International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 5. Hakodate 1998 
Outdoor Human Tracking using Spatio-Temporal Information 
Jun-ichi YAMAGUCHI and Nobue KOBAYASHI 
Engineering Institution 
Sogo Keibi Hosho Co., Ltd. 
Sokei Riverside Bldg. 2-14 Ishijima, Koutou-ku, Tokyo, 135-0014 
E-mail:yamaguchi@sok.co.jp 
JAPAN 
Commission V, Working Group IC V/II 
KEY WORDS: Human Finding, Tracking, Motion Vector, Hough Transform, Image Processing 
ABSTRACT 
In outdoor human finding using image processing, there is a problem of detection error in case of a background change caused by 
a branches and leaves trembling in the wind, a light reflection on the surface of the water and so on. Therefore, the automatic 
human finding is generally restricted within the adaptation to the scene of less disturbance. This paper describes an outdoor 
human tracking method which is uninfluenced by disturbance, using a spatio-temporal information of the image. In this method, 
the motion vector is detected by computing an image changing region. The position of the motion vector is shown by two 
parameters (the angle 0 andthe distance © ), and the shifting quantity is voted to the §- 0 space. The voting data obtained 
from passer-by is concentrated on a local region in the 0-0 space. On the other hand, the voting data obtained from disturbance 
shows a tendency to be distributed at random. The existence of the human is detected by estimating the vote data which continues 
at the same position in the 0-0 space. To track the human, the motion data which contributes to the continuous vote data is 
extracted. Tracking the coordinates of center-gravity of the extracted motion data, a locus of the human is detected. The results of 
the experiment, which was performed to verify the effectiveness of the proposed method, are demonstrated. 
1. INTRODUCTION 
The human tracking is useful for understanding a pattern of 
behavior, counting the number of passer-by, security and so on. 
Many methods on the human finding have proposed up to the 
present, and recently some of them are put to practical use. 
However, those methods are generally restricted within the 
adaptation to the scene of less disturbance. Particularly, the 
methods for outdoor are restricted, because of the influence of 
the disturbance which is a light changing, a branches and 
leaves trembling in the wind, a light reflection on the surface 
of the water and so on [1][2]. Therefore, it is required that the 
automatic human finding method has both the elimination of 
the influence of the disturbance and the elimination of the 
undetected error. 
The authors propose an outdoor human tracking method 
which is uninfluenced by the disturbance, using a spatio- 
temporal information of the image. This method finds the 
passer-by, using the straight line detection ability and noise 
elimination ability of Hough Transform [3]. In the method, the 
motion vector is detected by computing the image changing 
region. The position of the motion vector is shown by two 
parameters ( the angle 0 and the distance o ), and the 
shifting quantity is voted to the 0- o space. If a peak of the 
voted data continues at the same position in the 0- 0 space, 
the existence of the human is detected. To track the human, 
the motion data which contributes to the peak is extracted. 
Tracking the coordinates of center-gravity of the extracted 
motion data, a locus of the human is detected. 
This paper describes the method for tracking the outdoor 
human, using the spatio-temporal information of the image. 
The results of the experiment, which was performed to verify 
the effectiveness of the proposed method, are demonstrated. 
2. HUMAN TRACKING ALGORITHM 
2.1 DETECTION OF EXISTENCE 
The changing region of the image is extracted by comparing 
two images which are obtained at a constant time interval. 
The motion vector is detected by computing the correlation 
about the image changing region in a search area, using a 
matching method [4][5]. In human detection, there are some 
cases that it is difficult to detect the change of the image 
because of a speed, a moving course, a distance from the 
camera and so on. In such case, the detectable passer-by and 
the detectable area are restricted. And so, in order to eliminate 
such restriction as much as possible, the changing region is 
extracted by three steps as shown in fig. 1. The motion data in 
the step, which obtains more motion data than other two steps, 
is used for following processing. In detection of the motion data 
  
     
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
    
    
   
   
    
  
  
  
   
   
   
   
    
    
   
    
   
    
     
tn m