121 
  
  
   
     
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
| 400b dd PG ENRICO RENE UAE ll Mob. RE RA ER rdi per o a pudet i 
| S300} cuconousnnccnn one a use nucadece senc nuls deessent teeth queens een a NE UNE de EEN a Eee 2 te 4 5 .AnnL.....L.....4... See usen Nerea an eeeee eese cerae nete r ete iet eeeed ET. 00.0.2... 4 
are | exd re: oda Le 0 J L1 X a I eedem ; 
ia : d A m ee Man itas, : : : A us a : x i d SE 
JOO -.-------...-.-... 7... 7........ Ael. $e LAND o ee-: ren greens san € LJ... Teen e RAN nuvaesaneeeqentes A eer 4 
ons, / : >. Yeo $y far qq : "a 
0 hoc i ei ium wR aua S A ona =. 
0 50 100 150 200 250 100 150 200 250 
»rns Correlation Value Correlation Value 
| Figure 5: histogram of correlation values at different distances with texture intensity of 1.7 and 12.2 
ure, | 
ages | ; : 
sary | | Algorithm | Image Size | Add, Sub | Mult | Div [| Time on Sparc 10 | MOPS/s | 
lard | Bilinear Transformation 512x286 440 K 585 K | 145 K 780 ms 59 
this | 3x3 Lowpass 2 - 256x286 590 K v 150 K 1760 ms 38 
mal | Sobel-direction Filter 2 - 256x286 | 1320 K — 290 K 1760 ms 80 
| SAD-Correlation 256x286 660 K — 145 K 700 ms 40 
ther | Thresholding 256 x 286 73 K — — 90 ms 3.6 
OWS | Total o 3083 K 585 K | 730 K 5090 ms 220.6 
ion. 
ince | Table 3: Required operations and computing time per image and computing power for 50 frames/s 
and 
4.3 Required Computing power 
Algorithms for image processing often consist of simple operations applied to a large amount of data. For general 
| purpose processors this has the consequence, that the computing power for data handling (reading and writing pixel- 
and | data, processing loops) outperforms the computing power needed for data processing. Therefore it is often more efficient 
| to implement such algorithms completely in hardware. In table 3 the approximate number of operations and the 
(5) | computation time on a SPARC 10 is given. Computing power for data handling was not taken into account and only 
| the relevant part of the image (smaller than captured image) was considered. 
hich | 
tems 5. CONCLUSION AND FURTHER WORK 
han 
ctor | We presented a new method which checks a flexible defined separation surface between two workspaces for intruding 
n of | objects. A software prototype implementation showed that the method is feasible. In addition the achievable surface 
thickness, the detection rate of intruding objects and the minimal necessary texture intensity were measured. Current 
we work on the implementation of the method in hardware, so that the required processing can be done in video real 
; time with a pipeline delay of 40 ms (= 2 frames). As a next step we will work on the entire calibration of the cameras 
v— in order to be able to calculate the necessary transformation parameters for arbitrary surfaces. 
6. ACKNOWLEDGMENT 
= 7 We would like to thank Bruno Schneuwly for his advice and the many interesting and fruitful discussions during this 
d project. 
9 7. REFERENCES 
7x7 BAERVELDT, A-.J. 1992 (October). A Safety System for Close Interaction between Man and Robot. In: SAFECOMP’92, 
5x5] IFAC Int. Conf. on Safety, Security and Reliability of Computers. 
GouvIANAKIS, N., PARTHENIS, K., & DIMITRIADIS, B. 1991. A method for detection and tracking of moving objects 
e] in an industrial environment using stereo vision. Pages 349-56 of: TZAFESTAS, S.G. (ed), Engineering Systems with 
ed Intelligence. Concepts, Tools and Applications. Kluwer Academic Publishers Dordrecht, Netherlands. 
NISHIHARA, H.K. 1984. PRISM: a practical real time imaging stereo matcher. Pages 536-545 of: Optical engineering, 
m : vol. 23(5). 
aid RECHSTEINER, MARTIN, SCHNEUWLY, BRUNO, & TROESTER, GERHARD. 1994. Workspace monitoring system. Pages 
res) 689-696 of: EBNER, H., HEIPKE, C., & EDER, K. (eds), Spatial information from digital photogrammetry and computer 
09 vision, vol. 30. Munich, Germany: SPIE, Washington, USA, for ISPRS Commission III. 
SKIFSTAD, K., & JAIN, R. 1989. Illumination Independent Change Detection for Real World Image Sequences. In: 
eral Computer Vision, Graphics and Image Processing 46. 
VISCHER, D. 1992 (May, 10-15). Cooperating Robot with Visual and Taktile Skills. In: Proc. of the IEEE International 
Conference on Robotics and Automation. 
95 IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995