### Citations

3205 | Rapid object detection using a boosted cascade of simple features.
- Viola, Jones
- 2001
(Show Context)
Citation Context ... any vector projected (spanned) in the eigenvector space span(A i ) can be seamlessly expressed by the linear combination of these eigenvectors A i . So, no matter how is p chosen, the second term in =-=(5)-=- could always be minimized to 0. Therefore, i Ai(x) [ I(W(x;p) A0(x)] λ = ∑ ⋅ − (6) x∈s0 For the first term in (5), we will resort to inverse compositional image alignment method in the next subsectio... |

1825 | Robust Real-Time Face Detection,
- Viola, Jones
- 2004
(Show Context)
Citation Context ... is just reversing the roles of the trained model template image A(x) 0 and the newly input image I(W(W(x; Δ p);p)) in compositional method, which results in 2 0 ⊥ span(A i ) A (W(x; Δp)) − I(W(x;p)) =-=(7)-=- (5)sIn order to get the extremum, after taking the first order Taylor series expansion in terms of Δp at Δ p= 0, the solution to could be deducted as Δp ⎡ ∂W ⎤ Δ ∇ T ∑ ⎢ 0 ⎥ [ 0 ] −1 p= H A I(W(x;p))... |

826 | Detecting Faces in Images: A Survey,
- Yang, Kriegman, et al.
- 2002
(Show Context)
Citation Context ...ld be deducted as Δp ⎡ ∂W ⎤ Δ ∇ T ∑ ⎢ 0 ⎥ [ 0 ] −1 p= H A I(W(x;p))- (x) x ⎢⎣ ∂p p= 0⎥⎦ ⊥ span(A i ) T 0 0 x ⎢⎣ ∂p p p= 0⎥⎦ ⊥ ⎢ span(A ) ⎣ ∂ p= 0⎥⎦ ⊥ i span(A i) A (8) ⎡ ∂W ⎤ ⎡ ∂W ⎤ H = ∑ ⎢∇A ⎥ ⎢∇A ⎥ =-=(9)-=- ∂W where ∇A0 is a 1*2 gradient vector for every pixel in the trained model template image; is a 2* N Jacobian ∂p p= 0 matrix to express the relationship between the 2*1 warped point W(x;p) and the N ... |

614 | Active shape models-their training and application - Cootes, Taylor, et al. - 1995 |

565 | An extended set of haar-like features for rapid object detection
- Lienhart, Maydt
- 2002
(Show Context)
Citation Context ...ly expressed by the linear combination of these eigenvectors A i . So, no matter how is p chosen, the second term in (5) could always be minimized to 0. Therefore, i Ai(x) [ I(W(x;p) A0(x)] λ = ∑ ⋅ − =-=(6)-=- x∈s0 For the first term in (5), we will resort to inverse compositional image alignment method in the next subsection. 2.3 Inverse Compositional Image Alignment for AAM 2 0 ⊥ span(A i ) The ordinary ... |

451 | Active appearance models revisited - Matthews, Baker - 2004 |

355 | Statistical Models of Appearance for Computer Vision - Cootes, Taylor |

213 | Active shape models - ‘smart snakes
- Cootes, Taylor
- 1992
(Show Context)
Citation Context ...ove two centers. By using this simple method, the head pose could be roughly calculated after the newly incoming face is fitted. In a mathematical expression, it is: MC-CC) MC-CC) pose = F ; pose = F =-=(12)-=- ( ) ( face width face height x y ) 3.3 Trigger Intelligent Wheelchair by Head Pose The intelligent wheelchair can just be triggered by the above calculated head pose in a linear way. To improve the r... |

82 | FAME - a flexible appearance modelling environment - Stegmann, Ersboll, et al. - 2003 |

52 |
Time Face and Object Tracking as a Component of a Perceptual User Interface
- Bradski, “Real
- 1998
(Show Context)
Citation Context ...of Δp at Δ p= 0, the solution to could be deducted as Δp ⎡ ∂W ⎤ Δ ∇ T ∑ ⎢ 0 ⎥ [ 0 ] −1 p= H A I(W(x;p))- (x) x ⎢⎣ ∂p p= 0⎥⎦ ⊥ span(A i ) T 0 0 x ⎢⎣ ∂p p p= 0⎥⎦ ⊥ ⎢ span(A ) ⎣ ∂ p= 0⎥⎦ ⊥ i span(A i) A =-=(8)-=- ⎡ ∂W ⎤ ⎡ ∂W ⎤ H = ∑ ⎢∇A ⎥ ⎢∇A ⎥ (9) ∂W where ∇A0 is a 1*2 gradient vector for every pixel in the trained model template image; is a 2* N Jacobian ∂p p= 0 matrix to express the relationship between th... |

51 | Direct appearance models - Hou, Li, et al. - 2001 |

40 | Comparing active shape models with active appearance models - Cootes, Edwards, et al. - 1999 |

28 |
Head gesture recognition for hands‐free control of an intelligent wheelchair
- Gray, Jia, et al.
- 2007
(Show Context)
Citation Context ...ed in the following. Generally speaking, AAM fitting issue could be summarized as an optimization problem to minimize the following L2 norm: 0 M i= 1 M ∑ i= 1 i i 1 2 2 A 0(x)+ ∑ λiA i(x) − I(W(x;p)) =-=(4)-=- where A(x)+ λ A(x) comes from (3); p=( , , ) T p p L p is a vector of N shape parameters, refer to (2); W(x;p) is the destination pixel coordinates warped from the source pixel x by using the shape p... |

26 | Real-time non-rigid driver head tracking for driver mental state estimation - Baker, Matthews, et al. - 2004 |

17 |
The smart wheelchair component system
- Simpson, LoPresti, et al.
(Show Context)
Citation Context ...peaking, an AAM contains a shape model and a texture model. The shape of an AAM is defined as a vector of coordinates for v vertices, which actually makes up a mesh. s = [ x , y , x , y , L , x , y ] =-=(1)-=- 1 1 2 2 AAM shape model looks on an arbitrary face shape s as a base shape 0 plus a linear combination of N shape vectors . s s i s s p s 0 N i= 1 i i v v where p i are the shape parameters and si ar... |

11 |
Head Gesture based control of an Intelligent Wheelchair
- Jia, Hu
- 2005
(Show Context)
Citation Context ...the pixels inside the base mesh 0 . AAM texture model looks on an arbitrary texture A( x) as a base texture A(x) 0 plus a linear combination of M texture vectors A(x) i . A(x) = A (x) +∑ λ A (x) ∀x∈s =-=(3)-=- 0 M i= 1 i i 0 sswhere x ( , ) T = x y indicate the pixel coordinates, λ i are the texture parameters and A i ( x) are the orthonormal eigenvectors. 2.2 Fitting an AAM The key of AAM issue nowadays i... |

6 |
Electric powered wheelchairs
- Ding, Cooper
- 2005
(Show Context)
Citation Context ...arbitrary face shape s as a base shape 0 plus a linear combination of N shape vectors . s s i s s p s 0 N i= 1 i i v v where p i are the shape parameters and si are the orthonormal eigenvectors. = +∑ =-=(2)-=- The texture of an AAM is defined as a vector of coordinate-related intensities over all the pixels inside the base mesh 0 . AAM texture model looks on an arbitrary texture A( x) as a base texture A(x... |

6 |
Recent Advances in Face Detection
- Yang
(Show Context)
Citation Context ...anging, and the second biggest variance just comes from the up-bottom head pose changing as shown in the experimental results. Thus, our first proposed method is to directly tell the head pose by the =-=(10)-=- (11)sfirst two fitted AAM shape parameters. This method is of a straightforward idea, but it's not guaranteed that the training dataset has the above characters. 3.2 Simple Geometric Method So, anoth... |

1 | Lopar: Design and Implementation - unknown authors - 2006 |