Computer Surveillance and Scene Activity Tracking

On Moving Shadow Elimination and Gross-Activity Tracking for Scene Analysis

A shadow is formed when light from a source is intercepted by an opaque body in such a way that the other side of the body not facing the source is in darkness. Projection of this dark region on a surface behind the object is known as a shadow region.

Shadows, in general, can be categorized as static shadows or moving shadows depending upon whether the causal object is relatively static or moving [1]. Elimination of static shadows that usually form a part of the background has never been judged as a crucial pre-processing step as such shadows usually do not jeopardize the actual foreground object recognition process of surveillance systems [2]. On the other hand, shadows cast by dynamic objects or by objects suddenly brought into a background scene are often misclassified as the actual foreground objects leading to poor object segmentation and tracking [1]. Hence, a foreground shadow region elimination process has become an unavoidable pre-processing step for the development and implementation of a robust and reliable real-time video surveillance system.

A computational model that works in the RGB colour space has been developed to mark or eliminate moving shadow pixels in the video sequence [2]; the model exploits the fact that a shadow can be considered as a semi-transparent region in the image, which retains a representation of the underlying surface pattern, texture or colour value [3]. It estimates, in particular, the brightness and chromaticity distortion factor values separately for each pixel of the current frame with respect to the corresponding pixel of the expected background frame to determine whether it is a moving shadow pixel or not.

Mixture models have been considered to remove the ad hoc nature of the chosen thresholds for the above mentioned parameters. Moreover, the Kolmogorov-Smirnov test has been employed to verify whether the conceived model adequately describes the distribution or not [2].

A channel ratio test has been developed to increase the success rate in marking moving shadows cast on non-coloured wooden surfaces [2]. Evaluation metrics have been conceived to understand the significance of incorporating the channel ratio test in the overall method [2]. Moving region segmentation techniques based on statistical aberrant data detection processes have also been deployed so as to use relaxed threshold values; use of relaxed threshold values have helped as to mark not only the strong portion of the shadow, but also a considerable portion of the soft region [4].

Currently we have undertaken a work on detection of various body parts, and pose vectors to determine regularity in gross-activities.

[1] Nadimi S, Bhanu B. Physical Models for Moving Shadow and Object Detection in Video. IEEE Transactions on Pattern Analysis and Machine Intelligence 2004; 26 (8): 1079-87.

[2] Mitra BK, Kypraios I, Young R, Chatwin C. Development of a Moving Shadow Detection Method for Indoor Scene Activity Tracking, Under Review.

[3] Horprasert T, Harwood D, Davis LS. A Statistical approach for Real-Time Robust Background Subtraction and Shadow Detection. Proceedings IEEE International Conference on Computer Vision ('99 FRAME-RATE Workshop) 1999.

[4] Mitra BK, Young R, Chatwin C. On shadow elimination after moving region segmentation based on different threshold selection strategies, accepted for publication, Optics and Lasers in Engineering.