Variational Inference for Background Subtraction in Infrared Imagery
This work addresses background subtraction for infrared imagery, which is incremental as it applies an existing Bayesian approach to a specific domain with improvements in efficiency and accuracy.
The authors tackled background subtraction in infrared imagery by proposing a Gaussian mixture model with Bayesian variational inference, which automatically estimates parameters and avoids over/under fitting, resulting in outperforming other methods in precision and recall while maintaining real-time computational efficiency.
We propose a Gaussian mixture model for background subtraction in infrared imagery. Following a Bayesian approach, our method automatically estimates the number of Gaussian components as well as their parameters, while simultaneously it avoids over/under fitting. The equations for estimating model parameters are analytically derived and thus our method does not require any sampling algorithm that is computationally and memory inefficient. The pixel density estimate is followed by an efficient and highly accurate updating mechanism, which permits our system to be automatically adapted to dynamically changing operation conditions. Experimental results and comparisons with other methods show that our method outperforms, in terms of precision and recall, while at the same time it keeps computational cost suitable for real-time applications.