Multiple object tracking and object flow visualization

Date of Publication

8-17-2019

Document Type

Master's Thesis

Degree Name

Master of Science in Computer Science

Subject Categories

Computer Sciences

College

College of Computer Studies

Department/Unit

Software Technology

Thesis Adviser

Joel P. Ilao

Defense Panel Chair

Conrado R. Ruiz

Defense Panel Member

Florante R. Salvador
Joel P. Ilao

Abstract/Summary

Multiple object tracker could be used in different applications such as crowd move- ment control, abnormal behaviour detection, designing evacuation routes, activity
detection/monitoring systems, and improving product display strategies that lead
to business improvement. The main task of multiple object tracking is to link de- tections in different frames. This research performs multiple object tracking with
the use of object detector and single object tracker that is assigned to each object. Single object trackers require manual initialization and vulnerable to occlusions. To address such problem, this research aims to automatically initialize and assign
correlation filter-based object tracker to each object with the help of object de- tection. Trackers are updated based on the object detection result. In addition,multiple correlation filters in object tracker, assumption on direction of an object, and assumption on potential position of missing objects are introduced to address the occlusion problem.
Experiment shows that updating the tracker based on the object detection result performs better than the baseline. Baseline model updates correlation filter based on its own output. Neither object tracker is trained with object detection output nor additional measures are considered to solve occlusion problem. Fur- thermore, incorporating assumption on potential position improved the ability of correctly re-locating the missing object.
Further experiment is performed on benchmark dataset to compare the per- formance with other state-of-the-art trackers. The proposed outperforms some of the state-of-the-art models introduced in top conferences.
To maximize the multiple object tracker, obtained trajectories are smoothened and processed using various methods and heatmp for object flow visualization.
Perspective transformation is also performed on region of interest to provide vi- sualization in 2D perspective. The visualization distinguishes an area that is frequently passed by objects.
This research possibly be improved further by maximizing convolution map information (object detection) and appearance features (object tracker) for mul- tiple object tracking. Social force model may be incorporated as well to consider other objects and surroundings when tracking an object.

Abstract Format

html

Language

English

Format

Electronic

Accession Number

CDTG008263

Keywords

Image processing—Digital techniques; Computer vision

Upload Full Text

wf_no

Embargo Period

4-23-2025

This document is currently not available here.

Share

COinS