P. McAndrew. Recognising and Locating Objects in Two Dimensional Perspective Views. PhD thesis, Heriot-Watt University, Dept. of Computing and Electrical Engineering, , Sep 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....farmer processors, but the best implementation produced by this work uses a single processor farm with preloading of the input image. This is probably due to the dominance of communication in low level processing. The perspective inversion algorithm was studied in detail on sequential systems [McA90] but was subsequently implemented in a simpler form as an SML prototype and Occam2 implementation [WMM90] The SML prototype identified four major components suitable for implementation on processor farms; generation of point triples (generate.triples) calculation of the rotation translation ....

....Hough programme outputs end point specified line segments whereas the PIA programme requires junctions in the form of point triples. The methods required to perform the conversion are already available and have been well studied. The junctions can be formed by an approach developed by McAndrew [McA90] which is designed to give less weight to co linear lines. The conversion to point triples has already been written as part of the perspective inversion programme [WMM90] to allow input from model description files. It is interesting to note that this missing phase could be spotted by ....

P. McAndrew. Recognising and Locating Objects in Two Dimensional Perspective Views. PhD thesis, Heriot-Watt University, Dept. of Computing and Electrical Engineering, , Sep 1990.

....Hypotheses can be generated by random selection of model and scene features, Hough transformation, or geometry constrained tree search. The RANSAC approach of Fischler and Bolles [24] can be applied to model based vision, this was done by McAndrew to create the Perspective Inversion Algorithm [50, 49]. In the PIA, minimal subsets of the model and scene data are selected and used to determine the viewing parameters. After inversion, the model is projected into the scene and a measure of goodness is determined. The select invert test process continues until a good enough fit has been achieved. ....

.... is preserved and has been used to guide image understanding, e.g. early work on consistent labeling by Clowes [17] and Waltz [74] and later work on perceptual grouping by Lowe [47, 46] Perceptual grouping was used by McAndrew to constrain the RANSAC based perspective inversion algorithm [50]. Fischler and Bolles [24] show that a minimumof 3 point correspondences are required to invert the perspective transformation (though more than one solution may be produced) This allows a clustering method similar to that described for depth data to be employed for hypothesis generation. The ....

Patrick McAndrew. Recognising and locating objects in two dimensional perspective views. PhD thesis, Dept. of Computer Science, Heriot-Watt University, September 1990.

....The Hough program outputs end point specified line subsegments whereas the PIA program requires junctions in the form of point triples. The methods required to perform the conversion are already available and have been well studied. The junctions can be formed by an approach developed by McAndrew (McAndrew, 1990) which is designed to give less weight to co linear lines. The conversion to point triples had already been written as part of the perspective inversion program (Waugh et al. 1990) to allow input from model description files. The junction grouping phase was implemented as a sequential component ....

P. McAndrew. Recognising and Locating Objects in Two Dimensional Perspective Views. PhD thesis, Heriot-Watt University, Dept. of Computing and Electrical Engineering,, Sep 1990.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC