• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A target tracking method, based on a particle swarm optimization algorithm improved in the art of digital image processing, comprising the steps of first performing a box selection on a target to be tracked in a field. sequence of images and, by solving the one-dimensional features of a target region in a TSV color space, describing the target region; then adopting a linearly decreasing inertia weight adjustment strategy, adjusting an inertia weight in the algorithm, and balancing the development and particle search capacity in the algorithm; and finally, tracking the target in the image sequence using the two-swarm algorithm. Thus the learning ability of particles is balanced between an individual optimum position and an overall optimum position, which is favorable for position updating of the particles, and improves tracking efficiency and accuracy.
    公开号:BE1027208B1
    申请号:E20195435
    申请日:2019-07-05
    公开日:2020-11-24
    发明作者:Hui Sun；Jiabin Li；Rui Deng；Meng Li；Longlong Xia；Shigui Zou；Xiaolong Liao；Xuyu Wang
    申请人:Chengdu Qitai Zhilian Information Tech Co Ltd；
    IPC主号:
    专利说明:

    [1] [1] The present invention relates to the technical field of digital image processing, in particular to a target tracking method based on a particle swarm optimization algorithm. PRIOR TECHNIQUE
    [2] [2] A target tracking method is a line of research of particular interest in the field of computer vision. Primary target tracking methods currently include a centroid tracking method, a correlation tracking method, an optical flow method, an average offset tracking method, a Kalman filter tracking method, a tracking method. particulate filter, and the like. As a result of constant research by researchers, a variety of new tracking methods have been developed, intended for use in target tracking in image sequences or videos. The target tracking methods can be used in all aspects of vehicle tracking, pedestrian tracking, medical imaging, and the like. Solving the issue of inaccurate target tracking currently appears to be particularly important.
    [3] [3] The Particle Swarm Optimization Algorithm (PSO) is a new intelligent swarm optimization algorithm. The PSO algorithm is simulated and extracted from the predatory behavior of bird swarms or fish swarms. The particle swarm optimization algorithm has the particularity of being simple and easy to use, does not depend on specific problems and can be applied to different problems. However, the particle swarm optimization algorithm suffers from the phenomenon of diversity of particle losses, and particles are easily found in the state of local optimization, so as to cause premature convergence. Current methods of target tracking based on the particle swarm optimization algorithm include: 1. Yin Hongpeng, Liu Zhaodong, Luo
    [4] [4] In order to solve the technical problem, in accordance with the defects existing in the prior art, the present invention provides a target tracking method based on a particle swarm optimization algorithm so as to achieve an accurate tracking of - targets in a sequence of images.
    [5] [5] In order to solve the technical problem, the present invention adopts the technical scheme of a target tracking method based on an improved particle swarm optimization algorithm, in which the target tracking method comprises the steps following consisting of
    [6] [6] Step 1: read a sequence of images to be processed, and perform a box selection on a target to be continued in the first image frame, so as to obtain the position of the target
    [7] [7] Step 2: Convert the image in a target region from an RGB image to a TSV image according to the selected box target, and calculate one-dimensional characteristics of the target region in the TSV color space in order to describe the target region;
    [8] [8] convert the image in the target region from an RGB image to a TSV image, i.e. convert an RGB color space to a TSV color space, as shown in the following formula: V = max (R, G, B) S = max (R, G, B) -min (R, G, B) u max (R, G, B) 60x (GB) / (SxV), S # 0, max (R, G , B) = RH = 460x (2+ (BR) / (SxV)), S # 0, max (R, G, B) = G 60x (2+ (BR) / (S xV)), S # 0 , max (R, G, B) = B
    [9] [9] wherein in the above formula R represents red, G represents green, B represents blue, the range of values of R, G and B is [0, 255], H represents hue, S represents saturation, V represents visibility, the range of values of H is [0.360], the range of values of S and V is [0, 255];
    [10] [10] and based on the converted TSV color space and according to the visual resolving power of human eyes, divide the H-shade color space into 8 parts, divide the S saturation space into 3 parts, divide the visibility value space V into 3 parts, and construct the one-dimensional features of the target region on this basis to perform feature extraction of the selected target region, in which
    [11] [11] The construction formula for the one-dimensional characteristics M of the target region is as follows: M = 9H + 3S + V
    [12] [12] Step 3: adjust an inertia weight w in the swarm optimization algorithm
    [13] [13] the linearly decreasing inertia weight adjustment process is represented as follows: Woo Wei WE We iter max ter
    [14] [14] where Ym is the maximum inertial weight value in the particle swarm optimization algorithm obtained through a calculation based on the one-dimensional characteristics of the target region, W ,. is the minimum inertia weight value in the particle swarm optimization algorithm obtained via a calculation based on the one-dimensional characteristics of the target region, max / fer is the maximum number of iterations of the algorithm particle swarm optimization, and iter is the current number of iterations.
    [15] [15] Step 4: Track the target in the image sequence using a two-swarm particle swarm optimization algorithm, and output tracking results,
    [16] [16] in which the particle velocity update formula in the two-swarm particle swarm optimization algorithm is represented as follows: VO = WW + er (pbest, - x;) + cr ( gbest —x;)
    [17] [17] where, / = 1,2, L ‚n and n are the swarm size in the particle swarm algorithm, w is the linearly decreasing inertia weight, c, and c, are acceleration factors, 4 and 7, are both random numbers with a range of values of [0,1], vw is the speed of particle 7 at the 1st iteration, ÿ is the speed of particle / at the # + 1 -th iteration, pbest; is the optimum position of particle i at the -th iteration, gbhest ’is the overall optimum position of the particle in the particle swarm at the / -th iteration, and x! is the position of the target point at the # -th iteration.
    [18] [18] The particle position update formula is shown as follows: xx tv 5 [19] where x "represents the position of the target point at the / + 1 -th iteration.
    [20] [20] The target tracking method based on the improved particle swarm optimization algorithm, adopting the technical scheme, has the following beneficial effects: (1) in a traditional algorithm-based target tracking method PSO, the inertia weight w in the PSO algorithm is a constant which is not changed in the whole tracking process; a linearly decreasing inertia weight adjustment strategy is adopted to constantly change the magnitude of inertia weight during iteration, i.e. in an image frame particles start to first by finding the position of a target in a global space and then determining the approximate position of the target, so that the algorithm can accurately determine the position of the target. The method can reduce the number of iterations and improve the operating efficiency of the algorithm. (2) The two-swarm particle swarm optimization algorithm is used for target tracking in a sequence of images, so that the learning ability of particles is balanced between an individual optimum position and a overall optimum position, which is favorable for updating particle position, and tracking efficiency and accuracy can be further improved. DESCRIPTION OF FIGURES
    [21] [21] Figure 1 is a flowchart of a target tracking method based on an improved particle swarm optimization algorithm provided by the embodiment of the present invention;
    [22] [22] Figure 2 shows target tracking results obtained using a method provided by the embodiment of the present invention and the particle swarm optimization algorithm to track the 20th frame of a sequence of. 'images, where (a) is the continuation result of the process, and (b) is the continuation result of the particle swarm optimization algorithm; BE2019 / 5435
    [23] [23] Fig. 3 shows target tracking results obtained using a method provided by the embodiment of the present invention and the particle swarm optimization algorithm to track the 40th frame of a sequence d. 'images, where (a) is the continuation result of the process, and (b) is the continuation result of the particle swarm optimization algorithm;
    [24] [24] Fig. 4 shows the tracking time comparison diagram of two algorithms provided by the embodiment of the present invention;
    [25] [25] Figure 5 shows the tracking error comparison diagram of two algorithms provided by the embodiment of the present invention. DETAILED DESCRIPTION
    [26] [26] The specific embodiment of the present invention is further described in detail in combination with the figures and an embodiment. The following embodiment is used to explain the present invention, rather than to limit the scope of the present invention.
    [27] [27] An image sequence of 70 frames is exemplified in the embodiment, and the target tracking method based on the improved particle swarm optimization algorithm of the present invention is used to track a target in the image sequence.
    [28] [28] As shown in Figure 1, the target tracking method based on the improved particle swarm optimization algorithm comprises the following steps:
    [29] [29] Step 1: read a sequence of images to be processed, and perform a cell selection on a target to be continued in the first image frame, via the cell selection mode by a mouse, so as to obtain the position of the target at the level of the first image frame, namely determining the target to be tracked;
    [30] [30] Step 2: Convert the image in a target region from an RGB image to a TSV image according to the selected box target, and calculate one-dimensional characteristics of the target region in the TSV color space in order to describe the target region:
    [31] [31] convert the image in the target region from an RGB image to a TSV image $ Know how to convert an RGB color space into a TSV color space, as shown in the following formula: V = max (R, G , B) S = max (R, G, B) -min (R, G, B) u max (R, G, B) 60x (GB) / (SxV), S + 0, max (R, G, B) = RH = 460x (2+ (BR) / (SxV)), S # 0, max (R, G, B) = G 60x (2+ (BR) / (S xV)), S # 0, max (R, G, B) = B
    [32] [32] wherein in the above formula R represents red, G represents green, B represents blue, the range of values of R, G and B is [0, 255], H represents hue, S represents saturation, V represents visibility, the range of values of H is [0.360], the range of values of S and V is [0.255];
    [33] [33] and based on the converted TSV color space and according to the visual resolving power of human eyes, divide the H-hue color space into 8 parts, divide the S saturation space into 3 parts, divide the visibility value space V into 3 parts, and construct the one-dimensional features of the target region on this basis to perform feature extraction of the selected target region, in which
    [34] [34] The construction formula for the one-dimensional characteristics M of the target region is as follows: M = 9H + 3S + V
    [35] [35] In the embodiment, a hue space H is divided into 8 parts, the saturation space S is divided into 3 parts, and the visibility value V is divided into 3 parts, specifically represented as:
    [36] [36] Step 3: Fit an inertia weight w in the 5-particle swarm optimization algorithm using a linearly decreasing inertia weight adjustment method according to the one-dimensional characteristics of the target region in l TSV color space, and balance particle development and exploration capabilities in the particle swarm optimization algorithm; in which
    [37] [37] In the particle swarm optimization algorithm, the inertia weight is adjusted according to the number of iterations. The inertia weight is excellent at the initial stage of iterations and is used to find a target in the overall region. The inertia weight is low at the last stage of the iterations and is used to search around the target region to accurately find the position with an overall optimum solution. The inertia weight adjustment strategy adopted in the present invention is the linearly decreasing inertia weight adjustment strategy;
    [38] [38] the linearly decreasing inertia weight adjustment process is represented as follows: Woo Win: WZW iter max ter
    [39] [39] where Ym is the maximum inertial weight value in the particle swarm optimization algorithm obtained via a calculation based on the characteristics
    [40] [40] Step 4: Track the target in the image sequence using a two-swarm particle swarm optimization algorithm, and output tracking results,
    [41] [41] in which the particle velocity update formula in the two-swarm particle swarm optimization algorithm is represented as follows: tH1 __ ttttt VW = wv + er (pbest; Xx;) + cr (gbest -x)
    [42] [42] where, / = 1,2, L ‚n and n are the swarm size in the particle swarm algorithm, w is the linearly decreasing inertia weight, c, and c, are acceleration factors, 4 and 7, are both random numbers with a range of values of [0,1], vw is the particle speed / at the / -th iteration, v * is the speed of particle / at # + 1 -th iteration, pbest; is the optimum position of particle i at the -th iteration, gbhest ’is the overall optimum position of the particle in the particle swarm at the / -th iteration, and x! is the position of the target point at the level of the £ -th iteration. - [43] The particle position update formula is shown as follows: x - x + pt
    [44] [44] where x ”represents the position of the target point at the +1 -th iteration.
    [45] [45] In the embodiment, in the two-swarm particle swarm optimization algorithm, the acceleration factors of a swarm are as follows: c, = 0.5 and c, = 233, and acceleration factors of the other swarm are c = 2.3 and c, = 0.5 to ensure that the two swarms have different traces of movement, so that a larger solution-finding space exists to improve the efficiency of algorithm calculations in So 95635 together.
    [46] [46] In the embodiment, the target in the 70-frame image sequence is tracked using the particle swarm optimization method and algorithm, respectively, and the results are shown in Figs. 2 and 3. Figure 2 shows the results of target tracking of the 20th image frame by the two methods; Fig. 2 (a) shows the result of tracking the target in the image sequence using the particle swarm optimization algorithm; Fig. 2 (b) shows the result of tracking the target in the image sequence using the improved particle swarm optimization algorithm of the present invention; Fig. 3 shows the results of target tracking of the 40th image frame by the two methods; Fig. 3 (a) shows the result of tracking the target in the image sequence using the particle swarm optimization algorithm; and Fig. 3 (b) shows the result of tracking the target in the image sequence using the improved particle swarm optimization algorithm provided by the present invention. According to the above figures, it is possible to see that the method of the present invention has better tracking accuracy.
    [47] [47] In order to objectively assess the target tracking effect of the improved particle swarm optimization algorithm provided by the present invention, the errors of the tracking results and the tracking time between the two tracking methods. pursuit are compared. As shown in Figure 4 and Figure 5, Figure 4 shows the comparison of the target tracking time between the two methods. As shown in Fig. 4, it can be seen that the improved particle swarm optimization algorithm provided by the present invention has a shorter tracking time. Figure 5 shows the comparison of target tracking errors between the two methods. As shown in Fig. 5, it can be seen that the method of the present invention has better tracking accuracy.
    [48] [48] In the embodiment, an experimental comparison of a target tracking in the image sequence verifies that the improved particle swarm optimization algorithm of the present invention exhibits better tracking accuracy over the image. target in the image sequence and a shorter tracking time.
    [49] [49] Finally, it should be noted that the above embodiment is only used to illustrate the technical scheme of the present invention, rather than to limit the present invention.
    . in 2, , . BE7019 / 5435 invention; those skilled in the art should understand that the technical scheme recorded in the embodiment can be further modified, or that part of the technical characteristics or all of the technical characteristics can be replaced in an equivalent manner by the present invention which is illustrated in detail in detail in reference to the above embodiment; and the essence of the corresponding technical scheme does not depart from the scope limited by the claims of the present invention due to modifications or replacements.
    ; BE2019 / 5435 Fig 1 Perform box-selecting on a target to be tracked in | Perform a box selection on a target at the first frame of image so as to determine the target | continue in the first image frame of Convert the image in the target region from an RGB | Convert the image into the target region from image into an HSV image, and calculate the | from an RGB image to a TSV image, and calculate the target HSV color space in the TSV color space Convert the image in the target region from an RGB | Convert the image into the target region from one-dimensional features of the target region in the | one-dimensional characteristics of the target HSV color space region in the TSV color space two-swarm particle swarm optimization algorithm using the two-swarm particle swarm optimization algorithm
    Fig 4
    Fig5
    权利要求:
    Claims (5)
    [1]
    1. Method of target tracking based on an improved particle swarm optimization algorithm, characterized in that it comprises the following steps: Step 1: read a sequence of images to be processed, and perform a box selection on a target to be tracked in the first image frame, so as to obtain the position of the target at the level of the first image frame, namely to determine the target to be tracked; Step 2: Convert the image in a target region from an RGB image to a TSV image according to the selected box target, and calculate one-dimensional characteristics of the target region in the TSV color space to describe the region target: Step 3: Adjust an inertial weight w in the particle swarm optimization algorithm using a linearly decreasing inertia weight adjustment method according to the one-dimensional characteristics of the target region in space TSV color, and balance particle development and exploration capabilities in the particle swarm optimization algorithm; and Step 4: Track the target in the image sequence using a two-swarm particle swarm optimization algorithm, and output tracking results.
    [2]
    2. Method of target tracking based on an improved particle swarm optimization algorithm, according to claim 1, characterized in that in step 1, a box selection is performed on a target to be tracked in the first. image frame via box selection mode via mouse.
    [3]
    3. A method of target tracking based on an improved particle swarm optimization algorithm, according to claim 1, characterized in that in step 2, the method comprises in particular the step of: converting the image in the target region from an RGB image to a TSV image, i.e. converting an RGB color space to a TSV color space, as shown in the following formula: V = max (R, G, B)
    ; BE2019 / 5435 S = max (R, G, B) -min (R, G, B) max (R, G, B) 60x (GB) / (SxV), S # 0, max (R, G, B ) = RH = 460x (2+ (BR) / (SxV)), S # 0, max (R, G, B) = G 60x (2+ (BR) / (S xV)), S # 0, max (R, G, B) = B where in the above formula, R represents red, G represents green, B represents blue, the range of values of R, G and B is [0, 255] , H represents hue, S represents saturation, V represents visibility, the range of values of H is [0.360], the range of values of S and V is [0, 255]; and based on the converted TSV color space and according to the visual resolving power of human eyes, divide the H hue color space into 8 parts, divide the S saturation space into 3 parts, divide the 3-part visibility value space V, and construct the one-dimensional features of the target region on this basis to perform the feature extraction of the selected target region, wherein the construction formula of the one-dimensional feature B of the target region is: B = 9H + 3S + V.
    [4]
    4. A target tracking method based on an improved particle swarm optimization algorithm, according to claim 1, characterized in that in step 3, the linearly decreasing inertia weight adjustment method is indicated. as follows: Wx Wi WZW iter max ter where Wm is the maximum inertial weight value in the particle swarm optimization algorithm obtained via a calculation based on the one-dimensional characteristics of the target region, Wu is the minimum inertia weight value in the particle swarm optimization algorithm obtained via a calculation based on the one-dimensional characteristics of the target region, max / fer is the maximum number of iterations of the algorithm d particle swarm optimization, and iter is the current published number BE2019 / 5435 iterations.
    [5]
    5. A method of target tracking based on an improved particle swarm optimization algorithm, according to claim 4, characterized in that in step 4, the particle velocity update formula in the algorithm. particle swarm optimization is indicated as follows: tH1 __ ttttt VW = wv + er (pbest; Xx;) + cr (gbest -x) where, i = 12, L ‚n and n are the size d 'swarm in the particle swarm algorithm, w is the linearly decreasing inertia weight, c, and c, are acceleration factors, 4 and 7, are both random numbers with a range of values of [0,1], vw is the particle speed / at the / -th iteration, v * is the particle speed / at the # + 1 -th iteration, pbest; is the optimum position of particle i at the -th iteration, gbhest ’is the overall optimum position of the particle in the particle swarm at the / -th iteration, and x! is the position of the target point at the # -th iteration. The particle position update formula is shown as follows: x - x + pt where x ‘represents the position of the target point at the +1 -th iteration.
    类似技术:
    公开号 | 公开日 | 专利标题
    KR101698876B1|2017-01-23|Generation of ghost-free high dynamic range images
    US8824824B2|2014-09-02|Image processing apparatus, imaging apparatus, image processing method, and recording medium storing program
    JP5744437B2|2015-07-08|TRACKING DEVICE, TRACKING METHOD, AND PROGRAM
    CN104574423B|2017-02-22|Single-lens imaging PSF | estimation method based on spherical aberration calibration
    TWI293742B|2008-02-21|
    CN102685511B|2014-09-03|Image processing apparatus and image processing method
    US9886744B2|2018-02-06|Image processing apparatus, image processing method, and storage medium with image quality improvement processing
    US10210643B2|2019-02-19|Image processing apparatus, image processing method, and storage medium storing a program that generates an image from a captured image in which an influence of fine particles in an atmosphere has been reduced
    Cannell et al.2006|Image enhancement by deconvolution
    EP2457379B1|2020-01-15|Method for estimating a defect in an image capture system and associated systems
    US20160360081A1|2016-12-08|Control apparatus, image pickup apparatus, control method, and non-transitory computer-readable storage medium
    US10145790B2|2018-12-04|Image processing apparatus, image processing method, image capturing device and storage medium
    US8577140B2|2013-11-05|Automatic estimation and correction of vignetting
    TWI518635B|2016-01-21|Image restoration method and image processing apparatus using the same
    US20160088277A1|2016-03-24|Imaging device and imaging method
    BE1027208B1|2020-11-24|Target tracking method based on an improved particle swarm optimization algorithm
    CN104766284A|2015-07-08|Underwater color blurred image restoration method
    Stallinga et al.2012|The effect of background on localization uncertainty in single emitter imaging
    Cardinale et al.2009|Bayesian image analysis with on-line confidence estimates and its application to microtubule tracking
    FR2996668A1|2014-04-11|METHOD FOR ACCELERATING MONTE CARLO DELIVERIES
    CN108810415B|2020-07-14|Focusing method based on quantum particle swarm optimization algorithm
    Loke2017|Astronomical image acquisition using an improved track and accumulate method
    CN105785363B|2018-02-13|A kind of span Enhancement Methods of polarization synthetic aperture radar image
    JP6415108B2|2018-10-31|Image processing method, image processing apparatus, imaging apparatus, image processing program, and storage medium
    FR2937209A1|2010-04-16|Image correcting method for camera of mobile phone, involves obtaining three correction information by adding luminous intensity values associated to selected pixels, and determining blank deviation correction based on information
    同族专利:
    公开号 | 公开日
    BE1027208A1|2020-11-17|
    NL2023430B1|2020-10-06|
    CN110288634A|2019-09-27|
    LU101298B1|2020-11-10|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN103914124B|2014-04-04|2016-08-17|浙江工商大学|Energy-conservation Color Mapping Approach towards three-dimensional scenic|
    法律状态:
    2021-01-15| FG| Patent granted|Effective date: 20201124 |
    优先权:
    申请号 | 申请日 | 专利标题
    CN201910484131.4A|CN110288634A|2019-06-05|2019-06-05|A kind of method for tracking target based on Modified particle swarm optimization algorithm|
    [返回顶部]