Video processing system and method for deformation and occlusion resistant object tracking in video content
AssigneeSony Corporation
InventorHouman Rastgar
Life of this patent
Estimating the term…
Abstract
Various aspects of a video processing system and method for object tracking in video content are disclosed herein. The system includes one or more circuits in an electronic device that are configured to detect a current state of an object in a current image frame from a sequence of image frames based on a plurality of first parameters. The system is further configured to adjust a plurality of second parameters for computation of a plurality of features of the object in the current image frame, based on the detected current state. The system is further configured to track the object in the current image frame based on a first feature of the computed plurality of features of the object.
Claims21
A video processing system, comprising:
- one or more circuits in an electronic device, wherein said one or more circuits are configured to:
- capture a sequence of image frames of video content, wherein said sequence of image frames comprises at least a previous image frame and a current image frame, wherein each of said current image frame and said previous image frame comprises at least one object, and wherein said at least one object is in motion;
- determine a template matching score based on a similarity score between at least one feature of said at least one object in said current image frame and at least one feature of said at least one object in said previous image frame;
- determine an appearance similarity score based on a difference between a first color distribution of said at least one object in said current image frame and a second color distribution of said at least one feature of said at least one object in said previous image frame;
- determine a current state of said at least one object in said current image frame based on said appearance similarity score and said template matching score,
- wherein said determined current state corresponds to one of a normal state, an occlusion state, a first reacquisition state or a deformation state;
- adjust a plurality of parameters associated with said current image frame based on said determined current state;
- compute a plurality of features of said at least one object in said current image frame based on said adjusted plurality of parameters;
- and
- track said at least one object in said current image frame based on a first feature of said computed plurality of features of said at least one object in said current frame.
A video processing method, comprising:
- capturing a sequence of image frames of video content, wherein said sequence of image frames comprises at least a previous image frame and a current image frame, wherein said current image frame and said previous image frame comprises at least one object, and wherein said at least one object is in motion;
- determining a template matching score based on a similarity score between at least one feature of said at least one object in said current image frame and at least one feature of said at least one object in said previous image frame;
- determining an appearance similarity score based on a difference between a first color distribution of said at least one object in said current image frame and a second color distribution of said at least one feature of said at least one object in said previous image frame;
- determining a current state of said at least one object in said current image frame based on said appearance similarity score and said template matching score,
- wherein said determined current state corresponds to one of a normal state, and occlusion state, a first reacquisition state, or a deformation state;
- adjusting a plurality of parameters based on said determined current state;
- computing a plurality of features of said at least one object in said current image frame based on said adjusted plurality of parameters;
- and
- tracking said at least one object in said current image frame based on a first feature of said computed plurality of features of said at least one object.
Dependencies are inferred from the claim text.