February 20, 2010
School of Engineering, Amritapuri
A research paper written by Ms. Vinitha Panicker, Assistant Professor in the Department of Computer Science and Engineering at Amritapuri, was selected for the 2nd International Conference on Computer and Automation Engineering (ICCAE 2010).
The title of the paper that will be presented later this month in the conference at Singapore is Detection of Moving Cast Shadows using Edge Information.
Shadows of objects in 2-D photographs and pictures provide some 3-D information to viewers. But what are cast shadows?
A shadow of any object consists of two parts — self shadow and cast shadow. Self shadow is cast by that part of the object that is not illuminated by direct light. The cast shadow is the area projected by the object in the direction of direct light.
Why are cast shadows important?
Moving cast shadows pose a challenge in traffic monitoring and other visual surveillance applications. These shadows distort the true shape and color of target objects. The cast shadows are to be eliminated for correct object detection.
In recent years, there has been extensive research in the area of cast shadow detection as well as removal. Most existing techniques are based on brightness or color properties of the shadow. The problem with such approaches is that if the foreground has objects with similar values of brightness or intensity, misclassification occurs.
This paper introduces a new technique for detecting and thereby removing cast shadows, using edge information of shadows, rather than color information. The proposed method first removes the boundary of the cast shadow, while preserving the interior edges of the object. The coarse object shapes are then reconstructed using the object interior edges.
The cast shadow is finally detected by subtracting the reconstructed moving object from the original object. The detected shadow regions are replaced with corresponding background pixels for obtaining the removed frame of the shadow. This proposed method works even with thin shadows and shadows that are far away from the camera position.
The paper also compares this new method to five other popular shadow detection techniques, four based on color properties of shadows and one based on edge properties, like this one. The comparison results are favorable to this new technique. See Complete Abstract »
The paper will be included in the ICCAE 2010 Conference Proceedings that will be published by the IEEE. The proceedings will be included in IEEE Xplore, and indexed by Ei Compendex and Thomson ISI Proceeding.
Extraction of moving objects from a video sequence is a fundamental and crucial problem of many vision systems in a target detection / tracking and video-surveillance environment. In many vision systems the focus of image processing is on the robust shape detection of moving objects present in the scene. The accuracy and efficiency of detection is very crucial for these tasks.
However, the accuracy is marred by illumination changes such as shadows. As the shadows attached along with the moving object also have the same motion as that of objects, the detection of shadows as foreground objects is very common. This produces big errors in object localization and recognition. The effect of shadows has to be eliminated from those scenes to ensure the reliability of such systems; therefore separate methods have to be developed for shadow handling.
This paper introduces an effective method which uses edge information to detect moving cast shadows for traffic sequences. The proposed method initially removes the boundary of the cast shadow, preserving the object’s interior edges. The coarse object shapes are then reconstructed using the object interior edges. Finally, the cast shadow is detected by subtracting the reconstructed moving object from the change detection mask.
The method was implemented and tested using three benchmark videos. The efficiency of the proposed method was compared with five other popular shadow detection methods and the results proved its superiority over others.
