Citation Context

...The torso is located between the median coordinate and the head along the major axis of the silhouette. Examples of body part labeling results are shown in Fig. 16. 3.3 Tracking of Body Parts Gavrila =-=[12]-=- is a good survey on human body part tracking in 3D. We used a 2D approach to track body parts. W 4 uses template matching and motion prediction to track body parts for upright people. A second order ...

Citation Context

...outdoor scenes because they cannot handle the small motions of background objects such as vegetation (swaying tree branches). In this case, more than one process maybeobservedovertimeatasinglepixel.In=-=[11]-=-,amixtureof three normal distributions was used to model the pixel value for traffic surveillance applications to model road, shadow, and vehicle. In [13], pixel intensity is modeled by a mixture of K...

Citation Context

...ifferent postures. The similarities between the vertical projection of the current silhouette and those average single person vertical projection histogram templates are computed using the SAD method =-=[17]-=- and thresholds determined during training. An example of silhouettes which contain different number of people and their projection are shown in Fig. 8. 3 TRACKING PEOPLE: ISOLATED PERSON Here, we con...

Citation Context

... 4 combines the gray-scale textural appearance and shape information of person together in a 2D dynamic template called a textural temporal template, an extension to the temporal templates defined in =-=[4]-=- tos818 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 22, NO. 8, AUGUST 2000 Fig. 12. Examples of the order of the body parts on the silhouette boundary. recognize action using ...

Citation Context

...its minimum m…x† and maximum n…x† intensity values and the maximum intensity difference d…x† between consecutive frames observed during this training period. Other methods developed in our laboratory =-=[21]-=-, [10] have more sensitivity in detecting foreground regions, but are computationally more intensive. 2.1.1 Learning Initial Background Model W 4 obtains the background model even if there are moving ...

Citation Context

...city. Periodic motion is determined by self-similarity of silhouettes over time using the silhouettes projection histograms. A detailed explanation of periodic motion computation can be found in [8], =-=[19]-=-. W 4 analyzes the vertical projection histogram of the silhouettes to determine whether the foreground region contains multiple people. A set of average normalized vertical projection histogram templ...

Citation Context

...ront of a kiosk. It uses both color information, face detection, and stereo information for detection. However, when people are very close to the kiosk, it can only track a single person. TI's system =-=[29]-=- is a general purpose system for moving object detection and event recognition where moving objects are detected using change detection and tracked using first-order prediction and nearest-neighbor ma...

Citation Context

...rts; however, Pfinder assumes that there is only a single person in the scene and in an upright standing posture. W 4 allows multiple person groups and isolated people in different postures. KidRooms =-=[5]-=- is a tracking system based on ªclosedworld regions.º These are regions of space and time in which the specific context of what is in the regions is known. These regions are tracked in real-time domai...

Citation Context

...distribution of intensity of each pixel. It is designed for indoor surveillance and it cannot handle small motions of background objects. It is a single person tracking system. CMU developed a system =-=[25]-=- that allows a human operator to monitor activities over a large area using a distributed network of active video sensors. Their system can detect and track multiple people and vehicles within clutter...

Citation Context

...ACHINE INTELLIGENCE, VOL. 22, NO. 8, AUGUST 2000 Fig. 3. An example of change-map used in background model computation: (a) input sequence, (b) motion history map, and (c) detection map. MIT's system =-=[13]-=-, [14] uses a distributed set of sensors, and adaptive tracking to calibrate distributed sensors, classify detected objects, learn common patterns of activity for different object classes, and detect ...

Citation Context

...es, according to their sensor types (single or multiple camera, color or gray scale), and their functionality (track single person, multiple people, handle occlusion), as shown as in Table 1. Pfinder =-=[34]-=- has evolved over several years and has been used to recover a 3D description of a person in a large room size space. Pfinder has been used in many applications. It solves the problem of person tracki...

Citation Context

... scenes in which there is a single unoccluded person and fixed camera. Pfinder utilizes a 2D image analysis architecture with two complementary procedures for 2D tracking and initialization. Spfinder =-=[2]-=- is a recent extension of Pfinder in which a widebaseline stereo camera is used to obtain 3D models. Spfinder has been used in a smaller desk-area environment to capture accurate 3D movements of head ...

Citation Context

...ng to calibrate distributed sensors, classify detected objects, learn common patterns of activity for different object classes, and detect unusual activities. Lehigh's omnidirectional tracking system =-=[6]-=- has been used successfully to detect people in camouflage. It has been designed to handle multiple independent moving bodies undergoing nonrigid motion using background subtraction followed by connec...

Citation Context

... humans from other objects. People exhibit periodic motion while they are moving. Cutler previously introduced a robust, image-correlationbased technique to compute the periodicity of a moving object =-=[8]-=-. A computationally inexpensive version of [8], which requires less memory, is employed by W 4 to determine shape periodicity. Periodic motion is determined by self-similarity of silhouettes over time...

Citation Context

...od, a drastic change in sceneillumination,suchascloudsblockingthesun,canleadto gross detection failures. However, one could replace it with a more robust but slow detection algorithm such as in [14], =-=[10]-=-. Another problem that plagues W 4 is shadows. W 4 does not attempt to segment shadows from people's silhouettes. When shadows are large, W 4 has substantial difficulties in Fig. 2. The system archite...

Citation Context

...d method, a drastic change in sceneillumination,suchascloudsblockingthesun,canleadto gross detection failures. However, one could replace it with a more robust but slow detection algorithm such as in =-=[14]-=-, [10]. Another problem that plagues W 4 is shadows. W 4 does not attempt to segment shadows from people's silhouettes. When shadows are large, W 4 has substantial difficulties in Fig. 2. The system a...

Citation Context

...people in camouflage. It has been designed to handle multiple independent moving bodies undergoing nonrigid motion using background subtraction followed by connected component labeling. Oliver et al. =-=[28]-=- and Morris and Hogg [27] describe statistical modeling of human and object interactions. 2 BACKGROUND SCENE MODELING AND PEOPLE DETECTION This section focuses on detection of people in a single image...

Citation Context

...has been designed to handle multiple independent moving bodies undergoing nonrigid motion using background subtraction followed by connected component labeling. Oliver et al. [28] and Morris and Hogg =-=[27]-=- describe statistical modeling of human and object interactions. 2 BACKGROUND SCENE MODELING AND PEOPLE DETECTION This section focuses on detection of people in a single image using shape cues. We des...

Citation Context

...pecific context of what is in the regions is known. These regions are tracked in real-time domains where object motions are not smooth or rigid and where multiple objects are interacting. Smart Kiosk =-=[31]-=- is an application to detect and track people in front of a kiosk. It uses both color information, face detection, and stereo information for detection. However, when people are very close to the kios...

Citation Context

... estimated body postures are then reproduced in a 3D graphical character model by deforming the model according to the estimated data. The dynamics/kinematics model of human motion and Kalman filters =-=[35]-=- are utilized to help the tracking process, as well as to interpolate some 3D joint locations (i.e., elbows). The system runs on a network of Dual-Pentium 400 PCs at 28 frames per second. This project...