AUT LibraryAUT
View Item 
  •   Open Theses & Dissertations
  • Masters Theses
  • View Item
  •   Open Theses & Dissertations
  • Masters Theses
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Anomalies Detection and Tracking Using Siamese Neural Networks

An, Na
Thumbnail
View/Open
Thesis (4.547Mb)
Permanent link
http://hdl.handle.net/10292/13525
Metadata
Show full metadata
Abstract
In this thesis, we detect and track anomalies on the sidewalk using deep learning. The proposed network consists of two parts: The first part is an object detection network, namely, SSD(Single Shot MultiBox Detector) is employed to detect and classify objects, then we get the abnormal targets. The second one is to find data association of objects. The proposed model is based on the single-target tracking network SiamRPN, which assists multi-target tracking through a cyclic structure. We follow Hungarian algorithm for getting the final matching results. Both the networks are trained offline, their performance is well.

The contributions of this thesis are: (1) We implement the proposed model for object recognition, classification, and tracking for multiple types of anomalies. (2) We achieve multi-target tracking by combining our object detection algorithm and single-target tracking algorithm. (3) The proposed model is a novel type of deep neural networks to achieve anomaly detection, which has not been found in previous work.
Keywords
Anomalies tracking; Siamese Neural Network; SiamRPN; SSD
Date
2020
Item Type
Thesis
Supervisor(s)
Yan, Wei Qi
Degree Name
Master of Computer and Information Sciences
Publisher
Auckland University of Technology

Contact Us
  • Admin

Hosted by Tuwhera, an initiative of the Auckland University of Technology Library

 

 

Browse

Open Theses & DissertationsTitlesAuthorsDateThesis SupervisorMasters ThesesTitlesAuthorsDateThesis Supervisor

Alternative metrics

 

Statistics

For this itemFor all Open Theses & Dissertations

Share

 
Follow @AUT_SC

Contact Us
  • Admin

Hosted by Tuwhera, an initiative of the Auckland University of Technology Library