question that these image processing systems are expensive. This is particularly true when it comes to complex objects, such as those with a metallic sheen, which require precisely balanced imaging conditions.

To make matters worse, these extensive designs not only require expert design knowledge, but also need it for implementation and handling in day-to-day business. Simple and intuitive use by workers does not usually exist. The same applies when companies want to adapt their vision systems to new parts or even expand them to include other tasks. Due to the individual design, it is often simply impossible to continue using these systems. This considerably reduces planning reliability. With shrinking batch sizes and more frequent product iterations, this is a serious challenge. Accordingly, those responsible in companies take a critical view of automation solutions based on industrial image processing.

New image capture technology simplifies the process

The light field opens up a new dimension in automation here: compact 3D sensors based on a multi-camera array ensure robust image capture thanks to the unique technology. Light field sensors offer a whole range of advantages: less occlusion, reduction of gloss and reflections, for example on metal, plastic and fiber composites, precise and dense point clouds thanks to the multi-camera array, compact and lightweight, fast recording - no multiple exposures necessary.

The light field achieves all of this via the so-called plenoptic function P(θ, ζ, λ, t, _Vx, _Vy, _Vz). In addition to the light beam described by two angles (θ and ζ), the wavelength for color images λ, the time course t, it also contains the corresponding 3D coordinates _Vx, _Vy and Vz_{of the} camera position. This function can be used to reproduce any view at any time from any position and at any wavelength. The plenoptic function thus forms the basis for all further calculations for the use of light field technology - for example in industrial image processing. With the help of multi-camera arrays for recording and abstracting several image angles, this theory can be used in an industrially suitable manner.

Software must go hand in hand with hardware

In combination with high-performance software, this creates a new generation of machine vision solutions: complex, customized setups with confusing operation are replaced by systems that are quick, easy and intuitive to use. Solid image recognition using a light field makes it possible to solve various tasks with a compact multi-camera array combined into a sensor. Depending on the software configuration, these include bin picking and object handling, 3D scanning and AI quality inspection. These systems are equally easy to adapt and expand. What's more, new workpieces no longer require expert knowledge, but can be implemented in a very short space of time.

In this way, light field-based image processing, together with modern, user-friendly software, makes a valuable contribution to making industrial automation even more successful in the future. The intuitive self-serving approach transforms the role of employees from one of executing to one of planning and monitoring. In no time at all, they become experts for upcoming extensions and adjustments.

Mareike Schindler-Kotscha, Head of Marketing, HD Vision Systems