FAST OBJECT DETECTION METHOD BASED ON DEFORMABLE PART MODEL (DPM)

• US 20160210530A1
• Filed: 07/28/2015
• Published: 07/21/2016
• Est. Priority Date: 08/22/2014
• Status: Active Grant

• Current Assignee: Zhejiang Shenghui Lighting Company Limited (Tongxiang Shenghui Lighting Electrical Company Limited)
• Original Assignee: Zhejiang Shenghui Lighting Company Limited (Tongxiang Shenghui Lighting Electrical Company Limited)
• Inventors: LI, MAOLIN, SHEN, JINXIANG
• Granted Patent: US 9,846,821 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06V 10/422 for representing the structure of the pattern or shape of an object therefor
• CPC Class Codes: G06V 10/467 Encoded features or binary features, e.g. local binary patterns [LBP]
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 20/41 Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items segmenting video sequences G06V20/49
• CPC Class Codes: G06V 20/49 Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes

Abstract

A fast object detection method based on deformable part model (DPM) is provided. The method includes importing a trained classifier for object detection, receiving an image frame from a plurality of frames in a video captured by a camera, and identifying regions possibly containing at least one object via objectness measure based on Binarized Normed Gradients (BING). The method also includes calculating Histogram of Oriented Gradients (HOG) feature pyramid of the image frame, performing DPM detection for the identified regions possibly containing the at least one object, and labeling the at least one detected object using at least one rectangle box via non-maximum suppression (NMS). Further, the method includes processing a next frame from the plurality of frames in the captured video until the video ends and outputting object detection results.

Claims

1. A fast object detection method based on an deformable part model (DPM) implemented by an object detection system, comprising:

importing a trained classifier for object detection;

receiving an image frame from a plurality of image frames in a video captured by a camera;

identifying regions that may contain at least one object via objectness measure based on Binarized Normed Gradients (BING);

calculating Histogram of Oriented Gradients (HOG) feature pyramid of the image frame;

performing DPM detection for the identified regions that may contain the at least one object;

labeling the at least one detected object using at least one rectangle box via non-maximum suppression (NMS);

processing a next frame from the plurality of frames in the captured video until the video ends; and

outputting object detection results.

2. The method according to claim 1, further including:

processing each pixel in the first N levels of the HOG feature pyramid in each identified region that may contain the at least one object, wherein N is an integer greater than 1;

calculating a score of a root filter, scores of part filters, and a deformation cost, respectively; and

calculating a score of a window.

3. The method according to claim 2, further including:

when the score of the window is greater than a preset threshold, recording a current location.

4. The method according to claim 2, wherein:

a model for an object with n parts is formally defined by a (n+2)-tuple (F0, P1, . . . , Pn, b), wherein F0 is a root filter, Pi is a model for a i-th part and b is a real-valued bias term; and

provided that a location of each filter in a feature pyramid is z=(p0, . . . , pn), and pi=(xi, yi, li) specifies a level and position of the i-th filter, the score of the window is defined by;

5. The method according to claim 1, wherein calculating Histogram of Oriented Gradients (HOG) feature pyramid of the image frame further includes:

calculating the HOG feature pyramid of the image frame using a look-up-table to speed up the feature calculation.

6. A fast object detection system based on a deformable part model (DPM), comprising:

a camera configured to capture a video;

a receiving module configured to receive an image frame from a plurality of frames in the video captured by the camera;

an objectness measure module configured to identify regions that may contain at least one object via objectness measure based on Binarized Normed Gradients (BING);

a Histogram of Oriented Gradients (HOG) module configured to calculate a HOG feature pyramid of the image frame;

a DPM detection module configured to perform DPM detection for the identified regions that may contain the at least one object; and

a label module configured to label the at least one detected object using at least one rectangle box via non-maximum suppression (NMS).

7. The system according to claim 6, wherein the DPM detection module further includes:

a processing unit configured to process each pixel in the first N levels of the HOG feature pyramid in each identified region that may contain the at least one object, wherein N is an integer greater than 1; and

a calculation unit configured to calculate a score of a root filter, scores of part filters, a deformation cost, and a score of a window, respectively.

8. The system according to claim 7, wherein:

when the score of the window is greater than a preset threshold, a current location is recorded.

9. The system according to claim 7, wherein:

a model for an object with n parts is formally defined by a (n+2)-tuple (F0, P1, . . . , Pn, b), wherein F0 is a root filter, Pi is a model for a i-th part and b is a real-valued bias term; and

provided that a location of each filter in a feature pyramid is z=(p0, . . . , pn), and pi=(xi, yi, li) specifies a level and position of the i-th filter, the score of the window is defined by;

10. The system according to claim 6, wherein:

the HOG module is further configured to calculate the HOG feature pyramid of the image frame using a look-up-table to speed up feature calculation.