Human Pose Estimation Based on Deep Learning

Jedan od vrlo važnih zadataka u području računalnog vida je određivanje poze osobe iz slika. U okviru ovog rada predstavljena su dosadašnja istraživanja na tu temu. Odabran je prikladan model te se pristupilo implementaciji. Budući da modelu trebaju podatci na kojima će se učiti, provjeravati i ispitivati, potrebno je pretprocesirati ulazne podatke. Nakon procesa učenja slijedi optimizacija hiperparametara modela te, konačno, evaluacija.One of the very important tasks in computer vision field is human pose estimation from images. Previous research on the subject has been presented within this thesis. After the research, a suitable model was chosen and implemented. It was necessary to preprocess the input data since the model needs data for training, validation and testing. The learning process is followed by the optimization of model's hyperparameters. Finally, the implemented model was evaluated on the testing dataset