A Study of Automatically Tracking Pedestrian Flow

The purpose of this work is to analyze the movement of pedestrians in urban areas.I1l At present there is not enough data to analyze the movement of pedestrians a~curately.~~' In the architectural field, tracking and distribution data is now being done by hand.I31 In the image processing field it is difficult to track the movement of pedestrians the same way as a car, because the original image will change and you cannot predict in which direction people will m o ~ e . I ~ l [ ~ ~ It utilizes the algorithm system to show the trajectory of the image and which direction those objects (pedestrians) will move,even though this movement can't be seen yet. The data surveys have proven that this method is effective. 1 Automatic 'hacking Algorithm Please look at Fig. 1 to see how the Algorithm operates. The first part of this system is to extract the moving objects from the original image, and the second part is to track the movement of these objects. 2-1. Extracting Moving Objects After making a survey of the way pedestrians move in town, with the sequence of moving images one can make a background image"). I extract the mobile object(2) from the original image and calculate the difference between the original image and the background image, and label them so I can recognize each object. (Fig.2) 2-2. 'hacking Moving Objects 1) Segmentation of moving objects By using this Formula ( l ) , one can cluster and separate moving objects with the K-mean method, which is an easy way to gather data into one block. (Fig.3) dr,-,,,, =Jks(x2 + y 2 ) + k ~ ( r 2 + g 2 + b 2 ) ( 1 ) (05ks.kc51) (0 5 r,g, b 5 255) (x, y) represents exact location of each pixel. (r ,g ,b) represents each value. This has been done for all of the five dimensions (R, G, B color spaces and x, y position spaces ) .The moving objects have a hierarchical structure similar to the shape of a tree. * Adress : 8-1-1 Gakuennishi-machi, Nishi-ku, Kobe, Hyogo $5 1-2 1, Japan. E-mail : [email protected] [ ALGORITHM ] [ DATA ] . . . . . . Vlleo Images to dtgital Image Or~ginal Images Position of People > Hlerarehical Segmenting / , , , , . , PS , CLUS"IFR [ I PS U B E L 1 Fl:$!s 4 MCHKi PS CLUsnR PS UBEL PS . CLUSTER ( Mlrumum Ihffcrcncr Fig.1 Algorithm of Automatic Tracking