Ali Koksal Hocaoglu

Removed object dedection using one class classification

In this study, using surveillance videos, an operator interactive removed object detection algorithm is developed. Using dual foregrounds, static foreground regions are detected. Then, these detected regions are classified with a one-class classifier to decide whether a removed/stolen object is present in that region. For training the classifier, the region of interest is selected by the operator. Proposed algorithm further reduces the false alarms and detection delay with respect to the existing methods.

Improving track initiation performance by track validation algorithms for multi-target tracking in heavy clutter

In this study, for multi-target tracking in an heavy-clutter environment, line detection-based validation algorithms are proposed for improving track initiation performance by reducing false tracks. Track validation algorithm is applied to the results of the base track initiation algorithm periodically and deviation from line evaluated based on a defined measure. Then, by thresholding the deviation value, true and false tracks are discriminated, even for maneuvering targets. Thus, performance of both track initiation and tracker are improved by means of number of false tracks.

Target motion analysis with fitness adaptive big bang-big crunch optimization algorithm

In this paper, an evolutionary search approach is proposed for the bearing only target motion analysis problem. A problem spesific modification for the Big Bang-Big Crunch optimization algorithm, namely, Fitness Adaptive Big Bang-Big Crunch optimization algorithm is used as the evolutionary search method. The effectiveness of the proposed method is demonstrated by testing the algorithm in different tracking scenarios.