Detection and tracking of sphere markers

Abstract

Tracking and detection of sphere markers captured by a GoPro Hero 4 Silver camera at 240 frames per second is studied and discussed in this report. Different tools are designed such as, optimized color detection, optimized edge circle detection, and point tracking. A number of tools have been used such as Kernelized Correlation Filters, Optical Flow and Circle Hough Transform. The major problem of KCF trackers is their incapability of adjusting the scale changes of the target. It is solved by designing a point tracker unit. The marker used for this study is spherical and gray. Color detector algorithms are normally reliable for specific colors. In the worst case they cannot detect gray objects or it comes with high error. A learning algorithm was designed to optimize the color range in HSV as a signature of the marker. The Circle Hough Transform is a well-known method for detecting circles. This function accepts many inputs which effect the position of the circle. A learning method is developed to optimize its inputs by minimizing a defined error function. Finally, the results have been smoothen by applying Kalman Filter, an extremely accurate filter used in control industry and robotics to smooth or predict the position of robots. In this work we discuss the robustness of the developed algorithm against changes in brightness that shows how it would perform in more real-world conditions. The final results have been compared in accuracy of the algorithm versus three other well-known algorithms: CSRT, Boosting, and Median Flow. This comparison shows that the proposed method is reliable and the developed algorithm is in fact more accurate in detection and more reliable with fewer failures.

Date	17 Dec 2019
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Carlos Vázquez (Supervisor) & Felix Chenier (Co-supervisor)