IEEE Transactions on Pattern Analysis and Machine Intelligence
Efficient parallel object recognition
Efficient parallel object recognition
| Hi-index | 0.14 |
This paper describes a new formulation of the problem of object recognition under a bounded-error noise model and an object recognition methodology called PERFORM that finds matches by establishing correspondences between model and image features using this formulation. PERFORM evaluates correspondences by intersecting error regions in the image space. The algorithm is analyzed with respect to theoretical complexity as well as actual running times. When a single solution to the matching problem is sought, the time complexity of the sequential matching algorithm for 2D-2D matching using point features is of the order O(I3N2), where N is the number of model features and I is the number of image features. When line features are used, the sequential complexity is of the order O(I2N2). When a single solution is sought, PERFORM runs faster than the fastest known algorithm [8] to solve the bounded-error matching problem. The PERFORM method, which was developed with parallelizability as an important requirement, is shown to be easily realizable on both SIMD and MIMD architectures. The parallel versions of PERFORM are scalable, achieving linear speedups on both the MasPar and the KSR machines. When implemented in parallel, PERFORM does not require a large number of processors or memory, needs minimal to no inter-processor communication, requires no load balancing, and can produce all or just one solution to the matching problem. The communication-efficient version of PERFORM described in this paper has minimal memory requirements, since it only needs to store the model and image features and computes everything else on the fly.