e, September 1-3. 2010 
In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
59 
rtual detectors 
vehicles 
ite the a posteriori 
ariance P k+] . The 
es can be useful to 
dories diverge. 
ndabout in the city 
indycam TRV14E” 
es. The acquisition 
ition of the images 
las been used. The 
rames, with a time 
Guido et al. have 
btain the foreseen 
tan filter, we can 
•ed car on the left 
xmnding boxes of 
vant labelling: it is 
ing the shadowed 
has been obtained 
; figure 4, the red 
itified, compatible 
liction of Kalman 
les from 78840 to 
ved zone of figure 
for non maxima 
Dr a neighborhood 
'he result of this 
e outlines of the 
rame 78880. The 
jeen used for all 
Section has been 
Figure 13 shows the trajectories reconstructed for four vehicles 
present in figure 2. The image is obtained by using the regions 
detected in 8 consecutive frames. The trajectories are the lines 
connecting the centroids of the regions. 
The data manually obtained by Guido et al. for the next hour, 
used as reference data, have been compared with the results of 
the described procedure. Traffic ad meteo conditions was 
variable and influenced the performances. The best results were 
obtained for low traffic intensity and cloudy weather; in this 
case 95% of vehicles have been correctly recognized and 
tracked. Errors are mainly due to cars hidden by trucks or 
busses. The performances decrease in case of high traffic 
Figure 10. Frame 78830 with bounding boxes 
Figure 12. Frame 78880 with the outlines of the detected 
regions 
density, and the percentage of correctly recognized vehicles 
decreases to 80%. During the test 2475 vehicles crossed the 
roundabout; by using the described procedure without Kalman 
filter, only 1847 were correctly recognized and tracked; with 
Kalman filter 2153 vehicles (87%) were correctly recognized 
and tracked. 
Figure 13. Trajectories of four vehicles 
4. CONCLUSIONS 
A methodology has been described, for the vehicles detection in 
a roundabout and for the determination of their trajectories. 
The differences of images are used, and the frames obtained by 
a camera are compared with a base frame (background) to 
detect the presence and the position of the vehicles. The 
regions, corresponding to the possible vehicles present in the 
roundabout, are isolated by a segmentation operation, using 
edge extraction, region filling and dimensional filtering. 
Recognition and tracking are performed by comparing the 
extracted regions of a frame, and the vehicles present in the 
former frame. A compatibility table is built, which allows the 
vehicle recognition. The trajectory and velocity are obtained. 
The importance of an updated base frame has been underlined, 
and the technique used for the updating has been illustrated. 
The use of Kalman filter to foreseen the position of the vehicles 
has been described and the results of a test lead on a real case 
have been shown. 
The present work regards the optimization of Kalman filter and 
of the procedure. Next studies will be addressed to the 
resolution of the problems to face, with particular regard to the 
elimination of the shadows from the regions associated to the 
vehicles. 
REFERENCES 
Artese, G., 2008. Detecting and Tracking Vehicles in a 
Roundabout. In: International Archives of Photogrammetry 
and Remote Sensing and Spatial Geo informat ion Sciences, 
XXXVI, PART 5/C55, pp. 6-13 
Broggi, A., Dickmanns, E. D., 2000. Applications of computer 
vision to intelligent vehicles. Image and Vision Computing, 
18(5): pp. 365-366. 
Canny, J., 1986. A computational approach to edge detection. 
In: IEEE Transactions on Pattern Analysis and Machine