Object Detection for Object Tracking

01.09.2008 - 30.06.2010
Assigned research project
A variety of methods for object detection and object tracking have become popular over the last years. Object detection methods usually work with single images, while object tracking algorithms emphasize the temporal coherence of an image sequence. This projects mainly covers the topics of a master thesis, motivated by an industrial partner. The aim of this master thesis is to compare several state-of-the-art object identification methods in the face of their ability to track objects from video sequences. This approach is chosen because there are situations where classical object tracking algorithms, which rely on temporal coherence, are infeasible. One example can be video material with a low frame rate. Objects move too far from one frame to the next frame, therefore it is necessary to use methods which are able to identify objects for assigning. Other examples can be crowded places with lots of overlapping objects or non-stationary cameras. Methods which can handle video sequences with reduced frame rate are expected to boost tracking and identification performance for video sequences with full frame rate (25fps) as well. One of the main problems is the performance of object detection algorithms, so analyzing the algorithms capabilities to dealing with video sequences in real-time is an import part of the thesis. The evaluation will be done using ¿real-world¿ video data. Part of the evaluation process is the compilation of a corresponding video database.

People

Project leader

Institute

Contract/collaboration

  • Center Communication System GmbH

Research focus

  • Media Informatics and Visual Computing: 100%

Keywords

GermanEnglish
BildfolgenImage sequences
BildverarbeitungImage Processing
ObjekterkennungObject Recognition

Publications