Vehicle detection method based on thermal imaging

• US 9,773,174 B2
• Filed: 12/15/2015
• Issued: 09/26/2017
• Est. Priority Date: 12/15/2015
• Status: Active Grant

• Current Assignee: National Chung-Shan Institute of Science and Technology
• Original Assignee: National Chung-Shan Institute of Science and Technology
• Inventors: Huang, Shih-Shinh, Chien, Shih-Che, Chang, Feng-Chia
• Primary Examiner(s): Lefkowitz, Sumati
• Assistant Examiner(s): Wang, Carol
• Application Number:
  US14/969,217
• Publication Number:
  US 20170169299A1
• Time in Patent Office: 651 Days
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/295 Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
• CPC Class Codes: G06T 2207/10048 Infrared image
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/20112 Image segmentation details
• CPC Class Codes: G06T 2207/30236 Traffic on road, railway or crossing
• CPC Class Codes: G06T 2207/30252 Vehicle exterior; Vicinity of vehicle
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G06V 10/85 Markov-related models; Markov random fields
• CPC Class Codes: G06V 20/54 of traffic, e.g. cars on the road, trains or boats
• CPC Class Codes: G06V 20/584 of vehicle lights or traffic lights

Abstract

A vehicle detection method includes (1) vehicle likelihood region identifying step; (2) vehicle component locating step; and (3) vehicle detecting step. To reduce complexity of calculation and enhance accuracy of detection, the method uses a vehicle likelihood region identifying algorithm to eliminate background regions from a total thermal image and keep vehicle likelihood regions therein for use in further analysis and processing, detects obvious vehicle components, such as vehicle windows and vehicle bottoms, in the thermal image to thereby identify vehicle component likelihood regions, describes a space geometric relationship of vehicle components with a Markov random field model, defines vehicle detection as problems with maximum a posteriori probability, estimates the most likely configuration with an optimization algorithm, so as to effectuate vehicle detection.

Claims

1. A vehicle detection method based on thermal imaging, adapted to capture a total thermal image of a specific region, the vehicle detection method comprising:

an initial vehicle likelihood region identifying step in which a signature cutting algorithm discerns initial vehicle likelihood regions and a background region in the total thermal image and deducts the background region by a background deduction technique to identify the initial vehicle likelihood region;

a vehicle window locating step in which a border detection algorithm and a Hough transform detect an object component which has a feature that a pair of parallel horizontal lines are shown in a thermal image of the initial vehicle likelihood regions and a feature that a center of the thermal image is of low brightness, and the object component is regarded as a located vehicle window;

a vehicle bottom locating step in which the border detection algorithm and the Hough transform detect an additional object component which has a feature that the thermal image of the initial vehicle likelihood regions is slender and a feature that the thermal image is of high brightness, and the additional object component is regarded as a located vehicle bottom; and

a vehicle detecting step in which a space geometric relationship of the located vehicle window and vehicle bottom is described with a Markov random field, wherein, if the space geometric relationship conforms with a predetermined space geometric relationship, a region having the vehicle window and vehicle bottom with the space geometric relationship therebetween is regarded as an advanced vehicle likelihood region, thereby detecting a vehicle.

2. The vehicle detection method of claim 1, wherein the signature cutting algorithm comprises:

a pixel point value defining step for defining numerical values of pixel points included in the total thermal image and attributed to a thermal image value larger than a least thermal image value of the vehicle likelihood region;

a valid vertical area reserving step for calculating vertical projections of the numerical values of the pixel points of the total thermal image, cutting out invalid vertical areas with zero vertical projections, and reserving valid vertical areas with none-zero vertical projections;

a valid horizontal area reserving step for calculating horizontal projections of the numerical values of the pixel points of the valid vertical areas, cutting out invalid horizontal areas with zero horizontal projections, and reserving valid horizontal areas with non-zero horizontal projections; and

a vehicle likelihood region demarcating step for demarcating the valid horizontal areas as the vehicle likelihood regions.

3. The vehicle detection method of claim 2, wherein parameters of the vehicle window are defined as (c_x, c_y, w, h), wherein (c_x, c_y) denotes a center of the vehicle window, and (w, h) denotes width and height of the vehicle window.

4. The vehicle detection method of claim 3, wherein an upper horizontal line and a lower horizontal line are defined by equations as follows:

5. The vehicle detection method of claim 4, wherein u=(u_x, u_y) ε

M^(u) denotes a point of the upper horizontal line, and l=(l_x, l_y) ε

M⁽¹⁾ denotes a point of the lower horizontal line, so as to vote for upper and lower horizontal points S={ (u, l) |u_x=l_x} on a vertical line and parameter space vehicle window (c_y, h), wherein

6. The vehicle detection method of claim 5, wherein, before the Hough transform algorithm starts, a heat grayscale Gaussian model of the vehicle window is analyzed with a Gaussian model, and then a distance between the heat grayscale Gaussian model and a predetermined heat grayscale Gaussian model corresponding to the vehicle window is calculated, wherein, if the distance is larger than a predetermined threshold, the detection is regarded as wrong, so as to eliminate any wrongly identified vehicle window.

7. The vehicle detection method of claim 6, wherein the distance is expressed by an equation as follows:

8. The vehicle detection method of claim 7, wherein the border detection algorithm is a Sobel operator border detection algorithm.

9. The vehicle detection method of claim 1, wherein the Markov random field allows a space geometric relationship of the vehicle window and the vehicle bottom to be defined as a label problem, wherein, a graphical model G=(V, E) is provided, wherein V={v₁, v₂, . . . , v_n} denotes vertices and corresponds to all detected components of vehicles, wherein E={e_l, e₂, . . . , e_m} denote edges, indicating adjacent components of vehicles, wherein, according to the graphical model, a vehicle detection problem is described as how to match each vertex with one of three possible labels, namely false vertex (0), vehicle window (1), and vehicle bottom (2), wherein probability maximization-oriented labeling F={f₁, f₂, . . . , f_n}, and is defined as follows: