Video-based Activity Recognition through Tight Integration of Visual Reasoning and Plan Recognition

Abstract

Program Office Code 311 – Schwartz, Carey Project Summary Video-based Activity Recognition through Tight Integration of Visual Reasoning and Plan Recognition* PI: Baoxin Li & Co-PI: Subbarao Kambhampati The project objective is to develop a systematic approach to video-based activity/event recognition and visual reasoning that can overcome practical challenges through tight integration of low-level visual processing and high-level planning and plan recognition techniques. One key benefit enabled by such an approach is robust visual recognition in complex scenes where precise and complete prior models are in general unavailable. In addition, active perception enabled by high-level reasoning will effectively expand the operational range of the sensing units, making the system operational in situations challenging to a conventional system. These will bring about desired improvements to automated decision systems in Naval missions where visual data need to be acquired and automatically analyzed. The investigators, one Computer Vision researcher and one AI/Planning researcher, team up for developing new methodologies that tightly integrate low-level visual processing models and high-level planning techniques for video-based activity/event recognition. In doing so, we attempt to bridge the gap between two classes of existing approaches: vision approaches that are traditionally model-free (no high-level knowledge models explicitly used) and high-level reasoning such as plan recognition approaches that often assume complete knowledge/model of the problem. The proposed project is built upon the investigators’ respective recent relevant efforts: Li’s work on moving the vision approaches towards being more model-rich through inferencing semantic attributes from sensor data and Kambhampati’s work on model-lite planning and plan recognition with noisy/incomplete inputs. We envision that, by doing so, we will be able to deliver a complete solution with desired level of robustness for automated videobased activity/event recognition.

Document Details

Document Type
DoD Grant Award
Publication Date
Aug 08, 2016
Source ID
N000141512344

Entities

People

  • Baoxin Li

Organizations

  • Arizona State University
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Computer Vision.
  • Research Science/Academic Research

Technology Areas

  • AI & ML