Methods For Analyzing Three Dimensional Scenes

Several methods f o r analyzing three dimensional scenes are discussed w i t h respect to t h e i r advantages and d i f f i c u l t i e s i n g e t t i n g r e a l three d i mensional coordinates. Stereo p i c t u r e processing and l i g h t i n t e r s e c t i n g methods are described along w i t h the r e s u l t of a new hardware system. On the example of a software system some special problems w i t h the a b s t r a c t i o n of scenes w i t h p r i m i t i v e obj e c t s are mentioned. A short look ahead to the f u t u r e closes t h i s paper. 1 I n t r o d u c t i o n When designing a robot system, we have t o deal w i t h the problem of how the robot can get informat i o n about i t s environment. To get f u l l i n formation it seems to be necessary to design a v i s u a l input f a c i l i t y . Therefore there is the problem of how the automaton can get a three dimensional image of the world from v i s u a l i n f o r m a t i o n . This paper describes some systems, designed in the past, and t r i e s to p o i n t out t h e i r problems and d i f f i c u l t i e s . S t a r t i n g from these considerations, a new system i s presented w i t h which some of the w e l l known d i f f i c u l t i e s may be overcome. 2. Stereo p i c t u r e s For i n t e r p r e t a t i o n of stereo p i c t u r e s the system needs two p i c t u r e s of the same scene, taken from d i f f e r e n t viewing angles ( f i g . 1). Provided t h a t the two camera p o s i t i o n s P1 and P2 and the angles a and B are known, the three dimensional c o o r d i nates of the p o i n t B of the scene can be c a l c u l a ted. The main problem in stereo p i c t u r e processing i s t o d e f i n e two conjugate p o i n t s i n the stereo p a i r . A f t e r the d e f i n i t i o n o f one point i n the f i r s t p i c t u r e , the corresponding p o i n t i n the second p i c t u r e must be located. This can only be done i f the p i c t u r e p o i n t contains s i g n i f i c a n t i n formation, e.g. a corner p o i n t . This means t h a t we have t o deal w i t h two d i f f e r e n t problems: f i r s t , we have t o d e f i n e a s i g n i f i c a n t p o i n t i n a f i r s t p i c t u r e , and second, we have to recognize the corresponding p o i n t i n the second p i c t u r e . I n order t o solve the f i r s t problem we have t o apply timeconsuming preprocessing methods l i k e edge enhancement, e t c . The second problem can be solved by two dimensional l o c a l c o r r e l a t i o n methods f\/. Because o f these d i f f i c u l t i e s , stereo methods are not used in f u l l y automatic systems since now. 3. L i g h t i n t e r s e c t i n g methods I n a l l methods described below, i l l u m i n a t i o n o f the scene is regarded as a p a r t of the recognizing system. Then a l l v i s i b l e p o i n t s o f a scene f i t w i t h a s t r a i g h t l i n e of l i g h t or w i t h an i l l u m i n a t i n g plane o f l i g h t . With t h i s information the three d i mensional coordinates o f a l l v i s i b l e p o i n t s can b e c a l c u l a t e d from t h e i r two dimensional coordinates i n the p i c t u r e . Using t h i s p r i n c i p a l several systems were designed. 3.1 Laser range f i n d e r The laser range f i n d i n g system uses a plane of l i g h t w i t h known o r i e n t a t i o n i n space f o r i l l u m i n a t i o n of the scene. The plane of l i g h t is generated w i t h a laser and a c y l i n d r i c a l lens. In such an i l l u m i n a t e d scene the camera is able to see only the i n t e r s e c t i n g l i n e of the l i g h t plane and the surface of the scene. The camera takes a comp l e t e p i c t u r e o f t h i s image and transmits i t t o the computer f o r f u r t h e r processing and e x t r a c t i o n of three dimensional coordinates of the i n t e r s e c t i n g l i n e . By moving the l i g h t source step by step across th whole scene, the process of t r a n s m i t t i n g p i c t u r e s and c a l c u l a t i n g s p a t i a l coordinates i s r e peated. I t i s evident t h a t there i s a large amount of data being t r a n s m i t t e d and processed, one p i c tu r e per l i n e , containing very l i t t l e i n f o r m a t i o n . Therefore the system is i n t o l e r a b l y timeconsuming, Agin and B i n f o r d /5/ need about 5 to 10 minutes per scene. 3.2 P a r a l l e l g r i d i l l u m i n a t i o n In order to get quick r e s u l t s from the system, one must t r y to take the complete information of the scene w i t h only one p i c t u r e , i . e . the scene has to be i l l u m i n a t e d w i t h a l l planes of l i g h t a t the same time. The simplest way to do t h i s is to use an opt i c a l g r i d , being projected cnto the scene. Processing of the p i c t u r e could be done by t r a c i n g the images of the i n t e r s e c t i n g l i n e s in the p i c t u r e . Computation of three dimensional coordinates can b e done c o r r e c t l y only, i f i t i s possible t o i d e n t i f y the images o f the l i n e s o f l i g h t i n the p i c t u r e by t h e i r mathematical equations. This problem causes some r e s t r i c t i o n s t o t h i s method: a) The camera has to be adjusted in such a way t h a t the p i c t u r e s contain p a r t s of the ground surface. b) Shadows, i n t e r r u p t i n g the image o f a l i n e , cause the system to stop t r a c i n g the l i n e . c) The camera p o s i t i o n must be chosen in such a way t h a t l i n e s , running on the top surface of an object,can be d i s t i n g u i s h e d from those running on the ground surface behind the obj