Human head tracking using adaptive appearance models with a fixed-viewpoint pan-tilt-zoom camera

Abstract

We propose a method for detecting and tracking a human head in real time from image sequence. The proposed method has three advantages. 1) We employ a fixedviewpoint pan-tilt-zoom camera to acquire image sequence. With the camera, we eliminate the variations in the head appearance due to camera rotations with respect to the viewpoint. 2) We prepare a variety of contour models of the head appearances and relate them with the camera parameters. This allows us to adaptively select the model to deal with the variations in the head appearance due to human activities. 3) We use the model parameters obtained by detecting the head in the previous image to estimate those to be fitted in the current image. This estimation facilitates computational time for the head detection. Accordingly, the accuracy of the detection and required computational time are both improved and, at the same time, the robust head detection and tracking are realized in almost real time. Experimental results in the real situation show the effectiveness of our method.

Detecting Human Heads With Their Orientations

“…These observations remain invariant against changes in face orientation. We, therefore, model the contour of human-head appearances by the ellipse [1,12,19]. An ellipse has five parameters in the image ( Fig.…”

“…Most existing methods in the literatures, however, focus on only one of these two. Namely, methods to detect human heads/faces (see [1,12,19,20,24], for example) do not estimate orientations of the detected heads/faces, and methods to recognize face orientations/expressions (see [10,14,15,18,21], for example) assume that human faces in an image or an image sequence have been already segmented. Recently, a visual object detection framework was proposed and applied to face detection [16,17].…”

Detecting Human Heads and Face Orientations Under Dynamic Environment

“…We remark that the same evaluation was applied to the method (called the simple-evaluation method (cf. [1,16])) where the ellipse is evaluated only by (2.5), i.e., the gradient magnitude of intensity at the ellipse perimeter. For the position error and the difference of the size error from 1.0, the average and standard deviation over the image sequence were calculated, which is shown in Table 2.…”

“…These observations remain invariant against changes in face orientation. We, therefore, model the contour of human-head appearances by the ellipse [1,12,16].…”

“…Most existing methods in the literatures, however, focus on only one of these two. Namely, methods to detect human heads/faces (see [1,12,16,17,21], for example) do not estimate orientations of the detected heads/faces, and methods to recognize face orientations/expressions (see [10,[13][14][15]18], for example) assume that human faces in an image or an image sequence have been already segmented.…”