Tracking Objects with Full Occlusion in Video Sequence Using Modified Kalman Filter with S-Aces as a Measuring Tool

Document Type


Degree Name

Master of Science (MS)


Computer Science and Info Sys

Date of Award

Spring 2015


Over recent years, much research has been devoted to object tracking under occlusions. This holds because in the real-world tracking a target being partly or fully covered by obstacle, for an uncertain period of time, is common. It is a great challenge to detect an object that is overlapped or occluded by other objects. Further, we explain the methods to solve the occlusion problem in single object tracking. This paper proposes, moving object detection and tracking approach under partial and full occlusions in a video sequence, applying a Modified Kalman Filter (MKF) and Shrinking Active Contour (S-ACES). We used S-ACES as a closed curve that can move through a frame and capture an object boundary. By using the object mass center and the Modified Kalman Filter (MKF) we are able to determine the location of the object in the present frame. MKF is used for tracking and it consists of three stages while tracking the object: 1) Prediction of Object future location. 2) Reduction of noise introduced by inaccurate detections. 3) Keep tracking and estimating the position of the moving object during occlusion. In this study, we introduced velocity and acceleration components to the existing software. Adding them to the prediction stage improved the object tracking with occlusion results. To determine the target after a full disappearance, the active contour envelops the obstacle and once the object is covered, the radius points of the current frame and the previous frame are compared. If the software identifies any significant increase in radius, it means that the target is fully occluded by the obstacle. The software continues to compare the radius of the frames. Another significant change in the radius is an indication that the target is moving away from the obstacle get the maximum boundary distance point in x direction of this frame and subtracts with the target radius to get target center. And also software makes Measurement equal to Prediction equal to Estimation point to keep tracking continue. In another significant change in radius indicates that the target is splitting from the obstacle by using splitting algorithm. Thereafter, the matching method determines the target after its appearance. A number of experiments are performed using video sequences for full occlusions, with stable background.


Nikolay Metodiev Sirakov

Subject Categories

Computer Sciences | Physical Sciences and Mathematics