Person Detection Based on Convolutional Models

Problem detekcije objekata je jedan od najraširenijih problema u području računalnog vida, a u ovom radu se koncentriramo specifično na detekciju ljudi. Započinjemo teorijskim pregledom algoritama za detekciju gdje prvo opisujemo algoritam koji se ne temelji na konvolucijskim mrežama. Zatim detaljnije opisujemo nekoliko danas popularnih konvolucijskih modela i uspoređujemo ih na problemu detekcije ljudi. Modele uspoređujemo na nekoliko skupova podataka, a osim preciznosti, komentiramo i najčešće pogreške svakog modela uz primjere. Na kraju donosimo zaključke na temelju dobivenih rezultata.Object detection is one of the most widely researched areas in computer vision. In this thesis we focus on person detection. We start by giving a theoretical overview of the object detection algorithms. First, we introduce an algorithm that is not based on convolutional neural networks. We continue by describing today's most popular convolutional models and we compare them on person detection task. The models are evaluated on several datasets and the results are compared. Finally we identify the most common errors for each model and provide a conclusion based on the results