Ambrish Tyagi

Background

About Ambrish Tyagi

Researching emerging technologies and leading development of computer vision and machine learning algorithms for future devices and cloud services @ Amazon, Lab126. Health AI Lead for Amazon Halo. Computer Vision: 3D sensors and algorithms, multiple-view geometry, tracking, object detection, SLAM, activity recognition, gestures, optics, embedded system design. Deep Learning: CNNs for various computer vision applications such as object detection, semantic feature representation for tasks such as visual matching, latent modeling of feature spaces, video analysis (activity recognition, etc.), LSTM/RNNs, Siamese variations, etc. At Amazon, worked on 3D head tracking for Dynamic Perspective in Amazon Fire Phone. “CTO” for fashion AI (Echo Look). Cloud Cam. Echo Show. Currently lead CV + Deep Learning activities for Lab126. At Omron, my role involved R&D related to various computer vision techniques applicable to the domain of industrial automation. Responsibilities ranged from product conceptualization, prototyping, designing, manufacturing, and technical leadership. In past I've worked on the problem of human action recognition. I've worked in the area of 2D and 3D object tracking where we developed robust and efficient trackers for single and multiple camera systems. While working with multiple camera systems, I've also addressed the problems of model representation and information fusion across different sensors and views. As part of my work on online filtering on Riemannian manifolds developed a closed form recursive filtering solution for symmetric positive definite matrices with application to visual tracking. My internships at the IBM T. J. Watson research center provided opportunities to work on real-world large scale systems. Specifically, I worked on the indoor smart room CHIL project, and the outdoor surveillance network deployed at The Ohio State University. Research Areas: Computer vision, Machine Learning, Deep Learning, Human-Computer Interaction, Surveillance, Embedded Systems.