Background Attraction Problems

Background Subtraction Problems

According to Omar Javed, Khurram Shafique and Mubarak Shah (2001), the main problems of background subtraction approaches can be listed as follows.

Quick illumination changes

Quick illumination changes will affect the color characteristics of the background pixels on color or intensity based subtraction in background model. As a result, whole image appears as foreground. This kind of situation will occur in partially cloudy days.

Relocation of the background object

It induces change in two different sections in the image. It’s continuously acquired its position and previous position. If only the former will be identified as foreground region, any background subtraction system based on color variation is detected

For example, in a vehicle tracking, corners of vehicle can be selected as its features. When partial occlusion occurs, a fraction of these features is still visible. This may help us to overcome the partial occlusion problems.

According to P KaewTrakulPong and R Bowden (2001), they have used four models in their system to track objects. They listed as follows,

• Motion model – Kalman filters are used to track multiple objects. These objects based on measurement of their position, motion, simple shape details and color contents. The coordinates of the object’s centroid and the minimum bounding box of the object are modeled by a discrete time kinematics model. The centroid is modeled by using a white noise acceleration model. The height and width of the bounding box are modeled by using a white noise velocity

They are, Appearance model (above mentioned), Motion model and Probability model. Motion model uses two 2D motion model and one 3D model. Probability model depends on appearance model and Motion models in the tracking. The tracking of detected moving objects in each scene is formulated as a maximization of a joint probability model.

Moving object Features

The features of moving object can be represented as parameters for identifying moving object. These parameters can be taken from the images sequence. Normally we use matching algorithm to map the selected features to the current frame.

According to Aishy Amer (2003), we can use following features to track object, Size, Shape, Motion, Center of gravity and Distance.

• Size – basically this is area, width and height of the object

• Shape – minimum bounding box

• Motion – velocities of vertical and horizontal direction

• Center of gravity – it is the center of minimum bounding box

• Distance – the Euclidean distance between current object and the object in former