A method for estimating the motion direction of video targets based on bionic model

Abstract

The invention discloses a method for estimating the motion direction of a video target based on a bionic model, and relates to the technical field of bionic visual neural computing and video target tracking. In video target tracking, complex scenes will generate a large amount of interference noise, which brings challenges to the estimation of the horizontal, vertical and depth motion directions of targets in three-dimensional scenes. The insect visual nervous system can accurately and effectively estimate the spatiotemporal motion direction of targets in complex environments during high-speed flight. The invention utilizes the unique physiological characteristics of lobule plate tangential cells LPTC and lobule giant motion detector LGMD neurons in estimating the spatiotemporal motion direction of targets, as well as the unique sensitivity characteristics to target motion, to establish a bionic visual neural pathway model, which estimates the spatiotemporal motion direction of targets by calculating ON and OFF visual pathway signals, thereby being able to accurately and effectively estimate the spatiotemporal motion direction of targets in complex scenes, and having stronger robustness.

Description

Technical Field

[0001] The invention relates to the technical field of bionic visual neural computing and video target tracking, and in particular to a method for estimating the motion direction of a video target based on a bionic model. Background Art

[0002] In nature, estimating the spatiotemporal motion direction of targets has become one of the basic skills of insects, playing an important role in detecting predators, chasing prey, and finding mates. The lightweight spatiotemporal motion direction estimation neural system of insects provides a good example for establishing a bionic visual neural pathway model for video target three-dimensional motion direction estimation. Among the motor neurons of insects, there are two special types of neurons, called LPTC (Lobula Plate Tangential Cell) and LGMD (Lobula Giant Movement Derector). LPTC has a strong response to the visual stimulation of translational motion of large fields of view or prominent targets, and LGMD is very sensitive to the visual stimulation of collision motion of targets.

[0003] In terms of using bionic models to estimate the direction of target motion, researchers proposed the EMD (Flementary Motion Detector) computational model in the early stage, but its detection performance in complex three-dimensional scenes is not ideal and it is insensitive to the target's depth motion direction. Through further research on the computational models based on LPTC and LGMD motor neurons, it was found that the relevant computational models are independent of each other, that is, they realize their own functions separately, and the cooperation between different motor neurons is not integrated. By studying the separate visual pathways based on LPTC and LGMD motor neurons, the computational model based on LPTC motor neurons is mainly used to estimate the direction of target motion in the translation plane, and the computational model based on LGMD motor neurons is mainly used for obstacle avoidance, that is, estimating the direction of target motion in the depth plane. The complex three-dimensional scene is a three-dimensional space composed of the translation plane and the depth plane. The separate computational models based on LPTC and LGMD motor neurons cannot fully meet the estimation of the target motion direction in complex three-dimensional scenes. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for estimating the motion direction of a video target based on a bionic model, comprising the following steps:

[0005] S1. Construct a bionic visual neural pathway model, which includes the retinal neural layer, the lamina neural layer, the medullary neural layer and the lobular complex neural layer in sequence;

[0006] S2, in the retinal neural layer, discrete visual stimuli are captured frame by frame by photoreceptors;

[0007] S3, in the lamina neural layer, the lamina monopolar cells receive visual stimulation from the retinal neural layer and divide the visual stimulation into parallel ON visual pathway and OFF visual pathway, and encode the increase response and decrease response of brightness through the ON visual pathway and OFF visual pathway respectively;

[0008] S4, the neurons of the monopolar cells of the thin plate filter out visual flicker and irrelevant motion through a fast depolarization and slow repolarization mechanism;

[0009] S5, in the medullary neural layer, the upper, lower, left, and right sub-visual pathways of the ON visual pathway are calculated through T4 neurons;

[0010] S6, calculate the response strength of the ON visual pathway and OFF visual pathway signals respectively;

[0011] S7, in the lobule complex neural layer, the response strength of the upper, lower, left and right four sub-visual pathways of the OFF visual pathway is calculated through the T5 neuron; the T4 neuron and the T5 neuron are converged to the upper, lower, left and right four sub-layers of the lobule plate tangential cell neurons, and the horizontal and vertical response strengths are calculated through the response strength of the upper, lower, left and right four sub-visual pathways of the lobule plate tangential cell neurons;

[0012] S8, compare the ON visual pathway with the OFF visual pathway through the neuronal competition mechanism to estimate the depth motion and inhibit the translational motion to obtain the response strength in the depth direction;

[0013] S9, calculating the visual change response value and the corresponding sigmoid membrane potential value, and determining the depth movement direction mark;

[0014] S10. Use the Sigmoid membrane potential function to activate responses in the horizontal, vertical, and depth directions, and finally estimate the spatiotemporal motion direction through the Sigmoid membrane potential value.

[0015] The technical solution further defined in the present invention is:

[0016] Furthermore, in step S2, a first-order high-pass filter is used to simulate and capture the brightness change between two consecutive frames.

[0017] As described above, in a method for estimating the motion direction of a video target based on a bionic model, in step S3, a biologically reasonable space mechanism is introduced into the thin plate neural layer, and a Gaussian difference algorithm is used to simulate the filtering characteristics, and the obtained visual stimulus is divided into a parallel ON visual pathway and an OFF visual pathway through the Gaussian difference algorithm.

[0018] As described above, in a method for estimating the motion direction of a video target based on a bionic model, in step S4, the fast depolarization and slow repolarization mechanism includes the following steps

[0019] S4.1. Calculate the derivatives of the ON visual pathway and OFF visual pathway signals over time, and obtain the change instead of the continuous derivative by comparing the neural response between every two consecutive signals;

[0020] S4.2, use two different time constants to delay the ON and OFF visual pathways, representing fast depolarization and slow repolarization, respectively;

[0021] S4.3. Subtract the signal processed by the fast depolarization and slow repolarization mechanism from the original signal to filter out visual flicker and irrelevant motion.

[0022] As described above, the method for estimating the motion direction of a video target based on a bionic model, step S5 specifically includes the following sub-steps:

[0023] S5.1. Calculate the delayed ON visual pathway signal as shown below:

[0024]

[0025]

[0026] Among them, DL ON (x, y, t) and They represent the current signal and delayed signal of the ON visual pathway, respectively. (x, y) represents the spatial coordinates, t represents the time coordinates, γ1 represents the time coefficient, τ3 and τ i represents the time constant;

[0027] S5.2, T4 neurons use the current signal and delayed signal of the ON visual pathway to divide the input ON visual pathway signal into four sub-visual pathways: upper, lower, left, and right, as shown in the following formula:

[0028]

[0029] Among them, T4 U (x, y, t), T4 D (x, y, t), T4 L (x, y, t), T4 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T4 neurons, respectively, and sd represents the sampling distance.

[0030] As described above, in the method for estimating the motion direction of a video target based on a bionic model, in step S6, multiplication calculation and local threshold are used to control the response intensity of the ON visual pathway and the OFF visual pathway signal, as shown in the following formula:

[0031]

[0032] Among them, S ON (x, y, t) and S OFF (x, y, t) represent the response strength of the ON visual pathway and the OFF visual pathway signal, ε1 represents the influence coefficient, T s Indicates the threshold value.

[0033] As described above, in the method for estimating the motion direction of a video target based on a bionic model, in step S7, the lobule complex neural layer includes the lobule neural layer and the lobule plate neural layer; step S7 specifically includes the following sub-steps

[0034] S7.1. Calculate the delayed OFF visual pathway signal as shown below:

[0035]

[0036] Among them, DL OFF (x, y, t) and Represents the current and delayed signals of the OFF visual pathway, τ3 and τ i represents the time constant, γ1 represents the time coefficient;

[0037] S7.2, T5 neurons use the current signal and delayed signal of the OFF visual pathway to divide the input ON visual pathway signal into four sub-visual pathways: up, down, left, and right, as shown in the following formula:

[0038]

[0039] Among them, T5 U (x, y, t), T5 D (x, y, t), T5 L (x, y, t), T5 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T5 neurons, respectively, and sd represents the sampling distance;

[0040] S7.3, T4 neurons and T5 neurons are gathered into the upper, lower, left and right sublayers of the tangential cell neurons of the lobula plate, as shown in the following formula:

[0041]

[0042] Among them, LPTC u (t), LPTCD (t), LPTC L (t), LPTC R (t) represents the response intensity of the upper, lower, left, and right sub-visual pathways of the tangential cell neurons of the lobula plate;

[0043] S7.4. Compare the responses of the opposing movements by difference operation to obtain the response strength in the horizontal and vertical directions, as shown in the following formula:

[0044]

[0045] Among them, LPTC H (t) and LPTC V (t) represents the response strength in the horizontal and vertical directions, respectively.

[0046] As described above, in the method for estimating the motion direction of a video target based on a bionic model, in step S8, the ON visual pathway and the OFF visual pathway are compared through a neuron competition mechanism, as shown in the following formula:

[0047]

[0048] Among them, LGMD D (t) represents the response intensity in the depth direction.

[0049] As described above, the method for estimating the motion direction of a video target based on a bionic model, step S9 specifically includes the following sub-steps.

[0050] S9.1. Calculate the visual change response value of the background difference algorithm based on morphological processing, as shown in the following formula:

[0051]

[0052] Among them, S VCR (t) represents the set of visual change response values, BS represents the background difference algorithm based on morphological processing, P(x, y, t) represents the brightness change of the corresponding pixel in each frame, (x, y) represents the spatial coordinate, t represents the time coordinate, R and C represent the height and width of the input visual stimulus respectively;

[0053] S9.2. Calculate the sigmoid membrane potential value corresponding to the visual change response value by the following formula:

[0054]

[0055] Among them, U SMP (t) represents the sigmoid membrane potential value, S VCR(t) represents the set of visual change response values, R and C represent the height and width of the input visual stimulus, respectively, and k represents the influence coefficient;

[0056] S9.3. Determine the depth motion direction sign by comparing the sigmoid membrane potential values, as shown in the following formula:

[0057]

[0058] Among them, F D (t) indicates the depth movement direction sign, T smp Indicates the judgment threshold.

[0059] As described above, the method for estimating the motion direction of a video target based on a bionic model, step S10 specifically includes the following sub-steps.

[0060] S10.1, Sigmoid membrane potential activation function calculation formula is as follows, the output of Sigmoid membrane potential value is adjusted within [-1, 1] by the following formula,

[0061]

[0062] Among them, SMP H (t), SMP V (t), SMP D (t) represents the sigmoid membrane potential value of the horizontal, vertical and depth, k represents the influence coefficient, F D (t) indicates the depth motion direction flag, R and C indicate the height and width of the input visual stimulus, respectively. H (t), LPTC V (t), LPTC D (t) represents the response intensity in the horizontal, vertical and depth directions respectively;

[0063] S10.2. Estimate the target's spatiotemporal motion direction by the following formula:

[0064]

[0065] Among them, DE HM (t),DE VM (t), DF DM (t) represent the estimated horizontal, vertical and depth movement directions respectively, T h Indicates the judgment threshold.

[0066] The beneficial effects of the present invention are:

[0067] In the present invention, lobule plate tangential cell neurons and lobule giant motion detector neurons are used to estimate the spatiotemporal motion direction of targets in complex scenes. Not only can the motion direction of the target in the translation plane be estimated, but also the motion direction of the target in the depth plane can be estimated. It has two advantages at the same time. First, it has stronger robustness in estimating the spatiotemporal motion direction of the target; second, the estimation of the spatiotemporal motion direction of the target in a three-dimensional scene is similar to the estimation of the spatiotemporal motion direction of insects in a natural environment, and is more biologically reasonable. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a signal processing flow chart of the bionic visual neural pathway model in an embodiment of the present invention;

[0069] Figure 2 A network structure diagram of a bionic visual neural pathway model in an embodiment of the present invention;

[0070] Figure 3 Schematic diagram of the structure of the neuron competition mechanism in an embodiment of the present invention;

[0071] Figure 4 Schematic diagram of the first to fourth groups of original frames of the synthetic visual stimulus I in an embodiment of the present invention;

[0072] Figure 5 1 is a first group of comparison result diagrams of synthetic visual stimulation I in an embodiment of the present invention;

[0073] Figure 6 : is a second group of comparison result diagrams of the synthetic visual stimulus I in an embodiment of the present invention;

[0074] Figure 7 1 is a third group of comparison result diagrams of synthetic visual stimulation I in an embodiment of the present invention;

[0075] Figure 8 4 is a fourth group of comparison result diagrams of synthetic visual stimulation I in an embodiment of the present invention;

[0076] Fig. 9 Schematic diagram of the first to fourth groups of original frames of synthetic visual stimulation II in an embodiment of the present invention;

[0077] Fig.10 This is a first group of comparison result diagrams of synthetic visual stimulation II in an embodiment of the present invention;

[0078] Fig.11 This is a second set of comparison result diagrams of synthetic visual stimulation II in an embodiment of the present invention;

[0079] Fig.12 The third group of comparison results of synthetic visual stimulation II in the embodiment of the present invention;

[0080] Fig.13 This is the fourth group of comparison result diagrams of synthetic visual stimulation II in the embodiment of the present invention. DETAILED DESCRIPTION

[0081] This embodiment provides a method for estimating the motion direction of a video target based on a bionic model. Figure 1 As shown, the following steps are included

[0082] S1. Construct a bionic visual neural pathway model, such as Figure 2 As shown, the bionic visual neural pathway model includes the retinal neural layer, the lamina neural layer, the medullary neural layer and the lobule complex neural layer in sequence, and the lobule complex neural layer includes the lobule neural layer and the lobule plate neural layer.

[0083] S2. In the retinal neural layer, a large number of photoreceptors capture visual stimuli. The photoreceptors capture discrete visual stimuli frame by frame. In this embodiment, a first-order high-pass filter is used to simulate and capture the brightness change between two consecutive frames.

[0084] S3. In the lamina neural layer, visual stimulation from the retinal neural layer is received through LMC (Lamina Monoplar Cell), and the visual stimulation is divided into parallel ON visual pathway and OFF visual pathway, and the ON visual pathway and OFF visual pathway are used to encode the increase response and decrease response of brightness respectively.

[0085] The LMC neuron mechanism belongs to the spatial response mechanism. The specific LMC neuron mechanism is as follows: first, in order to have appropriate filtering characteristics, this embodiment places BRS (Biological Rationality Spatial) before the target motion estimation processing stage. This mechanism can maximize the transmission of useful information from complex scenes and improve the visibility of details such as moving edges; and use DoGs (Difference of Gaussians) to simulate filtering characteristics. DoGs is a filtering enhancement algorithm that can enhance the visibility of moving edges and other details, and describes the center-surround confrontation. The center represents the excitatory area and the surrounding represents the inhibitory area; then, in order to enhance the visibility of moving edges and other details, a narrower excitatory signal is used to subtract a wider inhibitory signal; finally, the obtained visual signal is divided into parallel ON and OFF visual pathways.

[0086] S4, the neurons of the lamina monopolar cells filter out visual flicker and irrelevant motion through the FDSR (Fast Depolarising Slow Repolarising) mechanism. The FDSR mechanism belongs to the time response mechanism. In this embodiment, the FDSR mechanism includes the following steps

[0087] S4.1. Calculate the derivatives of the ON visual pathway and the OFF visual pathway signals over time. Since digital signals do not have continuous derivatives, the changes are obtained by comparing the neural responses between every two consecutive signals instead of continuous derivatives.

[0088] S4.2, use two different time constants to delay the ON visual pathway and the OFF visual pathway, which are represented by FD (Fast Depolarising) and SR (Slow Repolarising), respectively;

[0089] S4.3. Subtract the signal processed by the fast depolarization and slow repolarization mechanism from the original signal to filter out visual flicker and irrelevant motion.

[0090] S5. In the medullary neural layer, the T4 neuron separates and processes the ON visual pathway signal, and calculates the upper, lower, left, and right sub-visual pathways of the ON visual pathway through the T4 neuron, which specifically includes the following sub-steps:

[0091] S5.1. Calculate the delayed ON visual pathway signal as shown below:

[0092]

[0093]

[0094] Among them, DL ON (x, y, t) and They represent the current signal and delayed signal of the ON visual pathway, respectively. (x, y) represents the spatial coordinates, t represents the time coordinates, γ1 represents the time coefficient, τ3 and τ i Represents the time constant.

[0095] S5.2, T4 neurons use the current signal and delayed signal of the ON visual pathway to divide the input ON visual pathway signal into four sub-visual pathways: upper, lower, left, and right, as shown in the following formula:

[0096]

[0097] Among them, T4 u (x, y, t), T4 D (x, y, t), T4 L (x, y, t), T4 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T4 neurons, respectively, and sd represents the sampling distance.

[0098] S6. Calculate the response strength of the ON visual pathway and the OFF visual pathway signals respectively. According to the biological research results in the visual nervous system of insects, this embodiment uses multiplication calculation and local threshold to control the response strength of the ON visual pathway and the OFF visual pathway signals, so as to estimate the collision motion of the target when the target is far away. The response strength of the ON visual pathway and the OFF visual pathway signals is shown in the following formula:

[0099]

[0100] Among them, S ON (x, y, t) and S OFF (x, y, t) represent the response strength of the ON visual pathway and the OFF visual pathway signal, ε1 represents the influence coefficient, T s Indicates the threshold value.

[0101] S7. In the lobula complex neural layer, the response intensity of the upper, lower, left and right four sub-visual pathways of the OFF visual pathway is calculated through the T5 neuron; the T4 neurons and T5 neurons are converged to the upper, lower, left and right four sub-layers of the LPTC (Lobula Plate Tangential Cell) neurons, and the horizontal and vertical response intensities are calculated through the response intensity of the upper, lower, left and right four sub-visual pathways of the lobula plate tangential cell neurons.

[0102] In the lobule neural layer, step S7 specifically includes the following sub-steps:

[0103] S7.1. Calculate the delayed OFF visual pathway signal as shown below:

[0104]

[0105] Among them, DL OFF (x, y, t) and Represents the current and delayed signals of the OFF visual pathway, τ3 and τ i represents the time constant, and γ1 represents the time coefficient.

[0106] S7.2, T5 neuron separates and processes the OFF visual pathway signal. T5 neuron uses the current signal and delayed signal of the OFF visual pathway to divide the input ON visual pathway signal into four sub-visual pathways: up, down, left, and right, as shown in the following formula:

[0107]

[0108] Among them, T5 U (x, y, t), T5 D (x, y, t), T5 L(x, y, t), T5 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T5 neurons, respectively, and sd represents the sampling distance.

[0109] S7.3, T4 neurons and T5 neurons are converged into the upper, lower, left, and right sublayers of the tangential cell neurons of the lobula plate. Each sublayer integrates the motion responses of the T4 and T5 sublayers with the same motion properties, as shown in the following formula:

[0110]

[0111] Among them, LPTC U (t), LPTC D (t), LPTC L (t), LPTC R (t) represents the response intensity of the upper, lower, left and right sub-visual pathways of the tangential cell neurons of the lobula plate.

[0112] S7.4. Compare the responses of the opposing movements by difference operation to obtain the response strength in the horizontal and vertical directions, as shown in the following formula:

[0113]

[0114] Among them, LPTC H (t) and LPTC V (t) represents the response strength in the horizontal and vertical directions, respectively.

[0115] S8. The ON visual pathway is compared with the OFF visual pathway through the neuronal competition mechanism to estimate the depth motion and inhibit the translational motion, and obtain the response intensity in the depth direction; the schematic diagram of the neuronal competition mechanism is shown in Figure 3 As shown, the competition mechanism between the ON visual pathway and the OFF visual pathway to judge the depth motion is:

[0116]

[0117] Among them, LGMD D (t) represents the response intensity in the depth direction.

[0118] S9, calculate the VCR (Vision Change Response) value and the corresponding SMP (Sigmoid Membrane Potential) value, specifically including the following steps

[0119] S9.1. In the absence of visual change response and sigmoid membrane potential, LGMD (Lobula Giant Movement Detector) cannot determine whether the target is moving closer or farther away, so the visual change response value of the background difference algorithm based on morphological processing is first calculated, as shown in the following formula:

[0120]

[0121] Among them, S VCR (t) represents the set of visual change response values, BS represents the background difference algorithm based on morphological processing, P(x, y, t) represents the brightness change of the corresponding pixel in each frame, (x, y) represents the spatial coordinate, t represents the time coordinate, R and C represent the height and width of the input visual stimulus respectively.

[0122] S9.2. Calculate the sigmoid membrane potential value corresponding to the visual change response value by the following formula:

[0123]

[0124] Among them, U SMP (t) represents the sigmoid membrane potential value, S VCR (t) represents the set of visual change response values, R and C represent the height and width of the input visual stimulus respectively, and k represents the influence coefficient.

[0125] S9.3. Determine the depth motion direction sign by comparing the sigmoid membrane potential values, as shown in the following formula:

[0126]

[0127] Among them, F D (t) indicates the depth movement direction sign, T smp Indicates the judgment threshold.

[0128] S10, using the Sigmoid membrane potential function to activate the horizontal, vertical and depth responses, and finally estimating the spatiotemporal motion direction through the Sigmoid membrane potential value, specifically including the following sub-steps

[0129] S10.1, Sigmoid membrane potential activation function calculation formula is as follows, the output of Sigmoid membrane potential value is adjusted within [-1, 1] by the following formula,

[0130]

[0131] Among them, SMP H (t), SMP V (t), SMPD (t) represents the sigmoid membrane potential value of the horizontal, vertical and depth, k represents the influence coefficient, F D (t) indicates the depth motion direction flag, R and C indicate the height and width of the input visual stimulus, respectively. H (t), LPTC V (t), LPTC D (t) represents the response intensity in the horizontal, vertical and depth directions respectively.

[0132] S10.2. Estimate the target's spatiotemporal motion direction by the following formula:

[0133]

[0134] Among them, DE HM (t),DE VM (t), DF DM (t) represent the estimated horizontal, vertical and depth movement directions respectively, T h Indicates the judgment threshold.

[0135] The new bionic visual neural pathway model proposed in the present invention is compared with the traditional EMD (Elementary Motion Detectot) calculation model. The original frames of the synthetic visual stimulus I and the synthetic visual stimulus TT are as follows: Figure 4 and Fig. 9 As shown in the figure, the comparison results of synthetic visual stimulus I are shown in Figures 5 to 8 As shown in Figure 2, the comparison results of synthetic visual stimulation II are shown in Figure 2. Figures 10 to 13 shown.

[0136] According to the experimental comparison results, it can be seen that the EMD calculation model cannot estimate the depth motion direction of the target, but can only estimate the planar motion direction of the target. However, the advantage of the insect visual nervous system is that it estimates the spatial motion direction of the target in complex three-dimensional scenes. However, the video target motion direction estimation method based on the bionic model proposed in the present invention can estimate the spatial motion direction of the target in complex three-dimensional scenes, and has stronger robustness and biological rationality.

[0137] In addition to the above embodiments, the present invention may also have other implementation modes. Any technical solution formed by equivalent replacement or equivalent transformation falls within the protection scope required by the present invention.

Claims

1. A method for estimating the motion direction of a video target based on a bionic model, characterized in that: The following steps are involved: S1. Construct a bionic visual neural pathway model, which includes the retinal neural layer, the lamina neural layer, the medullary neural layer and the lobular complex neural layer in sequence; S2, in the retinal neural layer, discrete visual stimuli are captured frame by frame by photoreceptors; S3, in the lamina neural layer, the lamina monopolar cells receive visual stimulation from the retinal neural layer and divide the visual stimulation into parallel ON visual pathway and OFF visual pathway, and encode the increase response and decrease response of brightness through the ON visual pathway and OFF visual pathway respectively; S4, the neurons of the monopolar cells of the thin plate filter out visual flicker and irrelevant motion through a fast depolarization and slow repolarization mechanism; S5, in the medullary neural layer, the upper, lower, left, and right sub-visual pathways of the ON visual pathway are calculated through T4 neurons; S6. Calculate the response strength of the ON visual pathway and the OFF visual pathway signals respectively; use multiplication calculation and local threshold to control the response strength of the ON visual pathway and the OFF visual pathway signals, as shown in the following formula: Among them, S ON (x, y, t) and S OFF (x, y, t) represent the response strength of the ON visual pathway and the OFF visual pathway signal, ε1 represents the influence coefficient, T s Indicates the threshold value; S7, in the lobule complex neural layer, the response strength of the upper, lower, left and right four sub-visual pathways of the OFF visual pathway is calculated through the T5 neuron; the T4 neuron and the T5 neuron are converged to the upper, lower, left and right four sub-layers of the lobule plate tangential cell neurons, and the horizontal and vertical response strengths are calculated through the response strength of the upper, lower, left and right four sub-visual pathways of the lobule plate tangential cell neurons; S8. The ON visual pathway is compared with the OFF visual pathway through a neuronal competition mechanism to estimate the depth motion and inhibit the translational motion, and the response strength in the depth direction is obtained; as shown in the following formula: Among them, LGMD D (t) represents the response intensity in the depth direction; S9, calculating the visual change response value and the corresponding sigmoid membrane potential value, and determining the depth motion direction mark; specifically including the following sub-steps: S9.

1. Calculate the visual change response value of the background difference algorithm based on morphological processing, as shown in the following formula: Among them, S VCR (t) represents the set of visual change response values, BS represents the background difference algorithm based on morphological processing, P(x, y, t) represents the brightness change of the corresponding pixel in each frame, (x, y) represents the spatial coordinate, t represents the time coordinate, R and C represent the height and width of the input visual stimulus respectively; S9.

2. Calculate the sigmoid membrane potential value corresponding to the visual change response value by the following formula: Among them, U SMP (t) represents the sigmoid membrane potential value, S VCR (t) represents the set of visual change response values, R and C represent the height and width of the input visual stimulus, respectively, and k represents the influence coefficient; S9.

3. Determine the depth motion direction sign by comparing the sigmoid membrane potential values, as shown in the following formula: Among them, F D (t) indicates the depth movement direction sign, T smp Indicates the judgment threshold; S10. Use the Sigmoid membrane potential function to activate responses in the horizontal, vertical, and depth directions, and finally estimate the spatiotemporal motion direction through the Sigmoid membrane potential value.

2. The method for estimating the motion direction of a video target based on a bionic model according to claim 1, characterized in that: In step S2, a first-order high-pass filter is used to simulate and capture the brightness change between two consecutive frames.

3. The method for estimating the motion direction of a video target based on a bionic model according to claim 1, characterized in that: In step S3, a biologically plausible spatial mechanism is introduced into the thin plate neural layer, and a Gaussian difference algorithm is used to simulate the filtering characteristics, and the obtained visual stimulation is divided into a parallel ON visual pathway and an OFF visual pathway by the Gaussian difference algorithm.

4. The method for estimating the motion direction of a video target based on a bionic model according to claim 1, characterized in that: In step S4, the fast depolarization slow repolarization mechanism includes the following steps: S4.

1. Calculate the derivatives of the ON visual pathway and OFF visual pathway signals over time, and obtain the change instead of the continuous derivative by comparing the neural response between every two consecutive signals; S4.2, use two different time constants to delay the ON and OFF visual pathways, representing fast depolarization and slow repolarization, respectively; S4.

3. Subtract the signal processed by the fast depolarization and slow repolarization mechanism from the original signal to filter out visual flicker and irrelevant motion.

5. The method for estimating the motion direction of a video target based on a bionic model according to claim 1, characterized in that: The step S5 specifically includes the following sub-steps: S5.

1. Calculate the delayed ON visual pathway signal as shown below: Among them, DL ON (x, y, t) and They represent the current signal and delayed signal of the ON visual pathway, respectively. (x, y) represents the spatial coordinates, t represents the time coordinates, γ1 represents the time coefficient, τ3 and τ i represents the time constant; S5.2, T4 neurons use the current signal and delayed signal of the ON visual pathway to divide the input ON visual pathway signal into four sub-visual pathways: upper, lower, left, and right, as shown in the following formula: Among them, T4 U (x, y, t), T4 D (x, y, t), T4 L (x, y, t), T4 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T4 neurons, respectively, and sd represents the sampling distance.

6. The method for estimating the motion direction of a video target based on a bionic model according to claim 1, characterized in that: In step S7, the lobule complex neural layer includes the lobule neural layer and the lobule plate neural layer; step S7 specifically includes the following sub-steps: S7.

1. Calculate the delayed OFF visual pathway signal as shown below: Among them, DL OFF (x, y, t) and Represents the current and delayed signals of the OFF visual pathway, τ3 and τ i represents the time constant, γ1 represents the time coefficient; S7.2, T5 neurons use the current signal and delayed signal of the OFF visual pathway to divide the input OFF visual pathway signal into four sub-visual pathways: upper, lower, left, and right, as shown in the following formula: Among them, T5 U (x, y, t), T5 D (x, y, t), T5 L (x, y, t), T5 R (x, y, t) represent the upper, lower, left, and right sub-visual pathways separated by T5 neurons, respectively, and sd represents the sampling distance; S7.3, T4 neurons and T5 neurons are gathered into the upper, lower, left and right sublayers of the tangential cell neurons of the lobula plate, as shown in the following formula: Among them, LPTC U (t), LPTC D (t), LPTC L (t), LPTC R (t) represents the response intensity of the upper, lower, left, and right sub-visual pathways of the tangential cell neurons of the lobula plate; S7.

4. Compare the responses of the opposing movements by difference operation to obtain the response strength in the horizontal and vertical directions, as shown in the following formula: Among them, LPTC H (t) and LPTC V (t) represents the response strength in the horizontal and vertical directions, respectively.

7. The method for estimating the motion direction of a video target based on a bionic model according to claim 6, characterized in that: The step S10 specifically includes the following sub-steps: S10.1, Sigmoid membrane potential activation function calculation formula is as follows, the output of Sigmoid membrane potential value is adjusted within [-1, 1] by the following formula, Among them, SMP H (t), SMP V (t), SMP D (t) represents the sigmoid membrane potential value of the horizontal, vertical and depth, k represents the influence coefficient, F D (t) indicates the depth motion direction flag, R and C indicate the height and width of the input visual stimulus, respectively. H (t), LPTC V (t), LGMD D (t) represents the response intensity in the horizontal, vertical and depth directions respectively; S10.

2. Estimate the target's spatiotemporal motion direction by the following formula: Among them, DE HM (t),DE VM (t),DE DM (t) represent the estimated horizontal, vertical and depth movement directions respectively, T h Indicates the judgment threshold.

Images