Objective

Education

• Proposed a new algorithm for searching and localizing people in images improving performance by 30 fold over the state-of-the-art; algorithm won 6 out of 10 competitions in international visual-object localization challenge organized by European Network of Excellence programme PASCAL.

• Awarded EGIDE (French agency for international scholars) scholarship for graduate and doctoral research.
• Ranked year's best master thesis project among 35 submitted to college of Vision and Robotics.

• Awarded Distinction; awarded merit-based scholarship during undergraduate years.
• Ranked 27th among 20,000 applicants in All-India examination for entrance in undergraduate school.

Book

• Pioneer book in the subarea connects the mean shift statistical theory to current applications in computer vision and pattern analysis.

Experience

• Developed feature spaces based on oriented histograms of image gradients for searching objects in images. The feature spaces represent dense map of substantially overlapped, locally normalized image blocks. They are very fast to compute and the overall search runs close to real time.
• Introduced new motion based feature spaces for searching people in videos. Local differential motion vectors (from two consecutive images) are mapped to oriented histogram space and fused with static appearance based features. New features capture characteristic motion of a person which can be
• further exploited for activity recognition. The overall performance gain is an order of magnitude better over static image descriptors.
• Investigated on parts based person detection which allows robust search of partly occluded people. Different specific body part detectors are fused with corresponding automatically determined pose information to create a reliable person detector.

• Developed algorithms for creating "motion panoramas" from challenging athletic video sequencescontaining significant camera motion and unconstrained athlete's movements. Motion panoramas creates a static background panorama where pixel-level video statistics are used to automatically remove the foreground athlete motion and a second panorama with athlete motion overlayed on the static panorama.
• Formulated constraints and derived equations for 3-D reconstruction of athlete's trajectories from a single camera; shows feasibility with excellent results on high jump and pole vault videos.
• Proposed time warping algorithm for automatic indexing and synchronization of two or more video sequences containing similar athletic events.

• Led cross-functional team of 5 non-direct reports (distributed across INRIA, France Telecom R&D and Philips Research) delivering a critical "Person Detection and Identification" module worth Euro 200K; the deliverable received special mention for quality of innovation at EU review of Euro 9.9M parent project.
• Represent INRIA at project meetings proposing design and implementation strategies, and exploring futuristic scenarios in digital media content analysis of EU technology commercialization project involving 15 industrial/academic partners and 60 researchers.

• Proposed a statistical solution for multi-scale detection of local anisotropic image structures. The algorithm estimates the optimal anisotropic fit over scale-space for points in n-d space.
• Principle inventor of US patent No. 2003P15288US01; status filed.

• Developed image interpolation algorithms in collaboration with Robotics Research Group, Oxford University, for realistic rendering of novel views from two or more images.
• Represented INRIA at EU project review and convinced reviewers of market potential and superiority (compared to competitors) of the technology developed during the project - resulting continuity of Euro 330,000project.

• Spearheaded development of the core module of image rendering product and helped gain investors' confidence in the startup; delivered the module within 25% of estimated schedule, achieving six monthslead in time-to-market.
• Developed automatic image annotation algorithm reducing graphic designers effort by 35%.
• Developed real-time novel image rendering algorithm essential for making product marketing feasible.

• Performed major architectural and development enhancements for compatibility of Windows CE office product, reducing support effort by 80% - saving $48,000 per annum.

Publications and Patents

• N. Dalal and D. Comaniciu. Multi-scale Detection of Local Image Structures. US Serial No. 2003P15288US01, filed September 2004.

• N. Dalal, B. Triggs and C. Schmid. Object Detection Using Histograms of Oriented Gradients. In preparation for submission to Pattern Analysis and Machine Learning.
• T. Rodriguez, I. Reid, R. Horaud, N. Dalal and M. Goetz. Image interpolation for virtual sports scenarios. In Journal of Machine Vision and Applications, June 2005.
• A. Bartoli, N. Dalal and R. Horaud. Motion Panoramas. In Journal of Computer Animation and Virtual Worlds, December 2004. 15(5): 501-517. (Also INRIA Research Report - 4771, 2003).

• N. Dalal, B. Triggs and C. Schmid. Human Detection Using Oriented Histograms of Flow and Appearance. In Proceedings of the European Conference on Computer Vision, Graz, Austria, May 2006.
• N. Dalal and B. Triggs. Histograms of Oriented Gradients for Human Detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, San Diego, USA, June 2005. Vol. II, pp. 886-893.
• K. Okada, D. Comaniciu, N. Dalal and A. Krishnan. A Robust Algorithm for Characterizing Anisotropic Local Structures. In Proceedings of the Eighth European Conference on Computer Vision, Prague, Czech Republic, May2004. Vol. I, pp. 549-561.
• N. Dalal and R. Horaud. Indexing Key Positions between Multiple Videos. In Proceedings of the IEEE Workshopon Motion and Video Computing, Orlando, Florida, USA, December 2002, pp. 65-71.
• A. Bartoli, N. Dalal, B. Bose and R. Horaud. From Video Sequences to Motion Panoramas. In Proceedings of theIEEE Workshop on Motion and Video Computing, Orlando, Florida, USA, December 2002, pp. 201-207.

• M. Everingham et al. The 2005 PASCAL Visual Object Classes Challenge. In F. d'Alche-Buc, I. Dagan, J. Quinonero (eds), Selected Proceedings of the first PASCAL Challenges Workshop, LNAI, Springer, (in press) 2006.

• N. Dalal, B. Triggs and C. Schmid. Parts Based Histogram Voting for Semi-occluded Human Detection. INRIA Research Report. In preparation.
• N. Dalal. Using Dynamics Contained in Athletic Video Sequences. MS thesis: Institut National Polytechnique de Grenoble, France, June 2002.

• Peer reviewer for IEEE Transactions on Pattern Analysis and Machine Intelligence (ad-hoc basis), International Conference on Computer Vision (2005), IEEE Conference on Computer Vision and Pattern Recognition (2005,2004), European Conference on Computer Vision (2004).

Additional

• Visiting Scholar: Robotics Research Group, Oxford University, UK, July 2002
• Software Skills: Excellent C/C++, Matlab, Linux/Unix, Windows.
• Multilingual: Fluent in English, Hindi; competent in French.
• Citizenship: Indian.
• References: Upon request
• Interests: Enjoy skiing, hiking, and long bike rides; won city level drawing competitions.