Illuminant-Invariant Bayesian Detection of Moving Video Objects

Y.-K. Wang and C.-H. Su (Taiwan)


Illuminant invariant, online parameter estimation, recursive update, Gaussian mixture models, moving object detection.*


Moving objects detection is a crucial step for video surveillance, MPEG-4 and many image and video processings. This paper proposes a Bayesian-based background subtraction approach for the detection of moving object. Our Bayesian approach needs two probability density distribution functions, which are considered as a random process and modeled by Gaussian Mixture Models. Efficient online recursive equations are devised to better estimate the parameters of the two probability density functions. The recursive update equations are adaptive to light change of environments. The proposed approach is compared to several methods in experiments of both simulated and real-world videos. Experiments show stable results, which verify the feasibility of our approach.

Important Links:

Go Back