J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose esti- mation from facial images with subspace neural networks. In Proc. Int. Neural Network and Brain Conf, pages 528-531, Beijing, China, 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....describe the flesh and hair color of the tracked person. The average error is claimed to be 6.8 (tilt) 5.7 (slant) and 2.9 (roll) but there was no investigation of stability. An appearance based approach similar to the one of [Schiele and Waibel, 1995] is investi gated in [Abraham Mumm, 1998; Bruske et al. 1998] The approach allows computation of slant and tilt. The head is again tracked as a color blob. A square at the detected blob position in the image defines a region of interest (RAT) Within the RaT, complex 2 D Gabor filters (see eq. 2.17) are homogeneously distributed. Different filtering ....

....of the eyes using a Hough transform. The circles of the irises deform to ellipses when the eyes rotate. The approach performs robust parameter estimation of the ellipses. The accuracy was 2.3 . 109 6. 3 Head Pose Estimation with Gabor Wavelet Networks The results reported in [ brahm Mumm, 1998; Bruske et al. 1998] were very promising. However, the filtering scheme that was used was rather rudimentary and straightforward. We will argue that this scheme has two major drawbacks that considerably limit the precision of the approach: In Section 2.6 we explained that for GWNs, precise positioning of novel ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose esti- mation from facial images with subspace neural networks. In Proc. Int. Neural Network and Brain Conf, pages 528-531, Beijing, China, 1998.

....present our approach for the estimation of the pose of a head using Gabor Wavelet Networks. There exist many different approaches for pose estimation, including pose estimation with color blobs [15; 16] pose estimation using a geometrical approach [17] stereo information [18] or neural networks [19], to cite just a few. While in some approaches, such as in [16] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [20] The precision of the geometrical approach [17] was extensively ....

....precise so that they could even be used as a basis for gaze detection such as in [20] The precision of the geometrical approach [17] was extensively tested and verified in [21] The minimal mean pan tilt error that was reached was 1.6 # . In comparison to this, the neural network approach in [19] resulted in a minimal pan tilt error of 0.64 # . The good results in [19] were achieved by first detecting the head using a color tracking approach. Within this region of interest, 16 sets of 4 complex Gabor filters with different orientations of 0, and # were evenly distributed on a ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, G. Sommer, Head-pose estimation from facial images with subspace neural networks, in: Proc. Int. Neural Network and Brain Conf., Beijing, China, 1998, pp. 528--531.

....section we will present results of our experiments for estimating the pose of a face. There exist many di erent approaches for pose estimation, including pose estimation with colour blobs [3; 4; 23] pose estimation applying a geometrical approach [9] stereo information [29] or neural networks [1], to cite just a few. While in some approaches, such as in [4; 23] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [27] The precision of the geometrical approach [9] was extensively ....

....net of just N = 52 odd Gabor wavelets, distributed over the inner face region. For optimization, the scheme that was introduced in section 2 was applied. and veri ed in [19] The minimal mean pan tilt error that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:58 . See table 2 for a summary of results of various approaches. The good result in [1] was reached by rst detecting the head using a color tracking approach. Within the detected color blob region, 8 8 sets of 4 complex Gabor lters with the di erent ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. of Int. Neural Network and Brain Conference, pages 528-531, Beijing, China, 1998.

....GWN In this section we will present the approach for the estimation of the pose of a head. There exist many di erent approaches for pose estimation, including pose estimation with color blobs [3; 17] pose estimation applying a geometrical approach [7] stereo information [24] or neural networks [1], to cite just a few. While in some approaches, such as in [17] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [22] The precision of the geometrical approach [7] was extensively ....

....using the reconstruction formula with an optimal wavelet net of just N = 52 odd Gabor wavelets, distributed over the inner face region. For optimization, the scheme that was introduced in section 2 was applied. that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:64 . The good result in [1] was reached by rst detecting the head using a color tracking approach. Within the detected color blob region, 4 4 sets of 4 complex Gabor lters with the di erent orientations of 0, 4 , 2 and 3 4 were evenly ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. Int. Neural Network and Brain Conf., pages 528-531, Beijing, China, 1998.

....GWN In this section we will present the approach for the estimation of the pose of a head. There exist many different approaches for pose estimation, including pose estimation with color blobs [3; 17] pose estimation applying a geometrical approach [7] stereo information [25] or neural networks [1], to cite just a few. While in some approaches, such as in [17] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [23] The precision of the geometrical approach [7] was extensively ....

....very precise so that they could even be used as a basis for gaze detection such as in [23] The precision of the geometrical approach [7] was extensively tested and verified in [15] The minimal mean pan tilt error that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:64 . The good result in [1] was reached by first detecting the head using a color tracking approach. Within the detected color blob region, 4 4 sets of 4 complex Gabor filters with the different orientations of 0, 4 , 2 and 3 4 were evenly ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. of Int. Neural Network and Brain Conference, pages 528--531, Beijing, China, 1998.

....In this section we will present results of our experiments for estimating the pose of a face. There exist many di erent approaches for pose estimation, including pose estimation with color blobs [3] pose estimation applying a geometrical approach [4] stereo information [13] or neural networks [1], to cite just a few. Color blob approaches give only approximate orientation information. The precision of the geometrical approach [4] was extensively tested and veri ed in [9] The minimal mean pan tilt error that was reached was 1:6 . In comparison to this, the neural network approach in ....

....[1] to cite just a few. Color blob approaches give only approximate orientation information. The precision of the geometrical approach [4] was extensively tested and veri ed in [9] The minimal mean pan tilt error that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:58 . The good result in [1] was reached by rst detecting the head using a color tracking approach. Within the detected color blob region, 4 4 sets of 4 complex Gabor lters with the di erent orientations of 0, 4 , 2 and 3 4 were evenly ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. of Int. Neural Network and Brain Conference, pages 528-531, Beijing, China, 1998.

....section we will present results of our experiments for estimating the pose of a face. There exist many different approaches for pose estimation, including pose estimation with colour blobs [2; 3; 22] pose estimation applying a geometrical approach [8] stereo information [28] or neural networks [1], to cite just a few. While in some approaches, such as in [3; 22] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [26] The precision of the geometrical approach [8] was extensively ....

....I 4;6 using formula (3) with an optimal wavelet net Y of just N = 52 odd Gabor wavelets, distributed over the inner face region. For optimization, the scheme that was introduced in section 2 was applied. that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:58 . See table 2 for a summary of results of various approaches. The good result in [1] was reached by first detecting the head using a color tracking approach. Within the detected color blob region, 8 8 sets of 4 complex Gabor filters with the ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. of Int. Neural Network and Brain Conference, pages 528--531, Beijing, China, 1998.

....section we will present results of our experiments for estimating the pose of a face. There exist many di erent approaches for pose estimation, including pose estimation with colour blobs [2; 3; 16] pose estimation applying a geometrical approach [7] stereo information [21] or neural networks [1], to cite just a few. While in some approaches, such as in [3; 16] only an approximate pose is estimated, other approaches have the goal to be very precise so that they could even be used as a basis for gaze detection such as in [19] The precision of the geometrical approach [7] was extensively ....

....very precise so that they could even be used as a basis for gaze detection such as in [19] The precision of the geometrical approach [7] was extensively tested and veri ed in [12] The minimal mean pan tilt error that was reached was 1:6 . In comparison to this, the neural network approach in [1] reached a minimal pan tilt error of 0:58 . See table 2 for a summary of results of various approaches. The good result in [1] was reached by rst detecting the head using a color tracking approach. Within the detected color blob region, 8 8 sets of 4 complex Gabor lters with the di erent ....

[Article contains additional citation context not shown here]

J. Bruske, E. Abraham-Mumm, J. Pauli, and G. Sommer. Head-pose estimation from facial images with subspace neural networks. In Proc. of Int. Neural Network and Brain Conference, pages 528-531, Beijing, China, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC