Object Matching in Videos - Isis Project No 1484

Isis Innovation, the technology transfer arm of the University of Oxford, now has for licensing a new approach to object matching in videos.

Marketing Opportunity

Retrieval of objects from video footage is a challenging problem because an object's visual appearance may be very different owing to the viewpoint and lighting, and it may be partially occluded, but some successful methods do now exist. Typically, in such methods, an image of an object is represented by a set of overlapping regions, each represented by a vector computed from the region's appearance. Other methods have been proposed whereby object matching and retrieval has been achieved based on colour and/or texture histograms. However, known methods for object retrieval possess several shortcomings; none are sufficiently robust against occlusion, clutter, viewpoint changes and image intensity changes. Additionally they tend not to be efficient enough if a search for an object is required to be carried out in respect of large numbers of individual images. Potential applications include security, archived library footage and screening of products for offensive material.

The Oxford Invention

The Oxford Invention describes an approach to object and scene retrieval that searches for and localizes all the occurrences of a user outlined object in a video. The method is analogous to text retrieval where matches on descriptors are pre-computed. Retrieval is immediate returning a ranked list of key frames or shots in the manner of a text search engine. We now have immediate run-time object retrieval throughout a movie database, despite significant viewpoint changes in many frames.

Patent Status

This work is the subject of patent application, and Isis would like to talk to companies interested in developing the commercial opportunity that this represents. Please contact the Isis Project Manager to discuss this further.

 

Request Further Information: Project Number 1484 - Object Matching in Videos