Human activity recognition (HAR) is a wide research topic in a field of computer science. Improving HAR can lead to massive breakthrough in humanoid robotics, robots used in medicine and in the field of autonomous vehicles. The system that is able to recognise human and its activity without any errors and anomalies would lead to safer and more empathetic autonomous systems. During this research work, multiple neural networks models, with different complexity, are being investigated. Each model is re-trained on the proposed unique data set, gathered on automated guided vehicle (AGV) with the latest and the modest sensors used commonly on autonomous vehicles. The best model is picked out based on the final accuracy for action recognition. Best models pipeline is fused with YOLOv3, to enhance the human detection. In addition to pipeline improvement, multiple action direction estimation methods are proposed.

/pdf/human-activity-recognition-based-path-planning-for-55wx8t43wu.pdf

Human activity recognition-based path planning for autonomous vehicles

Transportation management, as a part of the supply chain management, is a complex process that consists of planning and delivering goods to customers. The paper presents a complete multi-phase inte...

Adaptive multi-phase approach for solving the realistic vehicle routing problems in logistics with innovative comparison method for evaluation based on real GPS data

On using human activity recognition sensors to improve the performance of collaborative mobile manipulators: Review and outlook

Multiscale brings great benefits for people to observe objects or problems from different perspectives. It has practical significance for clustering on multiscale data. At present, there is a lack of research on the clustering of large-scale data under the premise that clustering results of small-scale datasets have been obtained. If one does cluster on large-scale datasets by using traditional methods, two disadvantages are as follows: (1) Clustering results of small-scale datasets are not utilized. (2) Traditional method will cause more running overhead. Aims at these shortcomings, this paper proposes a multiscale clustering framework based on DBSCAN. This framework uses DBSCAN for clustering small-scale datasets, then introduces algorithm Scaling-Up Cluster Centers (SUCC) generating cluster centers of large-scale datasets by merging clustering results of small-scale datasets, not mining raw large-scale datasets. We show experimentally that, compared to traditional algorithm DBACAN and leading algorithms DBSCAN++ and HDBSCAN, SUCC can provide not only competitive performance but reduce computational cost. In addition, under the guidance of experts, the performance of SUCC is more competitive in accuracy.

/pdf/a-fast-multiscale-clustering-approach-based-on-dbscan-3st8avx9vu.pdf

A Fast Multiscale Clustering Approach Based on DBSCAN

Human activity recognition is done based on the observation and analysis of human behavior to understand the performed activity. With the emergence of battery powered, low cost and embedded wearable sensors, it became possible to study human activity in various real-world scenarios. Together with the development in data collection, novel machine learning based modeling approaches show huge promise in modeling human activities accurately. Edge computing framework, that is capable of executing human activity recognition models at the edge of the network, is presented in this paper. Framework architecture and its implementation on a single board computer are presented. The framework allows the implementation of various machine learning models for human activity recognition in a standardized manner. The framework is demonstrated experimentally.

Edge Computing Framework for Wearable Sensor-Based Human Activity Recognition

Many companies own a significant number of vehicles. To ensure the undisturbed company workflow, all vehicles have to be tracked. The standard way of vehicle tracking is via a GPS device. Sometimes, GPS devices are sending fallacious data to the server. That data can cause significant errors in daily reports or in the vehicle route preview. This paper describes an efficient technique for finding different types of anomalies in GPS data. The paper describes a connection between finding a QRS complex in ECG signal and anomalies in GPS data. The algorithm is implemented and used as a part of the GPS tracking system that is used by distribution companies in Bosnia and Herzegovina.

Innovative GPS Data Anomaly Detection Algorithm inspired by QRS Complex Detection Algorithms in ECG Signals

In shared human-robot environments, control systems operate based on the information about both human and robot activities to facilitate the successful collaboration between the two. This paper contributes to the emerging field of human-robot collaboration (HRC) by unifying human action recognition (HAR) and high-level robot control technique into single control system. Approach in this paper includes artificial neural network based classifier for recognition of human activity and task-based control as an example of high-level control technique. Classifier is developed based on the data from wearable sensors attached on the human arms. Recognized human activity is used as the input for the selection of functions that describe robot's activity (task). This papers combines both the theoretical approach to the task-based control and it's synergy with HAR while the developed artificial neural network classifier is experimentally validated.

Task-Based Control and Human Activity Recognition for Human-Robot Collaboration

A crucial part to any warehouse workflow is the process of order picking. Orders can significantly vary in the number of items, mass, volume and the total path needed to collect all the items. Some orders can be picked by just one worker, while others are required to be split up and shrunk down, so that they can be assigned to multiple workers. This paper describes the complete process of optimal order splitting. The process consists of evaluating if a given order requires to be split, determining the number of orders it needs to be split into, assigning items for every worker and optimizing the order picking routes. The complete order splitting process can be used both with and without the logistic data (mass and volume), but having logistic data improves the accuracy. Final step of the algorithm is reduction to Vehicle Routing Problem where the total number of vehicles is known beforehand. The process described in this paper is implemented in some of the largest warehouses in Bosnia and Herzegovina.

Innovative Modular Approach based on Vehicle Routing Problem and Ant Colony Optimization for Order Splitting in Real Warehouses

In real datasets often occur cases, where variable or multiple variables have unusual values. These cases are known as anomalies or outliers. For any analysis, it is essential to detect them, because they can bias the analysis. In this paper, a robust anomaly detection method is presented, and it is based on median, rather then on mean value. The method is explained, as well as its parameters and the way how they affect the results. The method is then implemented, and used on Internal Banking Payment Systems. Analysis is given and results are presented.

Multi-level Generalized Clustering Approach and Algorithm for Anomaly Detection in Internal Banking Payment Systems

This paper introduces RVAF, a runtime verification (RV) extension of the Arrowhead Framework (AF) with container-based service-deployment and runtime-enforcement of a desired quality of service (QoS). AF is a service-oriented middleware architecture for IoT-applications, consisting of a set of core and auxiliary services and systems, respectively. The QoS manager (QoSM) is one AF’s most important auxiliary systems, which can be used to guarantee the application’s QoS for a wide set of parameters. In RVAF the QoS offered to a particular IoT-application is specified in signal temporal logic, and is continuously monitored by the RVAF-QoSM. In case of an imminent violation, RVAF automatically initiates a container-based reconfiguration, which is ensured to maintain the desired QoS. RVAF is beneficial to large IoT-applications, where the use of continuous-integration and continuous-deployment tools, is not only a recommended practice but also a necessity. Moreover, the use of RVAF is advantageous both during the development of an IoT application, and after its deployment. We describe the architecture of RVAF, provide its formal underpinning, and demonstrate the usefulness of RVAF supported by an industrial IoT application. The main contribution of this work is to show what it takes to incorporate RV concepts into modern SOA frameworks supporting the development of IoT applications.

Zlatan Tucakovic

Papers

Innovative GPS Data Anomaly Detection Algorithm inspired by QRS Complex Detection Algorithms in ECG Signals

Task-Based Control and Human Activity Recognition for Human-Robot Collaboration

Innovative Modular Approach based on Vehicle Routing Problem and Ant Colony Optimization for Order Splitting in Real Warehouses

Multi-level Generalized Clustering Approach and Algorithm for Anomaly Detection in Internal Banking Payment Systems

QoS for Dynamic Deployment of IoT Services