Skip to main content

The advantages of multi-object tracking

‘Recent breakthroughs in sensor-based sorting technology have made area scanning systems a feasible materials sorting solution,’ says Georg Maier of Germany’s Fraunhofer Institute. The next step, he believes, is multi-object tracking; which means information about individual objects present in the material stream can be ‘fused over time’ to reconstruct the trajectory of materials and, ultimately, boost sorting performance.

Multi-object tracking is a relatively new term, acknowledges Georg Maier of the Fraunhofer Institute of Optronics. The practice essentially enables predicting the position of each object for future points in time. ‘While conventional systems typically apply a rather simple motion model, our approach includes an individual motion model for each object. Based on the subsequent data, we can pinpoint at what time and at what position on the conveyor belt each different fraction will reach the separation stage,’ Maier explains.

Prototype specs

On behalf of Fraunhofer, he collaborated with various industry partners to fine-tune results, such as the Technical University of Berlin. ‘Together, we were able to create a prototype system extremely rapidly, which helped us obtain multiple observations for each scanned object to generate what is referred to as “ground truth data”,’ Maier comments.

The experiments were conducted using a camera of type Bonito CL-400 that allows frame rates up to approximately 192 fps. With respect to illumination, the team relied on a LED ring light with an inner diameter of 180 mm, while the prototype conveyor belt has a total length of 600 mm and width of 160 mm. Separation is performed by an array of 16 air jets whereas air released by a single jet spreads through five neighbouring nozzles.

1000+ objects scanned

Fraunhofer’s table-sized ‘robust and self-adaptive’ tracking system is provided unlabeled measurements as its input. Those measurements are represented by the centroid of the 2D projection of the objects contained in the material feed; these can be obtained by utilising image processing (pre-processing, segmentation, and connected component analysis).

In order to specify the location of all feedstock fragments, Maier’s team employed a standard Kalman filter with a constant velocity model. This yields an approximate position for each individual particle in the next camera frame. Maier is glad to report the researchers managed to track more than a thousand objects simultaneously with a temporal resolution of 100 fps.

High complexity?

He remarks that two approaches may be applied. ‘For instance, for frames containing a comparatively low number of objects to be tracked, we select a high quality algorithm with high complexity,’ Maier says. ‘On the other hand, for frames with a high load, we prefer to switch to an algorithm that is lower in complexity at the cost of lower quality with respect to the data associations.’

This is a teaser article from the upcoming edition of Recycling Technology

You might find this interesting too

Pollutec expo prepares for 40th anniversary
US army recycles plastics to repair military equipment

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Subscribe now and get a full year for just €119 (normal rate is €149) Subscribe