Comparative Study of Disparity Estimations with Multi-camera Configurations in Relation to Descriptive Parameters of Complex Biological Objects

This paper aims at evaluating multi-camera configurations as a function of the descriptive parameters of complex biological objects. Multi-baseline Stereo has potential to handle projective distortion at large baselines. Being close to the observed object and the orientation of object surfaces pointing toward the camera increase the projection distortion. An example is 3D reconstruction of plants where the leaves can be pointing steeply toward the cameras, while, sub-leaf reconstruction needs high depth resolution, because the leaves overlap closely to each other. The paper presents a new dissimilarity measure, called Sums of Individual Sums of Squared Differences (SISSD). It takes projection distortion and changing specular highlights into account by learning the gradual changing of the feature window. The method was included in the comparative study that used realistic ray traced plant models, where the descriptive parameters of the objects could be controlled. Other configurations in the experiment were the commonly used Multi-baseline Sum of the Sums of Squared Differences (SSSD), the popular binocular graph cuts, and two trinocular correlation techniques. Comparison is in regard to leaf type, texture and orientation, proportion of occlusion and proportion of changing highlights by computing the overall-, occluded-, and highlightedpercentage of bad matching pixels (pbmp, pbmpocc, and pbmphigh). The results showed a complicated relationship of trade-offs that points toward further development combining the strengths of the individual configurations.