Machine learning-based multiobject tracking for sensor-based sorting - Neural network system integration and preliminary results

Aug 5, 2026

Upload by Logic

Sensor-based sorting enables granular materials separation by using sensor observations for classification and actuator decisions, yet line-scanning systems provide only a single observation and lack movement information. Area-scan cameras allow observing each object over multiple time points, enabling multiobject tracking for followed-path estimation and per-object motion modeling. This paper develops a neural-network-based multiobject tracking system integrated into a laboratory sorting setup. Preliminary results indicate accuracy comparable to an optimized Kalman filter approach and reduce manual motion-model tuning through data-driven parameter learning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Motivation

Contribution

Machine learning-based multiobject tracking for sensor-based sorting Georg Maier1, Marcel Reith-Braun2, Albert Bauer3, Robin Gruna1, Florian Pfaff2, Harald Kruggel-Emden3, Thomas Lngle1, Uwe D. Hanebeck2, and Jrgen Beyerer1,4 1 Fraunhofer IOSB, Institute of Optronics, System Technologies and Image Exploitation, Karlsruhe, Germany 2 Intelligent Sensor-Actuator-Systems Laboratory, Karlsruhe Institute of Technology (KIT), Germany 3 Mechanical Process Engineering and Solids Processing (MVTA), TU Berlin, Germany 4 Vision and Fusion Laboratory (IES), Karlsruhe Institute of Technology (KIT), Germany Abstract Sensor-based sorting provides state-of-the-art solutions for sorting of granular materials. Current systems use line-scanning sensors, which yields a single observation of each object only and no information about their movement. Recent works show that using an area-scan camera bears the potential to decrease both the error in characterization and separation. Using a multiobject tracking system, this enables an estimate of the followed paths as well as the parametrization of an individual motion model per object. While previous works focus on physically-motivated motion models, it has been shown that state-of-the-art machine learning methods achieve an increased prediction accuracy. In this paper, we present the development of a neural network-based multiobject tracking system and its integration into a laboratory-scale sorting system. Preliminary results show that the novel system achieves results comparable to a highly optimized Kalman filter-based one. A benefit lies in avoiding tiresome manual tuning of parameters of the motion model, as the novel approach allows learning its parameters by provided examples due to its data-driven nature. Keywords Sensor-based sorting, machine learning, visual inspection, multiobject tracking G. Maier et al. 1 Introduction Sensor-based sorting provides state-of-the-art solutions for sorting of granular materials. This umbrella term describes a family of systems that enable the physical separation of individual objects from a material stream on the basis of information acquired by one or multiple sensors. Among other fields of application, it is considered a key technology for achieving a circular economy. In distinction to mechanical sorting processes such as screening, wind sifting, or float/sink processes, the technology is sometimes also referred to as indirect sorting [1], since particle classification and separation are performed in separate steps. In theory, any number of classes can be recognized for sorting, and separation into multiple fractions is also possible in principle. In industrial applications, however, the task is preferably implemented asa binary sorting task, i. e., sorting into “product” and “residue”, since multi-way sorting requires complex mechanical handling. The functional principle can be summarized as follows. First, the material is fed into the system by means of a conveyor mechanism. Subsequently, the material is transported further via a transport medium. In the course of the transport, sensor-based data acquisition takes place. The data collected is evaluated with the goal to detect and classify individual particles in the material stream. The result of the classification is the basis for the sorting decision, which is executed by means of an actuator. A particular strength of the sorting technology lies in the variety of industrially available sensors that are suitable for use in sensor-based sorting systems. This results in great flexibility with regard to the detectable material properties and thus the sorting criteria to be applied. Due to their suitability for systems with high material throughputs, imaging sensors dominate at this point. 1.1 Motivation Current systems use line-scanning sensors, which is convenient as the material is perceived during transportation. In case sorting criteria based on color, shape or texture suffi

FAQs

What limitation do line-scanning sensor systems have for sensor-based sorting?

They provide only a single observation of each object during transportation and do not capture object movement, requiring assumptions for timing and location at separation.

How does an area-scan camera improve sensor-based sorting?

With a sufficiently high frame rate, objects are observed at multiple time points, enabling multiobject tracking to estimate paths and parameterize per-object motion models for predictive actuation.

What is the main contribution of the proposed work?

A neural-network-based multiobject tracking system is developed and integrated into a laboratory-scale sorting system with an area-scan camera, aiming to avoid manual tuning by learning motion-model parameters from examples.

Aug 5, 2026

Upload by Logic