We’re teaching computers to understand video, recognize objects, and enrich video with metadata extracted from it.

    These are some of the ways our team uses deep learning, computer vision, machine learning, and video processing:
  • Focus on algorithms and platforms for effective interaction with multimedia content
  • Develop models for text recognition in natural scene images that are deployed in Watson
  • Extend image-based recognition models to video so they work efficiently and more accurately
  • Use multiple modalities to create video segmentation models and answer questions in video
  • Create deep learning models for video summarization and content detection in video
  • Collaborate with business units who consume our research assets to integrate them in actual products

Our goal is to advance AI capabilities through new methods that model dynamic temporal content like video and contextual data, to analyze and understand it more deeply.

Manager

Udi Barzelay, Manager Video AI Technologies, IBM Research - Haifa

Research

Video Enrichment/Retrieval/Summarization

Video Enrichment / Retrieval / Summarization

Using cognitive computing to discover insights from videos

Video Scene Detection

Video Scene Detection

A fundamental step in video processing aimed at dividing a video into its comprising temporal scenes.

Natural Scene Text Recognition

Natural Scene Text Recognition

A fundamental task in video analysis that supports indexing, surveillance and more.

Video Object Detection

Video Object Detection

Excellent for video analysis such as indexing, surveillance, and more.

CARMEL

CARMEL

We research and develop computer vision algorithms to tackle the challenges of video captured by moving platforms or monitoring equipment, in combination with geographical information (GIS), for applications such as dashboard and body-worn camera analytics.