3D Érzékelés és Mobilrobotika Kutatócsoport

2D/3D Képfeldolgozás

Computer Vision is one of the primary research areas of the 3DMR lab. Due to the large increase of computing power over the last decades, Computer Vision methods have seen a huge boost to performance and efficiency. The applications of these methods seems endless, ranging from medical applications, to surveillance, virtual reality and robotics. Computer Vision is perhaps the most complex perception problem, thus it is a fundamental problem a artificial intelligence. There is a wide range of research topics in our lab concerning vision problems. We research vision applications in virtual or augmented reality, such as precise object tracking, and recognition, as well as 3D reconstruction. Our most recent focus includes learning vision using deep learning technology, with applications, such as melanoma recognition and game-playing agents. We also develop vision algorithm for mobile robots, such as traffic sign recognition and collision avoidance.

Deep Learning for RoboCup - RoboCup is the annual world championship in robot soccer organized by the top researchers in the field of autonomous robots. Our research group collaborates with the Machine Intelligence and Pattern Analysis Lab (MIPAL) of the Australian Griffith University in order to enable the humanoid Nao robots to play soccer.

Melanoma Detection using Deep Learning - Melanoma is a form a malignant skin cancer that kills approximately 50,000 people annually. One of the major causes of death is late recognition, since melanomas can be safely removed in early stages. Melanomas are usually located on the skin and are often mistaken for moles by patients, still they have several recognisable features that can be used for automatic detection. Our goal is to develop a deep neural network to recognise potential melanomas visually, finding skin lesions that require the inspection of a dermatologist.

2D and 3D Object Tracking - Object tracking is one of the fundamental tasks in computer vision, with applications ranging from surveillance to augmented reality and mobile robotics. Our lab focuses on simultaneous model creation and tracking in both 2D and 3D settings. For the 2D case, we research the use of deep convolutional features and their use for accurate multi-object tracking and pose estimation.

Intelligent Agents for Playing computer Games - The idea of developing autonomous agents using reinforcement learning has seen many recent successes, including AlphaGo’s memorable victory. While using AI to play games may seem like a rather childish endeavour, games actually produce a very safe environment to test AI’s ability to develop an understanding of the physical world. Since most video games are made about certain aspects of the real world, developing AI that is able to play these games well may pave the way for autonomous robots that can interact with the physical world safely and productively.