Temporal Pattern-Based Active Marker Identification and Tracking

Marker tracking is widely used in applications like motion capture and pose estimation. This thesis describes the development of an active marker tracking algorithm using a Dynamic Vision Sensor (DVS) as recording device. The DVS’s event driven behaviour opens novel opportunities to develop a tracking algorithm with a high temporal resolution at low costs. The tracking algorithm combines both the spatial and temporal information that is induced by the DVS in response to changes in illumination. To ensure a continuous change in the illumination, blinking light emitting diodes (LED) are used as active markers. This changing illumination causes a continuous stream of events induced by the blinking of the LED. The thesis formulates a mathematical model to track these LEDs using the DVS’s key features: Its high temporal resolution and temporal accuracy. Especially accounting for the temporal information the developed method can handle external influences like camera movements, vibrations, background activity and noise. The thesis also explores the relation between the LED’s blinking pattern and the LED’s velocity to. It shows that a high LED velocity requires a high LED blinking frequency to accurately measure the LED’s trajectory. The thesis describes distinct features to uniquely identify a blinking LED and compares their advantages and disadvantages. Using the LED’s features the tracking algorithm is able to handle temporary object disappearance, occlusion and trajectory crossing. Aiming at demonstrating the basic capability the developed algorithm is able to identify up to 5 LEDs with blinking patterns between 100Hz and 1kHz.