Short biography

Suhani Singh is an Early-Stage Researcher at the Department of Automatic Control, Robotics, and Computer Vision (ARV), Universitat Politècnica de Catalunya, Barcelona. As part of the ITN 5GSmartFact, she will carry out her research on 3D Object Recognition and Localization of a Mobile Robot Platform Using 5G at Roboception GmbH, Munich. Before starting her PhD., she worked as a Senior Research Fellow at the National Institute of Technology, Rourkela, where she developed a ROS-based reconfigurable flight control module for multi-rotor drones. In 2019, she completed her master's degree in Robotics at the Defense Institute of Advanced Technology, India, focusing on path planning and force control of industrial and collaborative robots. She also earned a bachelor's degree in Electronics and Communication Engineering from Uttarakhand Technical University, Dehradun, India.

Short description of project objectives

The objectives of my work are:

1. Comprehensive research and testing of the capabilities of a distributed sensor network and a mobile platform leveraging edge cloud processing over a 5G network for perception and localization applications.

2. Employing machine learning and/or standard image processing methodologies to build solutions for 3D object recognition of known and unknown objects.

3. By making use of edge cloud server processing, assess a mobile platform's SLAM and path planning capabilities.

4. Investigate a remote supervision system using VR equipment for mobile platform surveillance, mapping, and teleoperation.

State of the research
Computer vision faces a processing challenge in order to produce results as quickly as possible. Because processing necessitates a substantial percentage of hardware resources that are not readily available in traditional infrastructures, algorithmic processes such as machine learning and deep learning processing typically take place in the cloud, where all data is analyzed, processed, and the results are sent to the final user. Cloud computing devices send visual data to the cloud for analysis, and the cloud responds with appropriate responses for further action. This can cause system response time delays.

Edge computing, on the other hand, is a type of distributed computing in which all computations occur outside of the cloud or at the "edge" of a centralized server. Edge computing devices help with this function by capturing visual data and performing computations locally on the device, closer to the source of the data, allowing that information to be used immediately. Edge has several advantages over cloud, including reach, speed, and privacy, but the most significant advantage is lower operational costs. An edge-based platform simplifies and reduces the cost of CV deployment at the edge.

Creating and deploying a high-quality computer vision application on the edge presents its own set of difficulties. When computation demand increases, we can scale instances in the cloud; however, our options with edge devices are limited to the power of the current device in use. Furthermore, without integrating additional packages, edge deployment may not provide direct access to device status.

APIs and containerization can assist in overcoming the aforementioned deployment challenges. A container-based deployment service can deploy and monitor multiple edge devices. While this approach does rely on some cloud connectivity, the actual data is not transmitted across the network, so the edge's speed and low cost are preserved.