Tracking of Moving Objects in Video Sequences od Tennis Matches

Tema ovog završnog rada je "Praćenje pokretnih objekata u videozapisima teniskih mečeva". Uz praćenje objekata, opisan je pojam detekcije objekata koji je sastavni dio praćenja te su prikazani neki od pristupa detekciji kao što su Viola-Jones algoritam, SSD te R-CNN modeli. Implementacija praćenja objekata zasnovana je na unaprijed istreniranom SSD modelu koji obavlja detekciju objekata te je napravljena usporedba s Faster R-CNN modelom. Za praćenje objekata izabran je jednostavni algoritam za praćenje položaja centroida. Algoritam je izrađen u programskom jeziku Python, uz korištenje biblioteka OpenCV i Tensorflow za jednostavniji rad s videozapisima i neuronskim mrežama. Navedene se mogućnosti poboljšanja algoritma za praćenje, kao i modeli koji daju bolje rezultate kod detekcije objekata.Topic of this work is "Tracking of Moving Objects in Video Sequences of Tennis Matches. Main concepts covered are object tracking and object detection, which is the base for most popular tracking algorithms. Few of most popular object detection models are described, such as Viola-Jones algorithm, SSD model and R-CNN models. Object tracking implementation is based on pre-trained SSD model which is used for object detection. As for object tracking algorithm, simple algorithm called centroid tracking is used. Algorithm is implemented in Python and OpenCV and Tensorflow libraries are used for easier handling with videos and neural networks. At the end, few improvements and better models are proposed in regards to object tracking algorithm and object detection models