25-08-2017, 09:32 PM
A Change Information Based Fast Algorithm for Video Object Detection and Tracking
5.pdf (Size: 523.58 KB / Downloads: 42)
Introduction
DETECTION and tracking of moving objects from a
video scene is a challenging task in video processing
and computer vision [1]–[4]. It has wide applications such
as video surveillance, event detection, activity recognition,
activity based human recognition, fault diagnosis, anomaly
detection, robotics, autonomous navigation, dynamic scene
analysis, path detection, and others [1]–[4]. Moving object
detection in a video is the process of identifying different
object regions which are moving with respect to the background.
More specifically, moving object detection in a video
is the process of identifying those objects in the video whose
movements will create a dynamic variation in the scene [2].
This can be achieved by two different ways: 1) motion detection/
change detection, and 2) motion estimation [2]. Change
or motion detection is the process of identifying changed and
unchanged regions from the extracted video image frames
when the camera is fixed and the objects are moving.
Proposed Algorithm for Object Detection
A block diagrammatic representation of the proposed
scheme is given in Fig. 1. Here we use two types of segmentation
schemes: one is a spatio-temporal spatial segmentation
and the other is a temporal segmentation. Spatial segmentation
helps in determining the boundary of the regions in the scene
accurately, and temporal segmentation helps in determining
the foreground and the background parts of it.
Spatio-Temporal Spatial Segmentation
In the spatio-temporal spatial segmentation scheme, we
have modeled each video image frame with compound MRF
model and the segmentation problem is solved using the
MAP estimation principle. For initial frame segmentation, a
hybrid algorithm is proposed to obtain the MAP estimate. For
segmentation of other frames, changes between the frames is
imposed on the previously available segmented frame so as
to have an initialization to find the segmentation result of
other frames. The total scheme is described in detail in the
subsequent sections.
Conclusion and Discussion
In this article, a change information based moving object
detection scheme is proposed. The spatio-temporal spatial
segmentation result of the initial frame is obtained by edgebased
MRF modeling and a hybrid MAP estimation algorithm
(hybrid of SA and ICM). The segmentation result of the initial
frame together with some change information from other
frames is used to generate an initialization for segmentation
of other frames. Then, an ICM algorithm is used on that
frame starting from the obtained initialization for segmentation.
It is found that the proposed approach produces better
segmentation results compared to those of edgeless and JSEG
segmentation schemes and comparable results with edgebased
approach.