System and method for tracking and recognizing people

• US 9,798,923 B2
• Filed: 01/21/2016
• Issued: 10/24/2017
• Est. Priority Date: 11/29/2011
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Yu, Ting, Tu, Peter Henry, Gao, Dashan, Akbay, Kunter Seref, Yao, Yi
• Primary Examiner(s): Thirugnanam, Gandhi
• Application Number:
  US15/002,672
• Publication Number:
  US 20160140386A1
• Time in Patent Office: 642 Days
• CPC Class Codes: G06F 18/213 Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
• CPC Class Codes: G06F 18/23 Clustering techniques
• CPC Class Codes: G06T 7/70 Determining position or orientation of objects or cameras camera calibration G06T7/80
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69
• CPC Class Codes: G06V 40/10 Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands

Abstract

A tracking and recognition system is provided. The system includes a computer vision-based identity recognition system configured to recognize one or more persons, without a priori knowledge of the respective persons, via an online discriminative learning of appearance signature models of the respective persons. The computer vision-based identity recognition system includes a memory physically encoding one or more routines, which when executed, cause the performance of constructing pairwise constraints between the unlabeled tracking samples. The computer vision-based identity recognition system also includes a processor configured to receive unlabeled tracking samples collected from one or more person trackers and to execute the routines stored in the memory via one or more algorithms to construct the pairwise constraints between the unlabeled tracking samples.

Claims

1. A method for tracking and recognition of people, comprising:

generating tracking samples from one or more person trackers of a tracking system;

receiving unlabeled tracking samples from the generated tracking samples into a data buffer for a time span;

generating weighted pairwise constraints between the unlabeled tracking samples;

generating clusters via spectral clustering of the unlabeled tracking samples with weighted pairwise constraints; and

utilizing, via a processor, continuously updated discriminative learning to create a respective appearance signature model for each respective cluster;

wherein the data buffer reaching a threshold size from the received unlabeled tracking samples activates the generation of the weighted pairwise constraints between the unlabeled tracking samples and the clusters and the continuously updated online discriminative learning of the respective appearance signature model for each respective cluster.

2. The method of claim 1, wherein the one or more person trackers comprise 3D ground plane-based trackers maintained in real-time, and generating tracking samples comprises extracting projected image regions from the 3D ground plane-based trackers.

3. The method of claim 1, receiving the unlabeled tracking samples, via batch processing, in an online and asynchronous mode.

4. The method of claim 1, wherein a portion of the received unlabeled tracking samples in the data buffer overlap from two successive time spans.

5. The method of claim 1, wherein the weighted pairwise constraints comprise a must-link constraint between two tracking samples from a single tracker and a cannot-link constraint between two tracking samples from different trackers.

6. The method of claim 1, wherein the respective appearance signature model comprises a new appearance signature model or an updated appearance signature model.

7. A non-transitory, computer-readable media comprising one or more routines which executed by at least one processor causes acts to be performed comprising:

receiving unlabeled tracking samples collected from one or more person trackers into a data buffer for a time span;

generating weighted pairwise constraints between the unlabeled tracking samples;

generating clusters via spectral clustering of the unlabeled tracking samples with weighted pairwise constraints; and

utilizing, via the at least one processor, continuously updated discriminative learning to create a respective appearance signature model for each respective cluster;

wherein the data buffer reaching a threshold size from the received unlabeled tracking samples activates the generation of the weighted pairwise constraints between the unlabeled tracking samples and the clusters and the continuously updated online discriminative learning of the respective appearance signature model for each respective cluster.

8. The non-transitory, computer-readable media of claim 7, wherein the weighted pairwise constraints comprise a must-link constraint between two tracking samples from a single tracker and a cannot-link constraint between two tracking samples from different trackers.

9. The non-transitory, computer readable media of claim 8, wherein the at least one processor utilizes a multi-class support vector machine to learn the respective appearance signature model for each respective cluster.

10. The non-transitory, computer readable media of claim 9, wherein the multi-class support vector machine comprises an incremental support vector machine that continuously updates itself upon receiving new data.