Robust object tracking system
First Claim
Patent Images
1. A method for tracking objects, comprising:
- identifying a target;
identifying a plurality of auxiliary objects related to the target; and
tracking the target using the plurality of auxiliary objects.
2 Assignments
0 Petitions
Accused Products
Abstract
A method for tracking objects includes identifying a target, identifying a plurality of auxiliary objects related to the target, and tracking the target using the plurality of auxiliary objects.
23 Citations
50 Claims
-
1. A method for tracking objects, comprising:
-
identifying a target; identifying a plurality of auxiliary objects related to the target; and tracking the target using the plurality of auxiliary objects. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25)
-
-
26. A tracking system configured to:
-
receive an image sequence; identify a target in the image sequence; identify a plurality of auxiliary objects related to the target; and track the target using the plurality of auxiliary objects. - View Dependent Claims (27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47)
-
-
48. A computer usable medium having computer readable program code embodied therein for causing a computer system to execute a method for tracking objects, comprising:
-
identifying a target; identifying a plurality of auxiliary objects related to the target; and tracking the target using the plurality of auxiliary objects.
-
-
49. A method for tracking a target head in an image sequence, comprising:
-
designating the target head in the image sequence; tracking the target head using an elliptical tracker; identifying a plurality of auxiliary objects in the image sequence related to the target head, wherein identifying comprises; splitting an image in the image sequence using a split-merge quad tree color segmentation algorithm to form a plurality of image regions; creating a plurality of item candidates by merging adjacent image regions of the plurality of image regions with similar appearances into item candidates; pruning item candidates inappropriate for tracking from the plurality of item candidates to obtain a reduced item candidate set; creating a plurality of quantized item candidates by k-means clustering of the reduced item candidate set; creating a plurality of transactions from the plurality of quantized item candidates by grouping neighboring quantized item candidates into transactions; building a transaction database from the plurality of transactions; removing at least one transaction from the plurality of transactions if the at least one transaction has a low co-occurrent frequency with the target head to obtain a remaining plurality of transactions; creating a plurality of candidate auxiliary objects from the remaining plurality of transactions; removing at least one candidate auxiliary object of the plurality of candidate auxiliary objects if the at least one candidate auxiliary object has a low motion correlation with the target head; and creating a plurality of auxiliary objects from the remaining plurality of candidate auxiliary objects; collaboratively tracking the target head using the plurality of auxiliary objects, wherein collaboratively tracking comprises; tracking the target head and each of the plurality of auxiliary objects using a plurality of independent trackers; and using a belief propagation technique to track the target head with the plurality of auxiliary objects, the belief propagation technique further comprising; using a mean-shift technique to determine motion correlations between the target head and the plurality of auxiliary objects; and applying the belief propagation technique to the target head and plurality of auxiliary objects using a star topology Markov random field, wherein the target head comprises the hub of the star topology Markov random field; verifying the consistency between the target head and plurality of auxiliary objects, wherein verifying comprises; identifying a plurality of statistical outliers among the target head and plurality of auxiliary objects using a robust fusion technique; excluding a result of an auxiliary object tracker if an auxiliary object of the plurality of auxiliary objects corresponding to the auxiliary object tracker is an outlier; stopping identification of the plurality of auxiliary objects temporarily if the target head is an outlier; and asserting a tracking failure if a majority of objects comprising the target head and the plurality of auxiliary objects are outliers.
-
-
50. A head tracking system configured to:
-
designate the target head in an image sequence; track the target head using an elliptical tracker; identify a plurality of auxiliary objects in the image sequence related to the target head, wherein identifying comprises; splitting an image in the image sequence using a split-merge quad tree color segmentation algorithm to form a plurality of image regions; creating a plurality of item candidates by merging adjacent image regions of the plurality of image regions with similar appearances into item candidates; pruning item candidates inappropriate for tracking from the plurality of item candidates to obtain a reduced item candidate set; creating a plurality of quantized item candidates by k-means clustering of the reduced item candidate set; creating a plurality of transactions from the plurality of quantized item candidates by grouping neighboring quantized item candidates into transactions; building a transaction database from the plurality of transactions; removing at least one transaction from the plurality of transactions if the at least one transaction has a low co-occurrent frequency with the target head to obtain a remaining plurality of transactions; creating a plurality of candidate auxiliary objects from the remaining plurality of transactions; removing at least one candidate auxiliary object of the plurality of candidate auxiliary objects if the at least one candidate auxiliary object has a low motion correlation with the target head; and creating a plurality of auxiliary objects from the remaining plurality of candidate auxiliary objects; collaboratively track the target head using the plurality of auxiliary objects, wherein collaboratively tracking comprises; tracking the target head and each of the plurality of auxiliary objects using a plurality of independent trackers; and using a belief propagation techniques to track the target head with the plurality of auxiliary objects, the belief propagation technique further comprising; using a mean-shift technique to determine motion correlations between the target head and the plurality of auxiliary objects; and applying the belief propagation technique to the target head and plurality of auxiliary objects using a star topology Markov random field, wherein the target head comprises the hub of the star topology Markov random field; verify the consistency between the target head and plurality of auxiliary objects, wherein verifying comprises; identifying a plurality of statistical outliers among the target head and plurality of auxiliary objects using a robust fusion technique; excluding a result of an auxiliary object tracker if an auxiliary object of the plurality of auxiliary objects corresponding to the auxiliary object tracker is an outlier; stopping identification of the plurality of auxiliary objects temporarily if the target head is an outlier; and asserting a tracking failure if a majority of objects comprising the target head and the plurality of auxiliary objects are outliers.
-
Specification