Bayesian modeling of dynamic scenes for object detection
read more
Citations
A survey of advances in vision-based human motion capture and analysis
ViBe: A Universal Background Subtraction Algorithm for Video Sequences
Traditional and recent approaches in background modeling for foreground detection: An overview
SuBSENSE: A Universal Change Detection Method With Local Adaptive Sensitivity
Segmenting salient objects from images and videos
References
Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images
Mean shift: a robust approach toward feature space analysis
Introduction to Statistical Pattern Recognition
On Estimation of a Probability Density Function and Mode
Fast approximate energy minimization via graph cuts
Related Papers (5)
Frequently Asked Questions (14)
Q2. What is the purpose of the joint feature space?
The joint feature space provides the ability to incorporate the spatial distribution of intensities into the decision process, and such feature spaces have been previously used for3031image segmentation, smoothing [4] and tracking [6].
Q3. How does the proposed approach handle the dynamic texture of the pool?
Since dynamic textures like the water do not repeat exactly, pixel-wise methods, like the mixture of Gaussians approach, handle the dynamic texture of the pool poorly, regularly27producing false positives.
Q4. What is the main argument that the model of image pixels is challenged?
The model of image pixels as independent random variables, an assumption almost ubiquitous in background subtraction methods, is challenged and it is further asserted that there exists useful structure in the spatial proximity of pixels.
Q5. What is the main source of spatial uncertainty of a pixel?
If the primary source of spatial uncertainty of a pixel is image misalignment, a Gaussian density would be an adequate model since the corresponding point in the subsequent frame is equally likely to lie in any direction.
Q6. What is the definition of a kernel density estimator?
The kernel density estimator is a nonparametric estimator and under appropriate conditions the estimate it produces is a valid probability itself.
Q7. What is the effect of the presence of multiple models on the pixel intensity?
Since most of these phenomenon are ‘periodic’, the presence of multiple models describing each pixel mitigates this effect somewhat by allowing a mode for each periodically observed pixel intensity, however performance notably deteriorates since dynamic textures usually do not repeat exactly (see experiments in Section III).
Q8. What is the need for a mixture model to describe spatial uncertainty?
Analogous to the need for a mixture model to describe intensity distributions, unimodal distributions are limited in their ability to model7 spatial uncertainty.
Q9. What was the first approach to pixel-wise modeling?
Early approaches operated on the premise that the color of a pixel over time in a static scene could be modeled by a single Gaussian distribution, N(µ, Σ).
Q10. What is the threshold for the detection using only the background model?
The threshold for the detection using only the background model was chosen as log(γ) (see Equation 7), which was equal to -27.9905.
Q11. What is the probability of observing a foreground pixel in the same proximity?
If an object is detected in the preceding frame, the probability of observing the colors of that object in the same proximity increases according to the second term in Equation 7.
Q12. What are the two categories of background modeling methods?
In the context of this work, these background modeling methods can be classified into two categories: (1) Methods that employ local (pixel-wise) models of intensity and (2) Methods5 that have regional models of intensity.
Q13. What was the first method to model the uncertainty of each pixel color?
While these methods were among the first to principally model the uncertainty of each pixel color, it was quickly found that the single Gaussian pdf was illsuited to most outdoor situations, since repetitive object motion, shadows or reflectance often caused multiple pixel colors to belong to the background at each pixel.
Q14. What is the probability of each pixel-vector belonging to the background?
Given this sample set, at the observation of the frame at time t, the probability of each pixel-vector belonging to the background can be computed using the kernel density estimator ([27], [31]).