Scalable Data Parallel Implementations of Object Recognition on Connection Machine CM-

Object recognition involves identifying known objects in a given scene. It plays a key role in image understanding, a Grand Challenge problem. Geometric hashing has been recently proposed as a technique f o r model based object recognition in occluded scenes. I n this paper , we present scalable data parallel algori thms f o r geometric hashing on Connection Machine CM-5. Given a scene consisting of S feaiure points, the parallel algorithm for one probe of the recognitionphase takes O ( $ l ogs ) tzme on a fat tree based archttecture. W e perform implementations of the proposed a[gorithms on CM-5, after a careful study of its computation and communication characteristics. Earlier parallel implementations of the geometric hashzng algorithm have been carried out on the Connection Machine CM-2 using O ( M n 3 ) processors, where M ts the number of models in the database and n is the n.umber of features an each model. In these zmplementations, th.e number of processors is independent of the size of the scene but depends on the size of model database which is usually very large. The algorithms presented zn this paper significanfly improve on the number of processors employed while a t the same t ime achieve much superior time performance. Earlier implementations claim 700 to 1300 msec f o r one probe of the recognition phase, assumzng ZOO feature points in the scene on an 8K processor CM-2. Our implementations run on a P processor Connection Machine CM-5, such that 1 5 P 5 S. Our results show that a probe of the recognition phase fo r a scene consisting of 1024 feature points takes less than 10 msec on a 256 processor CM-5. The implementations developed in this paper requzre number of processors independent of the sire of the .model database and are also scalable with the machine size.