Infrared monitoring video quantization parameter cascading method

Abstract

The invention discloses an infrared monitoring video quantization parameter cascading method. The method comprises the steps of determining texture characteristics of a video, determining motion characteristics of the video, determining a relationship between distortion and a code rate, determining a relationship between a Lagrange factor and a quantization parameter of a quantization parameter difference value of adjacent time layers, determining a relationship between the code rate and the quantization parameter, determining a Lagrange rate distortion cost function of the video, and determining an initial value of an ant colony method. The method comprises the following steps of: determining an initial value, determining a state transition probability, updating an initial value, updating pheromones, determining an optimal solution of a Lagrangian rate-distortion cost function, and determining a quantization parameter of each time layer. Compared with the prior art, the method has the advantages of good coding rate distortion performance, low time complexity and the like, and can be applied to the technical fields of video coding and communication.

Description

Technical Field

[0001] The invention belongs to the technical field of video encoding, and in particular relates to the selection of optimal quantization parameters of different time layers in the infrared monitoring video encoding process. Background Art

[0002] Video surveillance plays an increasingly important role in crime prevention, public safety and other fields, effectively protecting people's property safety and maintaining social stability. Due to insufficient light at night, visible light surveillance cannot record effective information, while infrared surveillance based on the principle of infrared imaging can clearly record what happens at night for easy reference. Infrared surveillance video needs to be shot for a long time in practical applications, and the amount of raw data generated during the acquisition process is very large. It is difficult to transmit and store without encoding, so encoding is one of the key technologies. The purpose of encoding is to optimize the picture quality of infrared surveillance video as much as possible while meeting the channel bandwidth and transmission delay constraints.

[0003] The currently used video coding standards adopt a hybrid coding framework based on prediction, transformation, quantization, and entropy coding. Among them, quantization is the root cause of the distortion of the encoded video. The quantization parameter is an important coding parameter in the quantization process, which significantly affects the distortion degree and bit rate of the encoded video. The method of selecting the optimal quantization parameter value for different time layer images in the video is the quantization parameter cascade method.

[0004] The currently proposed quantization parameter cascade method has the following problems:

[0005] The quantization parameter cascade methods proposed so far are all for ordinary natural light videos, while the quantization parameter cascade methods for infrared videos have not yet been proposed. In addition, due to the influence of inter-frame prediction, there will be a close rate-distortion dependency between images of different time layers in the video. However, the current quantization parameter cascade methods based on rate-distortion optimization ideas that consider rate-distortion dependency do not consider the typical content characteristics of infrared surveillance videos, and the rate-distortion models they construct cannot accurately reflect the rate-distortion characteristics of infrared surveillance videos. This ultimately leads to the low efficiency of these methods for infrared surveillance video encoding. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the disadvantages of the above-mentioned prior art and provide an infrared monitoring video quantization parameter cascading method with good coding rate distortion performance and low time complexity.

[0007] The technical solution adopted to solve the above technical problems consists of the following steps:

[0008] (1) Determine video motion characteristics

[0009] The background frame difference method is used to determine the motion characteristics of the video.

[0010] (2) Determine video texture characteristics

[0011] Determine the texture characteristic μ of the video according to formula (1):

[0012]

[0013] Among them, N H Indicates the total number of frames contained in a video, N B Indicates the number of luminance blocks contained in the i-th frame, N F represents the number of pixels contained in the oth block of the i-th frame, I(i,o,g) represents the brightness value of the gth pixel in the oth block of the i-th frame, N H 、N B 、N F is a finite positive integer.

[0014] (3) Determine the relationship between distortion and bit rate

[0015] According to formula (2), the relationship between the distortion and bit rate of each time layer of infrared surveillance video is determined:

[0016]

[0017] Where x represents the time layer number, x∈{0,1,…N L -1},N L Indicates the number of time layers, D x and R x Respectively represent the average coding distortion and bit rate of each image corresponding to the x-th time layer, α x , β x ,δ x 、m x 、n x , k x 、j x 、v x 、w x 、p x ,q x ,t x Represents an intermediate variable that determines the relationship between distortion and bit rate.

[0018] (4) Determine the difference in quantization parameters of adjacent time layers

[0019] According to formula (3), the quantization parameter difference △ of the adjacent time layers corresponding to the xth time layer is determined x :

[0020] △ x =Q x+1 -Q x (3)

[0021] Among them, Qx represents the quantization parameter corresponding to the x-th time layer, Q x A non-negative integer.

[0022] (5) Determine the relationship between Lagrangian factors and quantization parameters

[0023] According to formula (4), the relationship between the Lagrangian factor and the quantization parameter λ of each x time layer is determined x :

[0024]

[0025] Among them, i represents the intermediate variable; a and b represent model parameters, and a and b are positive real numbers.

[0026] (6) Determine the relationship between bit rate and quantization parameter

[0027] The relationship between the bit rate and quantization parameter of the xth time layer is determined according to formula (5):

[0028]

[0029] (7) Determine the Lagrangian rate-distortion cost function of the video

[0030] Determine the Lagrangian rate-distortion cost function J of the video according to formula (6):

[0031]

[0032] Among them, N x Indicates the number of images contained in the x-th time layer, N x is a finite positive integer.

[0033] (8) Determine the initial value of the ant colony method

[0034] Determine the initial value y of the ant colony method according to formula (7): σ :

[0035] y σ =ψ+θ (7)

[0036] θ=r×(μ η ) s +κ

[0037] Among them, ψ is an intermediate variable, σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, l is a finite positive integer, μ η represents the video content characteristic factor, r, s, κ are parameters, and r, s, κ are finite non-zero real numbers.

[0038] (9) Determine the state transition probability

[0039] Determine the state transition probability P according to formula (8):σ :

[0040]

[0041] in, Indicates pheromone, is a finite complex number, represents the minimum value of the pheromone corresponding to all ants, and f(·) represents the function shown in formula (6).

[0042] (10) Update initial value

[0043] Set the transition probability constant to P 0 , P 0 ∈[0,1], update the initial value cyclically according to the following steps:

[0044] 1) Local Search

[0045] P calculated in step (9) σ Less than P 0 , perform local search y′ according to formula (9) σ :

[0046]

[0047] Among them, γ is an intermediate variable, E is the local search step, E is a finite positive real number, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, and G is a finite positive integer.

[0048] 2) Global Search

[0049] P calculated in step (9) σ Greater than or equal to P 0 , perform global search y′(σ) according to formula (10):

[0050] y′ σ =y σ +(X max -X min )×(γ-0.5) (10)

[0051] Among them, X max and X min are the upper and lower bounds of the new initial value, respectively, and X max , X min is a finite integer.

[0052] 3) Limit the upper and lower boundaries of the new initial value

[0053] Limit the updated initial value to X min and X max between.

[0054] (11) Update pheromones

[0055] According to formula (11), the pheromone is updated cyclically.

[0056]

[0057] Where h is the pheromone evaporation coefficient, h∈[0.1,0.9], f(y′ σ ) represents the function value of the initial value of the σth cycle.

[0058] (11) Determine the optimal solution of the Lagrangian rate-distortion cost function

[0059] Repeat steps (9) to (11) until the number of iterations is equal to G, and determine the ant position corresponding to the optimal pheromone, that is, the optimal quantization parameter difference of the current adjacent time layer. Repeat steps (8) to (12) to determine the optimal quantization parameter difference of all adjacent time layers. x,opt .

[0060] (13) Determine the quantization parameters of each time layer

[0061] 1) Determine the quantization parameter Q′ corresponding to the 0th temporal layer of the video according to formula (12): 0

[0062] Q′ 0 =Q 0 -ζ (12)

[0063] Where ζ∈[-10,10].

[0064] 2) Determine the quantization parameters of the remaining time layers

[0065] Determine the optimal quantization parameter Q of the xth time layer according to formula (13): x,opt :

[0066] Q x,opt =Q x-1,opt +A(△ x,opt ) (13)

[0067] Among them, Q x,opt ∈[0,d], d represents the number of quantization parameter values ​​allowed by the video coding standard actually used, d is a finite positive integer, and A(·) is a rounding operation.

[0068] In the equation (5) for determining the relationship between distortion and bit rate in step (6) of the present invention, a and b represent model parameters, a∈(1,10), b∈(5,20).

[0069] In the formula (5) for determining the relationship between distortion and bit rate in step (6) of the present invention, a and b represent model parameters, the best value of a is 4.2005, the best value of b is 13.7122, Q x represents the quantization parameter corresponding to the x-th time layer, Q x A non-negative integer.

[0070] In the formula (7) for determining the initial value of the ant colony method in step (8) of the present invention, the σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, l is 5 to 50, μ η represents the video content characteristic factor, r∈(0,4), s∈(-2,0), κ∈(1,5).

[0071] In the formula (7) for determining the initial value of the ant colony method in step (8) of the present invention, the σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, and the best value of l is 20, μ η It represents the video content characteristic factor. r, s and k are parameters. The best value of r is 2.022, the best value of s is -0.697, and the best value of k is 3.424.

[0072] In the formula (9) for updating the initial value in step (10) of the present invention, γ is a finite real number, E is the local search step, E∈(0,10), C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, and G is the maximum number of iterations, G∈[5,100].

[0073] In the formula (9) of updating the initial value in step (10) of the present invention, the X max and X min are the upper and lower bounds of the new initial value, respectively, and X min ∈[-10,4],X max ∈[5,15].

[0074] In the formula (9) for updating the initial value in step (10) of the present invention, γ is an intermediate variable, E is a local search step, E is 0.5, 0.7, 2.0, 2.2 respectively, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, and the best value of G is 10.

[0075] In the formula (10) for updating the initial value in step (10), the X max and X min are the upper and lower bounds of the new initial value, respectively, and X min The best value is 0, X max The best value is 7.

[0076] In step (13) of the present invention, the quantization parameter Q′ of each time layer is determined 0 In formula (12), ζ∈[-10,10], and the best value of ζ is 3.

[0077] In step (13), the quantization parameter Q of each time layer is determined x,opt In formula (13), Q x,opt ∈[0,d], d represents the number of quantization parameter values ​​allowed by the actual video coding standard, and the optimal value of d is 63.

[0078] Since the present invention adopts the steps of determining the texture characteristics of the video, determining the motion characteristics of the video, and determining the relationship between the distortion and the bit rate, the technical problem that the existing method cannot accurately construct a quantization parameter cascade method suitable for infrared video is solved. The steps of determining the difference of quantization parameters of adjacent time layers, determining the relationship between the Lagrangian factor and the quantization parameter, determining the relationship between the bit rate and the quantization parameter, and determining the Lagrangian rate-distortion cost function of the video solve the problem that the rate-distortion model constructed by the existing method cannot accurately reflect the rate-distortion characteristics of the infrared surveillance video. The method of the present invention was used to carry out comparative simulation experiments, and the experimental results show that the average BD-rate obtained by the present invention is less than that of the comparative method, and the rate-distortion performance is greatly improved. The present invention has the advantages of good coding rate-distortion performance and low time complexity, and can be used in the fields of video coding and communication technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flow chart of embodiment 1 of the present invention.

[0080] Figure 2 1 is a rate-distortion curve of Example 1 of the present invention and a comparative experimental method. DETAILED DESCRIPTION

[0081] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the following embodiments.

[0082] Example 1

[0083] exist Figure 1 The infrared monitoring video quantization parameter cascade method of this embodiment is composed of the following steps:

[0084] (1) Determine video motion characteristics

[0085] The background frame difference method is used to determine the motion characteristics of the video.

[0086] (2) Determine the texture characteristics of the video

[0087] Determine the texture characteristic μ of the video according to formula (1):

[0088]

[0089] Among them, N H Indicates the total number of frames contained in a video, N B Indicates the number of luminance blocks contained in the i-th frame, N F represents the number of pixels contained in the oth block of the i-th frame, I(i,o,g) represents the brightness value of the gth pixel in the oth block of the i-th frame, N H 、N B 、N F is a finite positive integer.

[0090] (3) Determine the relationship between distortion and bit rate

[0091] According to formula (2), the relationship between the distortion and bit rate of each time layer of infrared surveillance video is determined as follows: x :

[0092]

[0093] Where x represents the time layer number, x∈{0,1,…N L -1},N L Indicates the number of time layers, D x and R x Respectively represent the average coding distortion and bit rate of each image corresponding to the x-th time layer. x , β x ,δ x 、m x 、n x , k x 、j x 、v x 、w x 、p x ,q x ,t x Represents an intermediate variable that determines the relationship between distortion and bit rate.

[0094] (4) Determine the difference in quantization parameters of adjacent time layers

[0095] According to formula (3), the quantization parameter difference △ of the adjacent time layers corresponding to the xth time layer is determined x :

[0096] △ x =Q x+1 -Q x (3)

[0097] Among them, Q x represents the quantization parameter corresponding to the x-th time layer, Q x A non-negative integer.

[0098] (5) Determine the relationship between Lagrangian factors and quantization parameters

[0099] According to formula (4), the relationship between the Lagrangian factor and the quantization parameter λ of each x time layer is determined x :

[0100]

[0101] Wherein, i represents an intermediate variable; a and b represent model parameters, a∈(1,10), b∈(5,20), and in this embodiment, the value of a is 4.2005, and the value of b is 13.7122.

[0102] (6) Determine the relationship between bit rate and quantization parameter

[0103] The relationship between the bit rate and quantization parameter of the xth time layer is determined according to formula (5):

[0104]

[0105] (7) Determine the Lagrangian rate-distortion cost function of the video

[0106] Determine the Lagrangian rate-distortion cost function J of the video according to formula (6):

[0107]

[0108] Among them, N x Indicates the number of images contained in the x-th time layer, N x is a finite positive integer.

[0109] (8) Determine the initial value of the ant colony method

[0110] Determine the initial value y of the ant colony method according to formula (7): σ :

[0111] y σ =ψ+θ (7)

[0112] θ=r×(μ η ) s +κ

[0113] Where ψ is an intermediate variable, σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, l is 5 to 30, and the value of l in this embodiment is 20, μ η It represents the video content characteristic factor, r, s, and κ are parameters, r∈(0,4), s∈(-2,0), κ∈(1,5), and in this embodiment, the value of r is 2.022, the value of s is -0.697, and the value of κ is 3.424.

[0114] (9) Determine the state transition probability

[0115] Determine the state transition probability P according to formula (8): σ :

[0116]

[0117] in, Indicates pheromone, represents the minimum value of the pheromone corresponding to all ants, and f(·) represents the function shown in formula (6). is a finite complex number.

[0118] (10) Update initial value

[0119] Set the transition probability constant to P 0 , P 0 ∈[0,1], update the initial value cyclically according to the following steps:

[0120] 1) Local Search

[0121] P calculated in step (9) σ Less than P 0 , perform local search y′ according to formula (9) σ :

[0122]

[0123] Among them, γ is an intermediate variable, E is the local search step size, and the values ​​of E in this embodiment are 0.5, 0.7, 2.0, and 2.2 respectively. C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, and G is the maximum number of iterations, G∈[5,20]. The value of G in this embodiment is 10.

[0124] 2) Global Search

[0125] P calculated in step (9) σ Greater than or equal to P 0 , perform a global search for y′ according to formula (10) σ :

[0126] y′ σ =y σ +(X max -X min )×(γ-0.5) (10)

[0127] Among them, X max and X min are the upper and lower bounds of the new initial value, respectively, and X min ∈[-10,4],X max ∈[5,20], X in this example max The value is 7,X min The value is 0.

[0128] 3) Limit the upper and lower boundaries of the new initial value

[0129] Limit the updated initial value to X min and X max between.

[0130] (11) Update pheromones

[0131] According to formula (11), the pheromone is updated cyclically.

[0132]

[0133] Where h is the pheromone evaporation coefficient, h∈[0.1,0.9], f(y′ σ ) represents the function value of the initial value of the σth cycle.

[0134] (12) Determine the optimal solution of the Lagrangian rate-distortion cost function

[0135] Repeat steps (9) to (11) until the number of iterations is equal to G, and determine the ant position corresponding to the optimal pheromone, that is, the optimal quantization parameter difference of the current adjacent time layer. Repeat steps (8) to (12) to determine the optimal quantization parameter difference of all adjacent time layers. x,opt .

[0136] (13) Determine the quantization parameters of each time layer

[0137] 1) Determine the quantization parameter corresponding to the 0th temporal layer of the video according to formula (12):

[0138] Q′ 0 =Q 0 -ζ (12)

[0139] Among them, ζ∈[-10,10], and the value of ζ in this embodiment is 3.

[0140] 2) Determine the quantization parameters of the remaining time layers

[0141] Determine the optimal quantization parameter Q of the xth time layer according to formula (13): x,opt :

[0142] Q x,opt =Q x-1,opt +A(△ x,opt ) (11)

[0143] Among them, Q x,opt ∈[0,d], d represents the number of quantization parameter values ​​allowed by the video coding standard actually used, the value of d in this embodiment is 63, and A(·) is a rounding operation.

[0144] Complete the infrared surveillance video quantization parameter cascade method.

[0145] Since the present invention adopts the steps of determining the texture characteristics of the video, determining the motion characteristics of the video, and determining the relationship between distortion and bit rate, the technical problem that the existing method cannot accurately construct a quantization parameter cascade method suitable for infrared video is solved, and the steps of determining the difference of quantization parameters of adjacent time layers, determining the relationship between the Lagrangian factor and the quantization parameter, determining the relationship between the bit rate and the quantization parameter, and determining the Lagrangian rate-distortion cost function of the video are determined, which solves the problem that the rate-distortion model constructed by the existing method cannot accurately reflect the rate-distortion characteristics of the infrared surveillance video.

[0146] Example 2

[0147] The infrared monitoring video quantization parameter cascading method of this embodiment is composed of the following steps:

[0148] (1) Determine video motion characteristics

[0149] This step is the same as in Example 1.

[0150] (2) Determine the texture characteristics of the video

[0151] This step is the same as in Example 1.

[0152] (3) Determine the relationship between distortion and bit rate

[0153] This step is the same as in Example 1.

[0154] (4) Determine the difference in quantization parameters of adjacent time layers

[0155] This step is the same as in Example 1.

[0156] (5) Determine the relationship between Lagrangian factors and quantization parameters

[0157] According to formula (4), the relationship between the Lagrangian factor and the quantization parameter λ of each time layer is determined x :

[0158] The expression of formula (4) is the same as that of Example 1.

[0159] In formula (4), i represents an intermediate variable; a and b represent model parameters, a∈(1,10), b∈(5,20), and in this embodiment, a is 1.1 and b is 5.1.

[0160] (6) Determine the relationship between bit rate and quantization parameter

[0161] This step is the same as in Example 1.

[0162] (7) Determine the Lagrangian rate-distortion cost function of the video

[0163] This step is the same as in Example 1.

[0164] (8) Determine the initial value of the ant colony method

[0165] Determine the initial value y of the ant colony method according to formula (7): σ :

[0166] The expression of formula (7) is the same as that of Example 1.

[0167] In formula (7), ψ is an intermediate variable, σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, l ranges from 5 to 30, and in this embodiment, l is 5, μ η It represents the video content characteristic factor, r, s, and κ are parameters, r∈(0,4), s∈(-2,0), κ∈(1,5), and in this embodiment, the value of r is 0.1, the value of s is -1.9, and the value of κ is 1.1.

[0168] (9) Determine the state transition probability

[0169] This step is the same as in Example 1.

[0170] (10) Update initial value

[0171] Set the transition probability constant to P 0 , P 0 ∈[0,1], update the initial value cyclically according to the following steps:

[0172] 1) Local Search

[0173] P calculated in step (9) σ Less than P 0 , perform local search y′ according to formula (9) σ :

[0174] The expression of formula (9) is the same as that of Example 1.

[0175] In formula (9), γ is an intermediate variable, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, G∈[5,100], and the value of G in this embodiment is 5. The meanings and values ​​of other parameters are the same as those in Example 1.

[0176] 2) Global Search

[0177] P calculated in step (9) σ Greater than or equal to P 0 , perform a global search for y′ according to formula (10) σ :

[0178] The expression of formula (10) is the same as that of Example 1.

[0179] In formula (10), X max and X min are the upper and lower bounds of the new initial value, respectively, and X min ∈[-10,4],X max ∈[5,20], X in this example max The value is 5, X min The value is -10.

[0180] The other steps are the same as those in Example 1. The infrared monitoring video quantization parameter cascade method is completed.

[0181] Example 3

[0182] The infrared monitoring video quantization parameter cascading method of this embodiment is composed of the following steps:

[0183] (1) Determine video motion characteristics

[0184] This step is the same as in Example 1.

[0185] (2) Determine the texture characteristics of the video

[0186] This step is the same as in Example 1.

[0187] (3) Determine the relationship between distortion and bit rate

[0188] This step is the same as in Example 1.

[0189] (5) Determine the difference in quantization parameters of adjacent time layers

[0190] This step is the same as in Example 1.

[0191] (5) Determine the relationship between Lagrangian factors and quantization parameters

[0192] The relationship between the Lagrangian factor and the quantization parameter of each time layer is determined according to formula (4):

[0193] The expression of formula (4) is the same as that of Example 1.

[0194] In formula (4), i represents an intermediate variable; a and b represent model parameters, a∈(1,10), b∈(5,20), and in this embodiment, a is 9.9 and b is 19.9.

[0195] (6) Determine the relationship between bit rate and quantization parameter

[0196] This step is the same as in Example 1.

[0197] (7) Determine the Lagrangian rate-distortion cost function of the video

[0198] This step is the same as in Example 1.

[0199] (8) Determine the initial value of the ant colony method

[0200] Determine the initial value y of the ant colony method according to formula (7): σ :

[0201] The expression of formula (7) is the same as that of Example 1.

[0202] In formula (7), ψ is an intermediate variable, σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, and the value of l in this embodiment is 30, μ η It represents the video content characteristic factor, r, s, and κ are parameters, r∈(0,4), s∈(-2,0), κ∈(1,5), and in this embodiment, the value of r is 3.9, the value of s is -0.1, and the value of κ is 4.9.

[0203] (9) Determine the state transition probability

[0204] This step is the same as in Example 1.

[0205] (10) Update initial value

[0206] Set the transition probability constant to P 0 , P 0 ∈[0,1], update the initial value cyclically according to the following steps:

[0207] 1) Local Search

[0208] P calculated in step (9) σ Less than P 0 , perform local search y′ according to formula (8) σ :

[0209] The expression of formula (9) is the same as that of Example 1.

[0210] In formula (9), γ is an intermediate variable, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, G∈[5,100], and the value of G in this embodiment is 100. The meanings and values ​​of other parameters are the same as those in Example 1.

[0211] 2) Global Search

[0212] P calculated in step (9) σ Greater than or equal to P 0 , perform a global search for y′ according to formula (10) σ :

[0213] The expression of formula (10) is the same as that of Example 1.

[0214] In formula (10), X max and Xmin are the upper and lower bounds of the new initial value, respectively, and X min ∈[-10,4],X max ∈[5,20], X in this example max The value is 20, X min The value is 4.

[0215] The other steps are the same as those in Example 1. The infrared monitoring video quantization parameter cascade method is completed.

[0216] In order to verify the beneficial effects of the present invention, a computer simulation comparison experiment was conducted using the method of Example 1 of the present invention (hereinafter referred to as the present invention) and the lambda method (hereinafter referred to as the comparison method). The experimental results are as follows:

[0217] 1. Shoot videos of different scenes

[0218] We took 16 infrared surveillance videos in different scenes, all shot at night, including a variety of common indoor and outdoor application scenes, and divided these videos into two groups. The first group includes 10 videos, the video names are: empty conference room, empty classroom, lakeside, playground, path with stationary vehicles, bicycle parking lot, garden under long lens, building corner under long lens, road with cars passing by shot from overpass, and laboratory corridor with people.

[0219] The second group includes 6 videos, the names of which are: conference room with people, dormitory corridor with people, overpass with people, iron gate with people passing by, canteen entrance with people passing by, overpass with no people. All test sequences are low-latency structures.

[0220] 2. Measuring rate-distortion performance

[0221] The video is encoded using the encoder version VTM16.2, the profile is main_10, the GOP size is 8, the initial QP is 19, 24, 29, 34, and the quantization parameter offset is the △ corresponding to each time layer x,opt The Bjontegaard-Delta rate-distortion ratio (BD-rate for short) is used to measure the rate-distortion performance of different methods.

[0222] The experimental results are shown in Figure 2 , Table 1.

[0223] Table 1 Rate-distortion performance of cascade methods with different quantization parameters

[0224]

[0225] As can be seen from Table 1, the average BD-rates of the first and second groups of videos obtained by the present invention are -5.49% and -6.47% respectively, which are both lower than those of the comparison method, and the rate-distortion performance is greatly improved.

[0226] exist Figure 2 In the figure, a is the rate distortion curve obtained by encoding the video named "No one in the meeting room", and b is the rate distortion curve obtained by encoding the video named "No one in the overpass". The JCTVC-X0038 QPC method is the benchmark method, the lambda-QPC method is the comparison method, and the ISV-QPC method is the method of the present invention.

Claims

1. A cascade method for quantization parameters of infrared surveillance video, characterized in that It consists of the following steps: (1) Determine video motion characteristics The background frame difference method is used to determine the motion characteristics of the video; (2) Determine video texture characteristics Determine the texture characteristic μ of the video according to formula (1): Among them, N H Indicates the total number of frames contained in a video, N B Indicates the number of luminance blocks contained in the i-th frame, N F represents the number of pixels contained in the οth block of the i-th frame, I(i,ο,g) represents the brightness value of the gth pixel in the οth block of the i-th frame, N H 、N B 、N F is a finite positive integer; (3) Determine the relationship between distortion and bit rate According to formula (2), the relationship between the distortion and bit rate of each time layer of infrared surveillance video is determined as follows: x : Where x represents the time layer number, x∈{0,1,…N L -1},N L Indicates the number of time layers, D x and R x Respectively represent the average coding distortion and bit rate of each image corresponding to the x-th time layer, α x , β x ,δ x 、m x 、n x , k x 、j x 、v x 、w x 、p x ,q x ,t x represents an intermediate variable that determines the relationship between distortion and bit rate; (4) Determine the difference in quantization parameters of adjacent time layers According to formula (3), the quantization parameter difference △ of the adjacent time layers corresponding to the xth time layer is determined x : △ x =Q x+1 -Q x (3) Among them, Q x represents the quantization parameter corresponding to the x-th time layer, Q x is a non-negative integer; (5) Determine the relationship between Lagrangian factors and quantization parameters According to formula (4), the relationship between the Lagrangian factor and the quantization parameter λ of each x time layer is determined x : Among them, i represents the intermediate variable; a and b represent the model parameters, and a and b are positive real numbers; (6) Determine the relationship between bit rate and quantization parameter According to formula (5), the relationship between the x-th time layer bit rate and the quantization parameter R is determined x : (7) Determine the Lagrangian rate-distortion cost function of the video Determine the Lagrangian rate-distortion cost function J of the video according to formula (6): Among them, N x Indicates the number of images contained in the x-th time layer, N x is a finite positive integer; (8) Determine the initial value of the ant colony method Determine the initial value y of the ant colony method according to formula (7): σ : y σ =ψ+θ (7) θ=r×(μ η ) s +k Among them, ψ is an intermediate variable, σ represents the number of ants, σ∈{1,2,…,l}, l represents the number of ants, l is a finite positive integer, μ η represents the video content characteristic factor, r, s, κ are parameters, and r, s, κ are finite non-zero real numbers; (9) Determine the state transition probability Determine the state transition probability P according to formula (8): σ : in, Indicates pheromone, is a finite complex number, represents the minimum value of the corresponding pheromone of all ants, and f(·) represents the function shown in formula (6); (10) Update initial value Set the transition probability constant to P0, P0∈[0,1], and update the initial value in a loop according to the following steps; 1) Local Search P calculated in step (9) σ Less than P0, perform local search y according to formula (9) σ ′: Among them, γ is an intermediate variable, E is the local search step, E is a finite positive real number, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, and G is a finite positive integer; 2) Global Search P calculated in step (9) σ Greater than or equal to P0, perform a global search for P′ according to formula (10) σ : and' σ =and σ +(X max -X min )×(γ-0.5) (10) Among them, X max and X min are the upper and lower bounds of the new initial value, respectively, and X max , X min is a finite integer; 3) Limit the upper and lower boundaries of the new initial value Limit the updated initial value to X min and X max between; (11) Update pheromones Update pheromone by looping l times according to formula (11) Where h is the pheromone evaporation coefficient, h∈[0.1,0.9], f(y′ σ ) represents the function value of the initial value of the σth cycle; (12) Determine the optimal solution of the Lagrangian rate-distortion cost function Repeat steps (9) to (11) until the number of iterations is equal to G, determine the ant position corresponding to the optimal pheromone, that is, the optimal quantization parameter difference of the current adjacent time layer, and repeat steps (8) to (12) to determine the optimal quantization parameter difference of all adjacent time layers. x,opt ; (13) Determine the quantization parameters of each time layer 1) Determine the quantization parameter Q0 corresponding to the 0th temporal layer of the video according to formula (12): ′ : Q0 ′ =Q0-ζ (12)where ζ∈[-10,10]; 2) Determine the quantization parameters of the remaining time layers Determine the optimal quantization parameter Q of the xth time layer according to formula (13): x,opt Q x,opt =Q x-1,opt +A(△ x,opt ) (13) Among them, Q x,opt ∈[0,d], d represents the number of quantization parameter values ​​allowed by the video coding standard actually used, b is a finite positive integer, and A(·) is a rounding operation.

2. The infrared monitoring video quantization parameter cascade method according to claim 1, characterized in that: In the equation (4) for determining the relationship between distortion and bit rate in step (5), a and b represent model parameters, a∈(1,10), b∈(5,20).

3. The infrared monitoring video quantization parameter cascade method according to claim 1 or 2, characterized in that: In the equation (4) for determining the relationship between distortion and bit rate in step (5), a and b represent model parameters, a is 4.2005, and b is 13.7122.

4. The infrared monitoring video quantization parameter cascade method according to claim 1, characterized in that: In the formula (7) for determining the initial value of the ant colony method in step (8), σ∈{1,2,…,l}, l represents the number of ants, l∈[5,50], r, s, κ are parameters, r∈(0,4), s∈(-2,0), κ∈(1,5).

5. The infrared monitoring video quantization parameter cascade method according to claim 1 or 4, characterized in that: In step (8) of determining the initial value of the ant colony method, in formula (7), the σ represents the number of ants, σ∈{1,2,…,l}, l is 20, μ η It represents the video content characteristic factor, r, s, and κ are parameters; r is 2.022, s is -0.697, and κ is 3.

424.

6. The infrared monitoring video quantization parameter cascade method according to claim 1, characterized in that: In the formula (9) for updating the initial value in step (10), γ is an intermediate variable, E is the local search step, E∈(0,10), C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, G∈[5,100]; In the formula (10) for updating the initial value in step (10), the X max and X min are the upper and lower bounds of the new initial value; X min ∈[-10,4],X max ∈[5,20].

7. The infrared monitoring video quantization parameter cascade method according to claim 1 or 6, characterized in that: In the formula (9) for updating the initial value in step (10), γ is an intermediate variable, E is the local search step, E is 0.5, 0.7, 2.0, 2.2 respectively, C is the number of iterations of the current ant colony method, C∈{1,2,…,G}, G is the maximum number of iterations, and G is 10; In the formula (10) for updating the initial value in step (10), the X max and X min are the upper and lower bounds of the new initial value, respectively, and X min The value is 0, X max The value is 7.

8. The infrared monitoring video quantization parameter cascade method according to claim 1, characterized in that: In step (13), in formula (12) for determining the quantization parameter Q′0 of each time layer, the value of ζ is 3; In step (13), the quantization parameter Q of each time layer is determined x,opt In formula (13), Q x,opt ∈[0,d], d represents the number of quantization parameter values ​​allowed by the actual video coding standard, and the value of d is 63.

Images