Siamska nevronska mreža za detekcijo gibanja v video sekvencah

Mlakar, Peter (author), Kristan, Matej (mentor)

V diplomski nalogi obravnavamo problem avtomatske detekcije gibanja v video sekvencah posnetih z video nadzornimi sistemi. Trenutno najuspešnejše metode uporabljajo konvolucijske nevronske mreže za reševanje tega problema. Bistvena omejitev teh pristopov je v tem, da potrebujejo ponovno učenje za različne video sekvence, kar zmanjša njihovo aplikativno vrednost. V diplomskem delu predstavimo novo metodo, ki temelji na arhitekturi siamskih konvolucijskih mrež. Mreža s pomočjo siamske arhitekture semantično opiše vhodno sliko sekvence ter model ozadja sekvence. Nadaljnji konvolucijski nivoji detektirajo relevantne razlike ter generirajo verjetnostno masko segmentacije gibanja. Z metodo lahko detekcijo gibanja izvajamo na različnih video sekvencah brez ponovnega učenja. Za izvajanje potrebujemo le referenčno sliko ozadja sekvence, ki jo nato tekom časa samodejno posodablja. Mrežo smo učili na podatkovni zbirki CDNET. Pridobljene rezultate smo primerjali s preostalimi metodami, objavljenimi na spletni strani CDNET. Naša metoda se je po uspešnosti uvrstila na osmo mesto izmed 46 objavljenih algoritmov. Mrežo smo ocenili tudi na evalvacijskih zbirkah Wallflower ter SGM-RGBD, kjer smo jo preizkusili v različnih okoliščinah ter podali kvalitativno analizo njenega delovanje.

• Language: Slovenian
• Keywords: računalniški vid, siamske konvolucijske nevronske mreže, detekcija gibanja, video nadzorni sistemi
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2018
• PID: 20.500.12556/RUL-102760
• Publication date in RUL: 07.09.2018

MLAKAR, Peter, 2018, Siamska nevronska mreža za detekcijo gibanja v video sekvencah [online]. Bachelor’s thesis. [Accessed 22 March 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=102760

Secondary language

• Language: English
• Title: Siamese neural network for motion detection in video sequences

Abstract

We examine the problem of automatic motion detection in video sequences captured by video surveillance systems. The state of the art methods use convolutional neural networks. Their main limitation is that they need to be retrained if they are to be applied on different sequences. In our thesis, we present a novel method which is based on the architecture of siamese convolutional neural networks. Our network semantically describes the input image from the sequence and the model of the background of the sequence. It does this by using the siamese architecture. It then applies convolutional layers to detect relevant differences and generates the final probability segmentation mask. Our approach allows detection on different video sequences without retraining the network on each new sequence. To detect motion only a reference background images is required. The method automatically updates the background image during application. We trained our network on the CDNET data set. We compared our method with the other methods published on the CDNET website. It ranked as the eight best method of the 46 published methods. We also evaluated our method on the Wallflower and SGM-RGBD data sets. There, we tested it in different circumstances and provided qualitative analysis of its performance.

• Keywords: computer vision, siamese convolutional neural networks, motion detection, video surveillance systems