Conference Special Session (September 11, 2017)

Developing Vision-Enabled Autonomous Systems Using MATLAB

Speaker: Dr. Amod Anandkumar, Mathworks, India

Single Propagation Techniques for the Unscented Kalman Filter to Reduce Processing Time

Speaker: Dr. Sanat Biswas, Indraprastha Institute of Information Technology Delhi, India

The Unscented Kalman Filter (UKF) is well-known for its better estimation performance than the Extended Kalman Filter (EKF) for highly non-linear system dynamics and observations. However, the UKF also require a large number of numerical operations to compute the a priori mean and error covariance which makes it difficult to implement in resource-constrained real-time applications. Two newly developed fast UKFs based on the Single Propagation Technique and the Extrapolated Single Propagation Technique will be presented. The a priori mean and error covariance computed by these new algorithms are to the first and second order Taylor Series terms. Theoretical analysis and a bench mark re-entry vehicle application show that these new algorithms can reduce the processing time of the UKF by up to 90%. A possible application of these algorithms in computer vision based navigation will be outlined.

Methods for enhancing printer imaging systems performance and better assessing print quality of stochastic clustered-dot halftones

Speaker: Dr. Puneet Goyal, Indian Institute of Technology Ropar, India

Printing industry is estimated at about $230 billion/year worldwide in revenue. Printing employs the process of halftoning i.e. transforming a continuous-tone image into an image with a limited number of tone levels. Direct binary search (DBS) is an iterative halftoning algorithm, considered as gold standard for generating best dispersed-dot halftones but it does not account well for printer-dots development and neighborhood dot interactions in electro-photographic (EP) printers. Recent advancements in estimating dot-interactions better using stochastic models and incorporating those models within the DBS framework will be discussed. New insights gained with the development of these printer models and how it can advance research for developing better printer models in future will also be highlighted.

Print quality assessment is best done by performing psychophysical experiments where subjects evaluate the print methods used for generating the halftones considered. However, this not only leads to subjectivity but is also inefficient in terms of time, cost, and effort. A new compactness measure that shows good potential to quantitatively assess and compare the print quality of different stochastic clustered-dot halftoning methods will be presented. Results using newer metric are almost in agreement with psychophysical experiments results reported earlier.

Back to Top