Target tracking method, device and equipment based on neurodynamic model

Abstract

The invention relates to the technical field of biological information processing, and discloses a target tracking method, device and equipment based on a neurodynamic model, and the method comprises the steps: obtaining the electrical activity data of a target neuron, generating a dynamic differential equation set according to the electrical activity data, carrying out the matrix arrangement of the dynamic differential equation set according to a target size, and obtaining a target neuron; and generating a two-dimensional neural field model, constructing a neurodynamic model based on at least one layer of two-dimensional neural field model, and inputting the to-be-identified data into the neurodynamic model to track a target object meeting a preset speed condition in the to-be-identified data. The problems of target detection and abnormal speed recognition in the field of computer vision are solved according to the biological vision intelligence principle, the parameter space dimension is reduced, training is not needed, the problem that a machine learning model is poor in anti-noise capacity is solved, and interpretability is achieved.

Description

Technical Field

[0001] The present invention relates to the field of biological information processing technology, and in particular to a target tracking method, device and equipment based on a neural dynamics model. Background Art

[0002] By simulating the structural characteristics of biological nervous systems, simplifying and building corresponding mathematical models to reproduce the intelligent behavior of organisms, it is one of the important research methods in the field of artificial intelligence, and has achieved great success in the fields of pattern recognition and machine vision. Taking visual perception as an example, this field is not only the foundation and focus of neuroscience research, but also one of the most important directions of research in the field of artificial intelligence. Convolutional neural networks (CNNs) based on the receptive field model of animal vision and parallel distributed processing, such as AlexNet, GoogLeNet, ResNet, etc., have reached or even exceeded the average level of humans in many fields such as target detection and face recognition. These deep convolutional networks (DCNNs) are not only used in image recognition and other fields, but have also gradually been extended from 2D image data to 2+1D video data processing, and have been widely used in the fields of moving target detection and action recognition. Despite this, DCNNs are still plagued by defects such as large training sets, poor interpretability, and susceptibility to adversarial sample attacks.

[0003] It can be seen that the current construction of artificial intelligence neural networks represented by machine learning and deep learning usually adopts a general model framework. Although these frameworks are inspired by the results of neuroscience research, such as the convolution process, which comes from the functional and structural characteristics of the primary visual cortex processing retinal input information in neuroscience, the subsequent network training and tuning all rely on data-driven objective function optimization methods. By using a large amount of training data, the optimal weights between network connections are mathematically found for the optimization goal, thereby achieving improved model performance. That is, compared with biological intelligence, the model construction and training method based solely on data-driven means that the entire process and results lack interpretability and are vulnerable to adversarial sample attacks. Summary of the invention

[0004] In view of this, the present invention provides a target tracking method, device and equipment based on a neurodynamic model to solve the problems of lack of interpretability and poor noise resistance of the traditional machine learning model training process and results when completing target detection tasks and abnormal speed recognition tasks.

[0005] In a first aspect, the present invention provides a target tracking method based on a neural dynamics model, the method comprising:

[0006] Acquiring electrical activity data of target neurons and generating a group of dynamic differential equations based on the electrical activity data; the group of dynamic differential equations is used to characterize the electrical signal feedback of target neurons under external current;

[0007] The dynamic differential equations are arranged in a matrix according to the target size to generate a two-dimensional neural field model;

[0008] Based on at least one layer of two-dimensional neural field model, a neural dynamics model is constructed; the neural dynamics model is used to perform target detection on data to be identified of target size; the data to be identified includes at least one of a picture to be tested and a video to be tested.

[0009] The data to be identified is input into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified.

[0010] The target tracking method based on the neural dynamic model provided by the embodiment of the present invention is based on the characteristics of the dorsal visual pathway of the biological brain. By acquiring the electrical activity data of a single neuron, a group of dynamic differential equations is generated to construct a two-dimensional neural field model. After cascading operations on multiple layers of the two-dimensional neural field model, a neural dynamic model is formed. The data to be identified is input into the dynamic model to achieve target detection and tracking of objects that meet preset conditions, so that the model framework design and module construction are interpretable. At the same time, through top-down feedback control of the model input layer signal, the noise of abnormal speed characteristics is suppressed, thereby solving the problem of poor noise resistance of the machine learning model.

[0011] In an optional embodiment, the calculation formula for generating the dynamic differential equation group based on the electrical activity data is:

[0012]

[0013] Where U is the instantaneous membrane potential of the target neuron, and τ is the time constant;

[0014] E L is the resting potential;

[0015] I ext is the external input current of a single neuron.

[0016] The target tracking method based on the neural dynamics model provided in the embodiment of the present invention improves the accuracy and dynamics of the model by mathematically characterizing the dynamic characteristics of the bioelectric activities of neurons and calculating the dynamic characteristics of the neuronal electrical activities through a set of differential equations, thereby making the target tracking method based on the neural dynamics model more flexible and predictable.

[0017] In an optional embodiment, the neural dynamics model includes: a visual perception input layer, a motion speed feature recognition layer, and a visual attention layer;

[0018] The visual perception input layer is used to detect the current position of the target object in the video to be tested;

[0019] The motion speed feature recognition layer is used to detect the current speed of the target object in the video to be tested;

[0020] Visual attention layer for identifying objects with abnormal speed.

[0021] The target tracking method based on the neural dynamics model provided in the embodiment of the present invention selects each two-dimensional neural field model according to the requirements of the model. Since the present invention is to achieve target detection and speed detection tasks, it is necessary to extract the position and speed characteristics of the moving target and classify and compare different speed characteristics. Therefore, a V1 neural field simulating the primary visual cortex is built as the visual perception input layer, the MT neural field simulating the MT cortex is built as the motion speed feature recognition layer, and the LIP neural field simulating the LIP decision cortex is built as the visual attention layer to simulate the physiological process of "visual attention focus-evidence accumulation-decision output".

[0022] In an optional implementation, when the data to be identified is a video to be tested, the data to be identified is input into a neural dynamics model to track a target object in the data to be identified that meets a preset speed condition, including:

[0023] For the data corresponding to the target frame in the video to be tested, the visual perception input layer in the neural dynamics model is used to perform spatial convolution processing on the data corresponding to the target frame of the video to be tested, so as to obtain detection feature data corresponding to the target frame; the detection feature data is used to characterize the target detection result corresponding to the target frame;

[0024] The motion speed feature recognition layer in the neural dynamics model performs spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the history frame to obtain motion feature data; the motion feature data is used to characterize the movement of objects in the target frame and the history frame;

[0025] Through the visual attention layer in the neural dynamics model, the motion detection results of each object are processed to obtain the target tracking trajectory to track the target object that meets the preset speed conditions.

[0026] The target tracking method based on the neural dynamics model provided by the embodiment of the present invention performs spatial convolution of the target frame image of the video to be tested with a specific convolution kernel size, and then uses the pixel value after the convolution as the external input current of the motion speed feature recognition layer. At this time, the LIF neuron at the current position of the target in the V1 layer generates a discharge, which is regarded as detecting the current position of the target. Then, the neuron will convolve the V1 output signal within a certain time window with convolution kernels of different sizes to achieve the effect of spatiotemporal filtering, reflecting the spatiotemporal integration characteristics of objects with different motion speeds, and achieving the task of obtaining the target tracking trajectory to track the target object that meets the preset speed conditions. The process has a suppressive effect on abnormal speed characteristics and solves the problem of poor noise resistance of the machine learning model.

[0027] In an optional implementation, the motion speed feature recognition layer in the neural dynamics model performs spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the historical frame to obtain motion feature data, including:

[0028] Acquire a first feature signal extracted by the motion speed feature recognition layer from a target frame of the video to be tested;

[0029] Perform spatiotemporal convolution on the first feature signal and perform normalization processing to obtain an attention weight matrix;

[0030] Acquire a second feature signal extracted by the motion speed feature recognition layer from a frame next to the target frame;

[0031] The second feature signal is weighted based on the attention weight matrix, and the attention weight matrix is ​​updated based on the second feature signal.

[0032] The target tracking method based on the neural dynamics model provided by the embodiment of the present invention performs spatiotemporal convolution on the speed feature signal output by the motion speed feature recognition layer. Since there is a numerical distribution difference between the abnormal speed signal and the normal speed signal, the position of the abnormal signal distribution will be highlighted in the output signal. This embodiment simulates the visual attention layer. After observing the abnormal feature signal, the biological brain will focus the eyeballs to the corresponding position, and only focus on the partial area near the abnormal signal, thereby narrowing the visual perception range and realizing target detection and tracking.

[0033] In an optional embodiment, inputting the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified also includes:

[0034] A decision is made based on the attention weight matrix whether there is an abnormal speed, and the distribution position of the abnormal speed signal converges;

[0035] Based on the distribution position of the abnormal speed signal, locate the actual position of the abnormal target object corresponding to the abnormal speed signal;

[0036] Based on the actual position of the abnormal target object, a regression analysis is performed on each historical position of the abnormal target object to determine the current position of the abnormal target object;

[0037] Output the current position of the abnormal target object.

[0038] The target tracking method based on the neural dynamics model provided by the embodiment of the present invention, after a period of dynamic processing, forms the convergence of the abnormal signal distribution position according to the attention weight matrix. At this time, the visual attention layer makes a decision on the existence of the abnormal speed signal, and locates the actual signal position at the position where the attention converges, and performs regression analysis on the historical motion trajectory of the target point, and finally determines the current position of the abnormal motion target point and outputs it, so as to realize the detection and tracking of the abnormal speed target object.

[0039] In a second aspect, the present invention provides a target tracking device based on a neural dynamics model, the device comprising:

[0040] An acquisition module is used to acquire the electrical activity data of the target neuron and generate a group of dynamic differential equations based on the electrical activity data; the group of dynamic differential equations is used to characterize the electrical signal feedback of the target neuron under the external current;

[0041] A generation module, used to matrix the dynamic differential equations according to the target size to generate a two-dimensional neural field model;

[0042] A construction module is used to construct a neural dynamics model based on at least one layer of a two-dimensional neural field model; the neural dynamics model is used to perform target detection on data to be identified of a target size; the data to be identified includes at least one of a picture to be tested and a video to be tested.

[0043] The tracking module is used to input the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified.

[0044] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the target tracking method based on the neurodynamic model of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0045] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the target tracking method based on a neurodynamic model according to the first aspect or any corresponding embodiment thereof.

[0046] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the target tracking method based on a neurodynamic model according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 is a flowchart of a target tracking method based on a neural dynamics model according to an embodiment of the present invention;

[0049] Figure 2 is a schematic diagram of a framework of a neural dynamics model according to an embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of the spatiotemporal filtering effect of the motion speed recognition layer according to an embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of abnormal feature recognition and dynamic convergence effect of the visual attention layer according to an embodiment of the present invention;

[0052] Figure 5 is a structural block diagram of a target tracking device based on a neural dynamics model according to an embodiment of the present invention;

[0053] Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0055] The current field of artificial intelligence, especially the construction of neural networks in machine learning and deep learning, generally adopts a highly versatile model architecture. Although these architectures are deeply influenced by the results of neuroscience research, such as convolution operations, they directly draw on the structural and functional characteristics of how the primary visual cortex parses retinal incoming information in neuroscience. However, in the subsequent network training and optimization stages, it is highly dependent on data-driven objective function optimization strategies. This strategy uses massive training data to seek the optimal configuration of network connection weights at the mathematical level for specific optimization goals, thereby improving model performance.

[0056] In contrast, this purely data-driven model building and training method, while showing strong capabilities, also exposes two major limitations: first, the model is not interpretable enough, making it difficult to intuitively understand the logic and basis behind its decisions; second, the model is vulnerable to adversarial samples, that is, carefully designed inputs can easily deceive the model and lead to prediction errors. These defects are particularly prominent compared to biological intelligence, which often has higher transparency and robustness in decision-making processes and results.

[0057] The embodiment of the present invention provides a target tracking method based on a neural dynamics model, which solves the problems of target detection and abnormal speed recognition in the field of computer vision through the principle of biological visual intelligence, reduces the dimension of parameter space, and does not require data training, thereby solving the problem of poor noise resistance of machine learning models and having interpretability.

[0058] According to an embodiment of the present invention, an embodiment of a target tracking method based on a neurodynamic model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0059] In this embodiment, a target tracking method based on a neural dynamics model is provided, which can be used in mobile terminals such as mobile phones, tablet computers, etc. Figure 1 is a flow chart of a target tracking method based on a neural dynamics model according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0060] Step S101, acquiring electrical activity data of a target neuron, and generating a group of dynamic differential equations according to the electrical activity data.

[0061] Specifically, a set of dynamic differential equations is used to characterize the electrical signal feedback of the target neuron under external current.

[0062] Furthermore, the target neurons are located in different parts of the brain structure according to their functions, such as the primary visual cortex V1, the middle temporal gyrus cortex MT, the lateral wall of the intraparietal sulcus LIP, etc.

[0063] Furthermore, electrical activity data mainly refers to membrane potential data, which is used to characterize changes in the potential difference between the inside and outside of the target neuron cell membrane.

[0064] Optionally, the electrical activity data also includes information transmitted between synapses of target neurons, such as the excitability or inhibition of the synapses, the delay time of synaptic transmission, etc.

[0065] Step S102, arrange the group of dynamic differential equations in a matrix according to the target size to generate a two-dimensional neural field model.

[0066] Specifically, the target size is determined according to the pixels of the data to be recognized.

[0067] Furthermore, the number of rows and columns of the matrix corresponds one-to-one to the pixels of the data to be identified. Specifically, if the pixels of the data to be identified are 1280*1280, then 1280*1280 target neurons are arranged in a matrix.

[0068] Furthermore, by defining the synaptic connections and arrangement between neurons, a matrix arrangement is performed and spliced ​​together to form a two-dimensional neural field model.

[0069] Optionally, the number of matrix rows and columns may not correspond one-to-one to the pixels of the data to be identified, but up-sampling and down-sampling may be performed according to calculation requirements.

[0070] Step S103, constructing a neural dynamics model based on at least one layer of two-dimensional neural field model.

[0071] Specifically, the neural dynamics model is used to perform target detection on the data to be identified of target size.

[0072] Furthermore, the framework of the neurodynamic model is built according to task requirements.

[0073] Furthermore, the data to be identified includes at least one of a picture to be tested and a video to be tested.

[0074] In some optional embodiments, the neural dynamics model includes: a visual perception input layer, a motion speed feature recognition layer, and a visual attention layer;

[0075] The visual perception input layer is used to detect the current position of the target object in the data to be identified;

[0076] The motion speed feature recognition layer is used to detect the current speed of the target object in the data to be recognized;

[0077] Visual attention layer for identifying objects with abnormal speed.

[0078] According to the target detection and speed detection tasks to be completed, a neural dynamics model is constructed to extract the position and speed features of the moving target and classify and compare different speed features. Therefore, a V1 neural field simulating the primary visual cortex is built as the visual perception input layer, an MT neural field simulating the neural activity pattern of the middle temporal lobe (MT) area is built as the motion speed feature recognition layer, and an LIP neural field simulating the neural activity pattern of the lateral intraparietal area (LIP) is built as the visual attention layer. Figure 2 shown.

[0079] Specifically, the V1 neural field simulating the primary visual cortex is based on the biological perception of the target's position and the neuronal firing generated at the corresponding position of the cortex after the highlight signal in the external input image is transmitted from the retina to the primary visual cortex. Therefore, the neurons of the V1 layer are arranged according to the size of the input image, so that the V1 layer can receive the pixel values ​​of the external image point-to-point. Since the primary visual cortex simulated by the V1 neural field also has the characteristics of spatial filtering for the external input image, the V1 layer performs spatial convolution of a specific convolution kernel size on each frame of the external input video, and then uses the pixel values ​​after convolution as the external input current of the visual attention layer, and inputs and stimulates the LIF neurons of the V1 layer point-to-point. After a specific time is calculated by the differential equation, it is used as the output. At this time, the LIF neurons of the V1 layer at the current position of the target are fired, which is regarded as detecting the current position of the target.

[0080] The MT neural field is used to simulate the MT cortex. The speed detection is based on the spatiotemporal filtering principle of "spatiotemporal scale-speed" mapping. The faster the moving object, the smaller the spatiotemporal accumulation of neuronal firing in the primary visual cortex. Therefore, if the output signal of the V1 layer is subjected to certain temporal convolution and spatial convolution, different speeds will form different integral values ​​in its convolution signal space. To simulate this "spatiotemporal scale-speed" mapping characteristic, a multi-layer and multi-scale LIF matrix is ​​constructed as the MT multi-layer neural field, in which LIF neurons of different scales will convolve the V1 output signal within a certain time window with convolution kernels of different sizes to achieve the effect of spatiotemporal filtering. The spatiotemporal integration characteristics of objects with different moving speeds are reflected in the MT output signal, such as Figure 3 As shown, the faster the object is, the blurrier it appears in the image.

[0081] The LIP neural field is used to simulate the LIP cortex, which is the area of ​​the biological brain that focuses visual attention, accumulates evidence, and makes decisions on motion features. Here, the LIF neuron equation group is used as the basic unit to build the LIP attention neural field with visual attention focusing function and the LIP decision neural field with decision output function to simulate the physiological process of "visual attention focusing - evidence accumulation - decision output".

[0082] Optionally, the neural dynamics model further includes a decision output layer for outputting the position of the target object.

[0083] Step S104, inputting the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified.

[0084] In some optional implementations, when the data to be identified is a video to be tested, the data to be identified is input into the neural dynamics model to track the target object in the data to be identified that meets the preset speed condition, including:

[0085] For the data corresponding to the target frame in the video to be tested, the visual perception input layer in the neural dynamics model is used to perform spatial convolution processing on the data corresponding to the target frame of the video to be tested, so as to obtain detection feature data corresponding to the target frame; the detection feature data is used to characterize the target detection result corresponding to the target frame;

[0086] The motion speed feature recognition layer in the neural dynamics model performs spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the history frame to obtain motion feature data; the motion feature data is used to characterize the movement of objects in the target frame and the history frame;

[0087] Through the visual attention layer in the neural dynamics model, the motion detection results of each object are processed to obtain the target tracking trajectory to track the target object that meets the preset speed conditions.

[0088] After the video to be tested is input into the neural dynamics model, the visual perception input layer performs spatial convolution processing on the data corresponding to the target frame in the video to be tested to obtain the detection feature data corresponding to the target frame. The motion speed feature recognition layer performs spatiotemporal convolution processing on the detection feature data and the detection results corresponding to the historical frames.

[0089] The visual attention layer simulates the visual attention and eye movement functions of the biological brain. After observing abnormal characteristic signals, the biological brain will focus the eyes on the corresponding position and only focus on the part of the area near the abnormal signal, which narrows the visual perception range. Therefore, after the visual attention layer performs spatiotemporal convolution on the speed characteristic signal output by the MT layer, since there is a numerical distribution difference between the signal representing the abnormal speed and the signal representing the normal speed in the MT output signal, please refer to Figure 4 Therefore, the location of the abnormal signal distribution will be highlighted in the output signal of the attention neural field to achieve detection and tracking of target objects that meet the preset speed conditions.

[0090] In some optional implementations, the motion speed feature recognition layer in the neural dynamics model performs spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the historical frame to obtain motion feature data, including:

[0091] Acquire a first feature signal extracted by the motion speed feature recognition layer from a target frame of the video to be tested;

[0092] Perform spatiotemporal convolution on the first feature signal and normalize it to obtain an attention weight matrix;

[0093] Acquire a second feature signal extracted by the motion speed feature recognition layer from a frame next to the target frame;

[0094] The second feature signal is weighted based on the attention weight matrix, and the attention weight matrix is ​​updated based on the second feature signal.

[0095] The feature signal of the target frame after spatiotemporal convolution is spatiotemporally convolved and normalized into an attention weight matrix. When the next frame of the target frame is used as a new input signal of the neural dynamics model, the signals of the V1 and MT layers are weighted to increase the attention on abnormal areas. At the same time, the attention weight matrix is ​​also updated during the processing of the next frame of the target frame.

[0096] In some optional implementations, inputting the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified further includes:

[0097] A decision is made based on the attention weight matrix whether there is an abnormal speed, and the distribution position of the abnormal speed signal converges;

[0098] Based on the distribution position of the abnormal speed signal, locate the actual position of the abnormal target object corresponding to the abnormal speed signal;

[0099] Based on the actual position of the abnormal target object, a regression analysis is performed on each historical position of the abnormal target object to determine the current position of the abnormal target object;

[0100] Output the current position of the abnormal target object.

[0101] The characteristic signal extracted by the neural dynamics model, namely the speed feature, is used as the input of the visual attention neural field, and the speed feature is classified and evidence is accumulated. The existence of abnormal speed signals is detected from the classified features. After a period of dynamic processing, the attention weight matrix will form a convergence of the distribution position of abnormal signals, see Figure 4 At this time, the LIP module will make a decision on the existence of abnormal speed signal, locate the actual signal position at the position where attention converges, and regress the historical motion trajectory of the target point through the abnormal signal position of multiple frames in the convergence state, and finally determine the current position of the abnormal motion target point and output it.

[0102] The target tracking method based on the neurodynamic model provided in this embodiment solves the target detection and abnormal speed recognition problems in the field of computer vision based on the principle of biological visual intelligence. The model parameters are pre-set by physiologically explainable principles, and the task can be completed without training. Specifically, the model imitates the V1 and MT layers of the spatiotemporal filtering characteristics of the primary visual cortex of the biological brain to extract feature signals from the input signals of the real data to be identified, which has a noise reduction effect, and also has a suppressive effect on the noise of abnormal speed features, solving the problem of poor noise resistance of machine learning models. The visual attention and decision-making modules built by imitating the decision cortex of the biological brain can dynamically locate abnormal features based on the time process of evidence accumulation in the biological brain, and perform top-down feedback control on the model input layer signals. This brain-like algorithm can realize an artificial neural network with the same functions as the existing machine learning algorithm, and its framework design and module construction are explainable. Compared with the existing machine learning and deep network models, the dimension of its parameter space is greatly reduced, and parameter tuning can be explained according to biological principles. The cost of parameter space search is also greatly reduced.

[0103] In some optional embodiments, the calculation formula for generating the dynamic differential equations according to the electrical activity data is:

[0104]

[0105] Among them, τ is the time constant; U is the instantaneous membrane potential of the target neuron, E L is the resting potential; I extThe external input current of a single neuron is used to characterize the dynamic characteristics of the neuron's bioelectric activity through a mathematical model. The dynamic characteristics of the neuron's electrical activity are calculated through a set of differential equations, which improves the accuracy and dynamics of the model and makes the target tracking method based on the neural dynamics model more flexible and predictable.

[0106] In this embodiment, a target tracking device based on a neurodynamic model is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" may be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0107] This embodiment provides a target tracking device based on a neural dynamics model. Figure 5 As shown, including:

[0108] An acquisition module 501 is used to acquire electrical activity data of a target neuron and generate a group of dynamic differential equations according to the electrical activity data; the group of dynamic differential equations is used to characterize the electrical signal feedback of the target neuron under an external current;

[0109] A generation module 502, for arranging the dynamic differential equations in a matrix according to the target size to generate a two-dimensional neural field model;

[0110] The construction module 503 is used to construct a neural dynamics model based on at least one layer of a two-dimensional neural field model; the neural dynamics model is used to perform target detection on the data to be identified of the target size; the data to be identified includes at least one of a picture to be tested and a video to be tested.

[0111] The tracking module 504 is used to input the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified.

[0112] In some optional implementations, the acquisition module 501 includes:

[0113] The computational submodule is used to generate a set of dynamic differential equations based on the electrical activity.

[0114] In some optional implementations, the tracking module 504 includes:

[0115] The spatial convolution submodule is used to perform spatial convolution processing on the data corresponding to the target frame in the video to be tested through the visual perception input layer in the neural dynamics model to obtain the detection feature data corresponding to the target frame; the detection feature data is used to characterize the target detection result corresponding to the target frame.

[0116] The temporal convolution submodule is used to perform spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the historical frame through the motion speed feature recognition layer in the neural dynamics model to obtain movement feature data; the movement feature data is used to characterize the movement of objects in the target frame and the historical frame.

[0117] The processing result submodule is used to process the motion detection results of each object through the visual attention layer in the neural dynamics model to obtain the target tracking trajectory to track the target object that meets the preset speed conditions.

[0118] In some optional implementations, the tracking module 504 includes:

[0119] The first feature signal submodule is used to obtain the first feature signal extracted by the motion speed feature recognition layer from the target frame of the video to be tested.

[0120] The convolution submodule is used to perform spatiotemporal convolution on the first feature signal and perform normalization processing to obtain an attention weight matrix.

[0121] The second feature submodule is used to obtain a second feature signal extracted by the motion speed feature recognition layer from the next frame of the target frame.

[0122] An updating submodule is used to perform weighted processing on the second feature signal based on the attention weight matrix, and to update the attention weight matrix based on the second feature signal.

[0123] In some optional implementations, the tracking module 504 further includes:

[0124] The decision submodule is used to make a decision on whether there is an abnormal speed based on the attention weight matrix and obtain the distribution position convergence of the abnormal speed signal.

[0125] The position determination submodule is used to locate the actual position of the abnormal target object corresponding to the abnormal speed signal based on the distribution position of the abnormal speed signal.

[0126] The regression analysis submodule is used to perform regression analysis on each historical position of the abnormal target object based on the actual position of the abnormal target object to determine the current position of the abnormal target object.

[0127] The output submodule is used to output the current position of the abnormal target object.

[0128] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0129] The target tracking device based on the neurodynamic model in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0130] The embodiment of the present invention also provides a computer device having the above Figure 5 The target tracking device based on the neural dynamics model is shown.

[0131] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0132] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0133] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0134] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0135] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0136] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0137] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0138] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0139] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0140] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A target tracking method based on a neural dynamics model, characterized in that: The method comprises: Acquiring electrical activity data of a target neuron, and generating a group of dynamic differential equations according to the electrical activity data; the group of dynamic differential equations is used to characterize the electrical signal feedback of the target neuron under an external current; Arranging the group of dynamic differential equations in a matrix according to the target size to generate a two-dimensional neural field model; Based on at least one layer of two-dimensional neural field model, a neural dynamics model is constructed; the neural dynamics model is used to perform target detection on data to be identified of target size; the data to be identified includes at least one of a picture to be tested and a video to be tested; The data to be identified is input into a neural dynamics model to track a target object that meets a preset speed condition in the data to be identified.

2. The method according to claim 1, characterized in that The calculation formula for generating the dynamic differential equation group according to the electrical activity data is: Wherein, U is the instantaneous membrane potential of the target neuron, and τ is the time constant; E L is the resting potential; I ext is the external input current of the target neuron.

3. The method according to claim 2, characterized in that The neural dynamics model includes: a visual perception input layer, a motion speed feature recognition layer, and a visual attention layer; The visual perception input layer is used to detect the current position of the target object in the data to be identified; The motion speed feature recognition layer is used to detect the current speed of the target object in the data to be recognized; The visual attention layer is used to identify the target object with abnormal speed.

4. The method according to claim 3, characterized in that When the data to be identified is a video to be tested, the data to be identified is input into a neural dynamics model to track a target object in the data to be identified that meets a preset speed condition, including: For the data corresponding to the target frame in the video to be tested, spatial convolution processing is performed on the data corresponding to the target frame of the video to be tested through the visual perception input layer in the neural dynamics model to obtain detection feature data corresponding to the target frame; the detection feature data is used to characterize the target detection result corresponding to the target frame; The motion speed feature recognition layer in the neural dynamics model performs spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the historical frame to obtain motion feature data; the motion feature data is used to characterize the movement of objects in the target frame and the historical frame; The motion detection results of each object are processed through the visual attention layer in the neural dynamics model to obtain a target tracking trajectory to track the target object that meets the preset speed condition.

5. The method according to claim 4, characterized in that The step of performing spatiotemporal convolution processing on the detection results corresponding to the target frame and the detection results corresponding to the historical frame through the motion speed feature recognition layer in the neural dynamics model to obtain the motion feature data includes: Acquire a first feature signal extracted by the motion speed feature recognition layer from the target frame of the video to be tested; Performing spatiotemporal convolution on the first feature signal and normalizing it to obtain an attention weight matrix; Acquire a second feature signal extracted by the motion speed feature recognition layer from a frame next to the target frame; The second feature signal is weighted based on the attention weight matrix, and the attention weight matrix is ​​updated based on the second feature signal.

6. The method according to claim 5, characterized in that The step of inputting the data to be identified into a neural dynamics model to track a target object in the data to be identified that meets a preset speed condition also includes: Making a decision on whether there is an abnormal speed based on the attention weight matrix, and obtaining a convergence of the distribution position of the abnormal speed signal; Based on the distribution position of the abnormal speed signal, locating the actual position of the abnormal target object corresponding to the abnormal speed signal; Based on the actual position of the abnormal target object, performing regression analysis on each historical position of the abnormal target object to determine the current position of the abnormal target object; The current position of the abnormal target object is output.

7. A target tracking device based on a neurodynamic model, characterized in that: The device comprises: An acquisition module, used to acquire electrical activity data of a target neuron and generate a group of dynamic differential equations according to the electrical activity data; the group of dynamic differential equations is used to characterize the electrical signal feedback of the target neuron under an external current; A generation module, used for arranging the group of dynamic differential equations in a matrix according to a target size to generate a two-dimensional neural field model; A construction module, for constructing a neural dynamics model based on at least one layer of a two-dimensional neural field model; the neural dynamics model is used to perform target detection on data to be identified of a target size; the data to be identified includes at least one of a picture to be tested and a video to be tested; The tracking module is used to input the data to be identified into the neural dynamics model to track the target object that meets the preset speed condition in the data to be identified.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the method according to any one of claims 1 to 6.

Images