Diego Montoya, Benjamin Menhorn, and Frank Slomka
stereo vision, disparity, parallel computation
This paper presents a passive method to calculate the disparity of stereoscopic images. A probabilistic approach is adopted, where two different criteria are used to find a probability density function (PDF) from which the final result can be calculated. The first criterion is a dissimilarity function used to compare the images pixel by pixel. The second is a ``sharpness'' criterion: Both images are overlapped with an offset in direction of the axis separating the two recording devices. As the objects in both images align better, edges overlap. The energy of the high frequency band increases while the energy in the low frequency band decreases. Due to differences in the perspective, this is only valid locally and therefore only applied to a given neighborhood. The results of both criteria are merged to a final PDF providing the final disparity. The main advantage of our approach is the parallelization ability, and thus a scalable implementation, achieving a compromise between accuracy and ease of realization. Furthermore, with minor adjustments our approach can also be used for software refocusing.
Important Links:
Go Back
