System and method for detecting and tracking multiple parts of an object

A method of identifying at least one predetermined part of an object, comprising:

• obtaining a model of said object;
• said model describing at least one relationship between said at least one predetermined part and said object;
• extracting an object shape of said object from a first image;
• computing a characteristic dimension value of said at least one predetermined part from said object shape based on said model by:
• determining a plurality of segmentation boundaries based on a first set of thresholds;
• extracting said predetermined part from said object shape based on said segmentation boundaries;
• identifying said at least one predetermined part based on a second set of thresholds;
• measuring dimensional values of said at least one predetermined parts;
• and
• calculating said characteristic dimension value based on said dimensional values of said predetermined parts;
• constructing a mask based on said characteristic dimension value;
• extracting a profile of said at least one predetermined part from said object shape by applying said mask on said object shape;
• and
• identifying said at least one predetermined part from said profile.

A system of detecting a posture of at least one predetermined part of an object, comprising:

• a first image sensor that captures at least one first image of said predetermined part from a first view angle;
• and
• an image processing subsystem coupled to said first image sensor, wherein said image processing subsystem takes said at least one first image as input and produces at least one first feature point location as output;
• said image processing subsystem includes:
• an object extractor module that extracts an object shape of said object from said input;
• a body part locator that locates said at least one predetermined part within a predetermined region of said object shape and determines a characteristic dimension value of said at least one predetermined part by:
• determining a plurality of segmentation boundaries based on a first set of thresholds;
• extracting said predetermined region from said first object shape based on said segmentation boundaries;
• identifying said at least one predetermined part within said predetermined region based on a second set of thresholds;
• measuring dimensional values of said at least one predetermined parts within said predetermined region;
• and calculating said characteristic dimension value based on said dimensional values of said predetermined parts within said predetermined region;
• and
• a feature point locator that identifies at least one first location feature point within said at least one predetermined part in a profile, said profile obtained by masking said first object shape with a mask that is constructed based on said characteristic dimension value;
• wherein said feature point locator further couples said at least one first location feature point to said output of said image processing subsystem;
• thereby said posture can be detected by tracking said at least one first feature point location obtained from said image processing subsystem.