Fig 4 - uploaded by Jonas Stenzel
Content may be subject to copyright.
Source publication
In this study, a novel approach for the detection of parcel loading positions on a pallet is presented. This approach was realized as a substantial change in comparison with traditional system design of contour detection in de-palletizing processes. Complex 3D-vision systems, costly laser scanners or throughput decreasing local sensor solutions int...
Contexts in source publication
Context 1
... contemporary cost for the MESA sensor exceed the ifm camera about 4 times. Figure 4 pictures the detection results of the two cameras in comparison. As expected, the MESA sensor accomplishes a higher accuracy (less than 1 cm) compared with the ifm sensor (less than 2.6 cm). ...
Context 2
... number of points per parcel face was set to s = 100. The results demonstrate the invariance of the parcel position towards the detection rate Trade-off: Parcel Dimensions vs. Accuracy Figure 14 depicts the accuracy in terms of varying parcel dimensions. More specifically, the size of the generated cube varied from 0.01 to 1 m and n was set to 1500. ...
Context 3
... of points per parcel face was set to s = 100. The results demonstrate the invariance of the parcel position towards the detection rate Trade-off: Parcel Dimensions vs. Accuracy Figure 14 depicts the accuracy in terms of varying parcel dimensions. More specifically, the size of the generated cube varied from 0.01 to 1 m and n was set to 1500. In Fig. 14(a) Θ was set to 10 mm and in Fig. 14(b) Θ = 30 mm was used. Additionally, s was set to 100 (solid) and 1000 (dashed). MSAC fails if the parcel length approximately approaches 23 cm while RANSAC already fails when the parcel length reaches 35 cm (s. Fig. 14(b)). Figure 14(a) roughly shows the same results except that both RANSAC and ...
Context 4
... s = 100. The results demonstrate the invariance of the parcel position towards the detection rate Trade-off: Parcel Dimensions vs. Accuracy Figure 14 depicts the accuracy in terms of varying parcel dimensions. More specifically, the size of the generated cube varied from 0.01 to 1 m and n was set to 1500. In Fig. 14(a) Θ was set to 10 mm and in Fig. 14(b) Θ = 30 mm was used. Additionally, s was set to 100 (solid) and 1000 (dashed). MSAC fails if the parcel length approximately approaches 23 cm while RANSAC already fails when the parcel length reaches 35 cm (s. Fig. 14(b)). Figure 14(a) roughly shows the same results except that both RANSAC and MSAC require a rela- tively larger ...
Context 5
... the size of the generated cube varied from 0.01 to 1 m and n was set to 1500. In Fig. 14(a) Θ was set to 10 mm and in Fig. 14(b) Θ = 30 mm was used. Additionally, s was set to 100 (solid) and 1000 (dashed). MSAC fails if the parcel length approximately approaches 23 cm while RANSAC already fails when the parcel length reaches 35 cm (s. Fig. 14(b)). Figure 14(a) roughly shows the same results except that both RANSAC and MSAC require a rela- tively larger parcel length in order to work properly. This may be justified by the reduced accuracy which is associated with the higher value of Θ (allowing more points to be used for setting up the consensus set). The impact of s is ...
Context 6
... 14(b)). Figure 14(a) roughly shows the same results except that both RANSAC and MSAC require a rela- tively larger parcel length in order to work properly. This may be justified by the reduced accuracy which is associated with the higher value of Θ (allowing more points to be used for setting up the consensus set). ...
Similar publications
This paper analyzes the market integration process of nominal prices, develops a model to analyze market integration, and presents a test of increasing market integration. A distinction is made between the economic concepts of price convergence in mean and variance. When both types of convergence occur, prices are said to converge in distribution....
![Figure 6. Scheme of the predictive combustion model reported in [8].](publication/321394116/figure/fig2/AS:631650622906384@1527608652153/Scheme-of-the-predictive-combustion-model-reported-in-8_Q320.jpg)
The present study is focused on the development of a new control-oriented semi-empirical model to predict nitrogen oxide (NOx) emissions in a light-duty diesel engine under both steady-state and transient conditions. The model is based on the estimation of the deviations of NOx emissions, with respect to the nominal engine-calibration map values, a...
Real convergence of economies in macroeconomic terms is accompanied by convergence of nominal variables. The paper provides an analysis of the relationship of the GDP per capita development as an indicator of the real convergence and the development of the comparable price level as a representative of the nominal convergence in relation to the EU a...
Citations
... A model fitting algorithm for detecting payload in the form of pellets, followed by estimating the position of the pellets in a frontal view is presented in [13]. Weichert, Skibinski et al. [14] propose a similar approach to detect box-shaped payloads on euro pellets in point clouds obtained using 3D ToF cameras, and present successful results for objects represented with a sufficient number of points. An algorithm for detecting cardboard packages from RGB-D images by fitting cuboid models to detected box faces is described in [15]. ...
This paper presents a hybrid algorithm for real-time instance segmentation of packages from scenes represented by 2D distance maps (range images). The paper introduces a novel approach combining deep learning-based methods and digital signal processing methods to enable accurate package recognition, using a small training dataset with high variability and distance measurement errors characteristic of Time-of-Flight-based scanning. Two convolutional neural networks with architecture optimized for training with a limited number of samples perform an initial segmentation of package components (sides and edges). An algorithm based on digital signal processing methods performs refinement of intermediate results, and combines package components into packages. Training and evaluation of the algorithm were performed on a custom dataset containing scenes of packages, shipping bags, and packaging of irregular shapes with various sizes, orientations, and degrees of occlusion, organized either in ordered stacks or arbitrary order. The convolutional neural networks provide a reliable distinction between components of packages and components of other types of packaging and surroundings. Package sides containing a sufficient number of distance points are correctly combined into packages. Thus, the proposed algorithm represents a solid basis for fully automated loading/unloading of packages with arbitrary sizes and materials from transport trailers and storage spaces. The dataset and annotations for box side surfaces are available at: https://dipteam.feit.ukim.edu.mk/results-package-detection.html.
... The authors conclude that the PMD system is characterised by a greater accuracy than a typical stereo camera system. On the basis of such an insight, a solution for pallet loading and de-palletising detection employing a PMD camera has been introduced in [67]. Again, such approach requires the introduction of ad hoc hardware, at the expense of cost and maintenance. ...
The problem of autonomous transportation in industrial scenarios is receiving a renewed interest due to the way it can revolutionise internal logistics, especially in unstructured environments. This paper presents a novel architecture allowing a robot to detect, localise, and track (possibly multiple) pallets using machine learning techniques based on an on-board 2D laser rangefinder only. The architecture is composed of two main components: the first stage is a pallet detector employing a Faster Region-Based Convolutional Neural Network (Faster R-CNN) detector cascaded with a CNN-based classifier; the second stage is a Kalman filter for localising and tracking detected pallets, which we also use to defer commitment to a pallet detected in the first stage until sufficient confidence has been acquired via a sequential data acquisition process. For fine-tuning the CNNs, the architecture has been systematically evaluated using a real-world dataset containing 340 labelled 2D scans, which have been made freely available in an online repository. Detection performance has been assessed on the basis of the average accuracy over k-fold cross-validation, and it scored 99.58% in our tests. Concerning pallet localisation and tracking, experiments have been performed in a scenario where the robot is approaching the pallet to fork. Although data have been originally acquired by considering only one pallet as per specification of the use case we consider, artificial data have been generated as well to mimic the presence of multiple pallets in the robot workspace. Our experimental results confirm that the system is capable of identifying, localising and tracking pallets with a high success rate while being robust to false positives.
... Forks with an integrated weighing system [38] provide a measure to ensure the forklift is not overloaded. It is a simple assistance system for handling and easy to install and use because only the forks have to be replaced. ...
A forklift is a piece of common equipment used to move, lift and transload goods and materials in warehouses, distribution centers and manufacturing facilities. Handling errors by the forklift driver, insufficient safety awareness, improper workplace design or a combination of these factors can cause injuries or fatal accidents. Sensor-based assistance systems are able to minimize safety risks and increase work efficiency. However, there is no comprehensive overview and no structuring scheme of sensor-based assistance systems for forklifts. For this reason, we provide a classification scheme of assistance systems for forklifts. We also offer a proposal for choosing forklift assistance system according to the logistics processes that may have a huge potential for error.
... This can be achieved by using depth cameras and methods of object recognition and pose estimation [51]. Similar applications are de-palletizing of goods [52] and robot-based order picking in warehouses or distribution centers [53]. Such technologies are also suitable to control conveyors, which perform complex intralogistics tasks such as package layering or singulation of packages [54] [55]. ...
In this article, applications and trends of image processing in logistics will be presented. The term image processing refers to the entirety of systems that capture images and operations that are applied to images and either produce enhanced images or extract data from the images. The information captured by image processing is used to control or monitor logistics processes and thus contributes to the smartness of smart logistics zones. --- Published in: Schenk, Michael (Ed.): 11th International Doctoral Students Workshop on Logistics, June 19, 2018, Magdeburg: Institut für Logistik und Materialflusstechnik an der Otto-von-Guericke-Universität Magdeburg. ----
... The authors conclude that the PMD system is characterised by a greater accuracy than a typical stereo camera system. On the basis of such an insight, a solution for pallet loading and de-palletising detection employing a PMD camera has been introduced in [59]. ...
The problem of autonomous transportation in industrial scenarios is receiving a renewed interest due to the way it can revolutionise internal logistics, especially in unstructured environments. This paper presents a novel architecture allowing a robot to detect, localise, and track multiple pallets using machine learning techniques based on an on-board 2D laser rangefinder. The architecture is composed of two main components: the first stage is a pallet detector employing a Faster Region-based Convolutional Neural Network (Faster R-CNN) detector cascaded with a CNN-based classifier; the second stage is a Kalman filter for localising and tracking detected pallets, which we also use to defer commitment to a pallet detected in the first step until sufficient confidence has been acquired via a sequential data acquisition process. For fine-tuning the CNNs, the architecture has been systematically evaluated using a real-world dataset containing 340 labeled 2D scans, which have been made freely available in an online repository. Detection performance has been assessed on the basis of the average accuracy over k-fold cross-validation, and it scored 99.58% in our tests. Concerning pallet localisation and tracking, experiments have been performed in a scenario where the robot is approaching the pallet to fork. Although data have been originally acquired by considering only one pallet, artificial data have been generated as well to mimic the presence of multiple targets in the robot workspace. Our experimental results confirm that the system is capable of identifying, localising and tracking pallets with a high success rate while being robust to false positives.
... The authors conclude that the PMD system had a greater accuracy than a stereo camera system. The paper [23] presented a solution for load detection and de-palletizing using a PMD camera. Moreover, other vision-based methods for pallet recognition and pose estimation have been presented in [24][25][26][27][28][29]. ...
The problem of developing an autonomous forklift that is able to pick-up and place pallets is not new. The same is true for pallet detection and localization, which pose interesting perception challenges due to their sparse structure. Many approaches have been presented for solving the problems of extraction, segmentation, and estimation of the pallet based on vision and Laser Rangefinder (LRF) systems. Here, the focus of attention is on the possibility of solving the problem by using a 2D LRF. On the other hand, machine learning has become a major field of research in order to handle more and more complex detection and recognition problems. The aim of this thesis is to develop a new and robust system for identifying, localizing, and tracking the pallets based on machine learning approaches, especially Convolutional Neural Network (CNN). The proposed system is mainly composed of two main components: Faster Region-based Convolutional Network (Faster R-CNN) detector and CNN classifier for detecting and recognizing the pallets, and a simple Kalman filter for tracking and increasing the confidence of the presence of the pallet. For fine-tuning the proposed CNNs, the system is tested systematically on real-world data containing 340 labelled object examples. Finally, performance is evaluated given the average accuracy over k-fold cross-validation. The computational complexity of the proposed system is also evaluated. Finally, the experimental results are presented, using MATLAB and ROS, verifying the feasibility and good performance of the proposed system. The best performance is achieved by our proposed CNN with an average accuracy of 99.58% for a k-fold of 10. Regarding the tracking task, the experiments are performed while the robot was moving towards the pallet. Due to availability, the experiments are carried out by considering only one pallet, and consequently to check
the robustness of our algorithm, an artificial data are generated by considering one more pallet in the environment. It is observed that our system is able to recognize and track the positive tracks (pallets) among other negatives tracks with high confidence scores.
... Weichert, Skibinski et al. [14] propose a system that can detect payloads on euro pallets using 3D TOF cameras. Their system is capable of detecting individual packages on a pallet using a multiple stage model fitting algorithm that aligns simple box models with the observed data. ...
... Approaches for autonomous load handling use different types of sensors to obtain an understanding about the environment. In (Weichert et al., 2013) the authors discuss the advantages of several sensors for this task. We will group these approaches into two main categories based on the sensors used: range sensors and vision-based sensors (monocular or stereo cameras). ...
... In this section, a brief introduction to the state-of-the-art concerning depalletizing is given and applicable sensors for low-cost load detection are presented. A comprehensive survey of the technical background is given in [21] and [8]. ...
... Sect. 2.3) is broadly evaluated [21,10,9] and constitutes a competitive alternative in comparison to existing complex and costly 3D-vision systems. The presented solution reduces the costs of contour checking in the automated de-palletizing process from an economic point of view concurrently having a moderate run time (less than 1 s on the employed test system). ...
In this chapter, novel approaches for the detection of logistical objects (loading units) in the field of material flow applications are comparative presented, focusing on solutions using low cost 3D sensors. These approaches realize substantial changes in comparison to traditional system design of logistic processes. Complex 3D-vision systems, costly laser scanners or throughput decreasing local sensor solutions integrated in grippers are substituted by low cost Photonic Mixing Device (PMD) cameras or structured light sensors (like Asus Xtion or Microsoft Kinect). By using low cost sensors and modern point cloud processing algorithms for detection and classification in logistic applications like de-palletizing, automation of usually manual processes will be economically feasible. Besides the description of different basic solution concepts for 2.5D and 3D, two practical applications are presented.
... We will discuss related object detection methods and also provide an analysis of current systems in use. The papers [1], [2] discuss solutions for load detection and de-palletizing. Their work focuses on parcel detection and handling using a photonic mixing device (PMD) camera. ...
The paper presents a method for automatically detecting pallets and estimating their position and orientation. For detection we use a sliding window approach with efficient candidate generation, fast integral features and a boosted classifier. Specific information regarding the detection task such as region of interest, pallet dimensions and pallet structure can be used to speed up and validate the detection process. Stereo reconstruction is employed for depth estimation by applying Semi-Global Matching aggregation with Census descriptors. Offline test results show that successful detection is possible under 0.5 seconds.
