Moving object tracking using hybrid method

Abstract

Moving object tracking is a method used to estimate the trajectory, detect and analyze changes that occur in an observed object in a video. The moving observed object can be a single object or plural objects. This research focused on a single object within dataset video consists of several moving objects, object difference to the background, nonlinear object movement, camera movement, and object zoom in/out. Sometimes the object being tracked is obvious, but the tracking result is less precise. Some of the reasons are low quality video, system noise, small object, and other factors. In order to improve the precisions of the tracked object especially with dataset video above, we propose a new hybrid method for better and faster tracking result. The object tracking process used a hybrid method of Camshift method as the main tracking technique and Kalman filter for prediction and correction. The Camshift method has several advantages including tracking in various histogram condition and varying object color. Kalman Filter method has advantages to predict object movement in next frame based on previous frame. The computational complexity and large memory requirements for the implementation of tracking were reduced and the precision of the tracked target was good. Based on the trial of tracking the whole video, we can conclude that by adding prediction process with Kalman Filter makes object tracking results become more precise. The addition of Kalman Filter also makes the average tracking time faster than the tracking time using the Camshift method only in the whole video.