Object tracking apparatus and method and camera

An object tracking method for determining, according to a predetermined object region containing an object in an initial image of an image sequence, an object region estimated to contain the object n each subsequent image of the image sequence, a shape or a size of the object changing in the image sequence, the object tracking method comprising:

• generating a confidence map for the each subsequent image according to the object region of a previous image of the subsequent image of the image sequence, a value of each pixel in the confide indicating a confidence with which a pixel in the subsequent image with a same position as a pixel in the confidence map belongs to the object;
• determining, using a first object tracking method, a first candidate object region of the each subsequent image in the confidence map of the subsequent image, a size of the first candidate object region being fixed for the each subsequent image;
• determining, using a second object tracking method different from the first object tracking method, a second candidate object region of the each subsequent image in the confidence map of the subsequent image based on the first candidate object region of a subsequent image, for the each subsequent image the second candidate object region being adapted to each shape or size of the object in the each subsequent image;
• generating color histograms respectively for the predetermined object region of the initial image, and the first candidate object region and the second candidate object region of the each subsequent image,
• comparing the color histograms of the first and the second candidate object regions of the each subsequent image respectively with the color histogram of the predetermined object region of the initial image to determine each weight for the first candidate object region and the second candidate object region of the each subsequent image, wherein the smaller the difference from the color histogram of the predetermined object region of the initial image is, the larger the weight is;
• and
• calculating, using a processor, a weighted sum of lengths, widths, and centers of the first candidate object region and the second candidate object region of the each subsequent image respectively according to the each weight of the first candidate object region and the second candidate object region of the each subsequent image.

A camera comprising:

• a processor configured to:
• acquire an image sequence and determine an object region containing an object in an initial image of the image sequence;
• determine an object region estimated to contain the object in each subsequent image of the image sequence according to the object region of the initial image, a shape or a size of the object changing in the image sequence;
• generate a confidence map for the each subsequent image according to the object region of a previous image of the subsequent image of the image sequence, a value of each pixel in the confidence map indicating a confidence with which a pixel in the subsequent image with a same position as a pixel in the confidence map belongs to the object;
• determine, using a first object tracking method, a first candidate object region of the each subsequent image in the confidence map of the subsequent image, a size of the first candidate object region being fixed for the each subsequent image;
• determine, using a second object tracking method different from the first object tracking method, a second candidate object region f the each subsequent image in the confidence map of the subsequent image based on the first candidate object region of a subsequent image, for the each subsequent image the second candidate object region being adapted to each shape or size of the object in the each subsequent image;
• generate color histograms respectively for the predetermined object region of the initial image, and the first candidate object region and the second candidate object region of the each subsequent image;
• compare the color histograms of the first and the second candidate object regions of the each subsequent image respectively with the color histogram of the predetermined object region of the initial image to determine each weight for the first candidate object region and the second candidate object region of the each subsequent image, wherein the smaller the difference from the color histogram of the predetermined object region of the initial image is, the larger the weight is;
• and
• calculate a weighted sum of lengths, widths, and centers of the first candidate object region and the second candidate object region of the each subsequent image respectively according to the each weight of the first candidate object region and the second candidate object region of the each subsequent image.

A non-transitory storage medium including machine readable program codes which when being executed on an information processing machine, allow the information processing machine to perform a method for determining, according to a predetermined object region containing an object in an initial image of an image sequence, an object region estimated to contain the object in each subsequent image of the image sequence, a shade or a size of the object changing in the image sequence, the method comprising:

• generating a confidence map for the each subsequent image according to the object region of a previous image of the subsequent image of the image sequence, a value of each pixel in the confidence map indicating a confidence with which a pixel in the subsequent image with a same position as a pixel in the confidence map belongs to the object;
• determining, using a first object tracking method, a first candidate object region of the each subsequent image in the confidence map of the subsequent image, a size of the first candidate object region being fixed for the each subsequent image;
• determining, using a second object tracking method different from the first object tracking method, a second candidate object region of the each subsequent image in the confidence map of the subsequent image based on the first candidate object region of a subsequent image, for the each subsequent image the second candidate object region being adapted to each shape or size of the object in the each subsequent image;
• generating color histograms respectively for the predetermined object region of the initial image, and the first candidate object region and the second candidate object region of the each subsequent image;
• comparing the color histograms of the first and the second candidate object regions of the each subsequent image respectively with the color histogram of the predetermined object region of the initial image to determine each weight for the first candidate object region and the second candidate object region of the each subsequent image, wherein the smaller the difference from the color histogram of the predetermined object region of the initial image is, the larger the weight is;
• and
• calculating a weighted sum of lengths, widths, and centers of the first candidate object region and the second candidate object region of the each subsequent image respectively according to the each weight of the first candidate object region and the second candidate object region of the each subsequent image.