| Description |
Good but fast methods for tracking (following) objects in video sequences are on high demand in many applications such as augmented reality (AR), virtual reality (VR), e-toys, or even commerce. Robust object tracking methods are computational expensive and cannot be run as is in online systems (networks) or on mobile devices. The first major problem with mobile devices (online systems) is that visual data provided are often noisy, of low resolution, have strong compression artifacts, degraded due to lighting and other adverse effects. The second major problem is limited computing capabilities, memory, and power. On the other side, diverse input sensors such as cameras, GPS, IMU, or orientation sensors, enable the fusion of data, which in turn support video processing algorithms under these challenging visual conditions. One solution to the first problem is parameter optimization of the tracking method so to achieve good QoS/QoE (Quality of Service/Quality of Experience) at a low cost. A second solution is cloud computing where tracking is to perform on the cloud and communicate results to the cell phone. Another solution is to use other input sensors than the camera to support object tracking. The objective of this project is to implement a video tracking method on a simplified online network or/and a mobile device so it runs on real-time using either parameter optimization (tracker configuration), cloud/server computing, and/or depth and orientation data. What objects to track for what application is an option to be selected based on the student interest and creativity, for example, they may select video tracking of children in schools, or of tourist attractions, etc. |