Binocular vision depth calculation system and brain-like calculation model

Abstract

By means of constructing neuron-like models, the present application achieves neural information encoding, inter-neuron connection self-organization, and neural calculation and output of depth estimation encoded information. A system and models provided in the present application comprise: constructing two paths of visual receiver models; constructing a visual sequence model, which, under small-amplitude, reciprocating, irregular movements of the visual receiver models, outputs visual sequence information; constructing a neural network model, which receives information input from the visual receiver models, wherein the inputted information drives a basic neural model to operate. Under visual stimulation, light-receiving units of the visual receiver models connect to the neural network in a self-organized manner, and color information and edge information of an object are separated and outputted; stimulation at a boundary line causes a general sequence model to operate, wherein all points at the boundary line are sequentially excited to form an edge sequence combined signal, the signal at this time comprising edge color sequence information and edge direction sequence information; under the action of two paths of edge sequences, a same-sequence seeking model operates, outputs an original sequence, and identifies depth information.

Description

A binocular vision depth calculation system and brain-like computing model Technical Field

[0001] This invention belongs to the field of cognitive neuroscience and has applications in passive visual depth estimation and brain-inspired computing. More specifically, it proposes a neuronal binocular vision depth calculation system and a related brain-inspired computing model. Background Art

[0002] Computer stereo vision depth estimation is currently an important method for passive visual depth estimation. In practical applications, binocular stereo matching algorithms are key to depth estimation. This invention provides a brain-inspired computing stereo matching method and a binocular vision depth calculation system based on the bioelectric characteristics of neurons. Unlike the storage and computation separation algorithms of general computer systems, this invention designs and constructs a brain-inspired computing model based on the basic functions of neurons. This model implements functions such as neuronal information encoding, interneuronal connection self-organization, and neuronal computation output of coded information.

[0003] Summary of the Invention

[0004] The present invention proposes a binocular stereo matching method for stereo vision based on known neuron characteristics and a related brain-inspired computing model, providing a new method and system for binocular stereo matching algorithms in stereo vision. The specific method and system include the following steps:

[0005] In step S1, the visual receptor model receives external light stimulation, forms physical object boundaries, and generates color and direction information of the object boundaries. The binocular vision depth calculation system first constructs two visual receptor models, which input perceived visual information into the neural network model.

[0006] In step S2, the visual receiver model performs small, random back-and-forth motions, and the visual sequence model generates visual sequence pulses. The two visual receiver models of the binocular visual depth calculation system continuously perform small, random back-and-forth motions to acquire visual information from the objective world. This further constructs the visual sequence model, and the light receiving units of the visual receiver models are connected to the visual sequence model. The visual sequence model then sequentially outputs visual information, generating visual sequence pulses.

[0007] Step S3 realizes the separation of visual color information and edge information. The binocular vision depth calculation system further constructs a basic neuron model (perception model). The two-way visual receiver model inputs pulses to the neural network, and the stimulation causes the basic neuron model (perception model) to form and run. During this period, the stimulation sequence received by the basic neuron model (perception model) in the patch area is stable, and is inhibited by the model inhibitory neurons for a certain period of time, and the signal excitation is in an inhibited state; the stimulation sequence received by the basic neuron model (perception model) at the edge or boundary constantly changes to form continuous excitation. In this way, a temporal distinction is formed, and the color mode and the edge mode are separated.

[0008] In step S4, the color and orientation information of the object's edge is processed and extracted, integrated to form local edge neural pulse sequence information. The binocular vision depth calculation system further constructs a general sequence model. The stimulation received by the basic neuron model (perception model) at the edge or boundary triggers the operation of the general sequence model. Each point at the edge or boundary is sequentially stimulated through the general sequence model to form an edge sequence combination pulse. This pulse contains both the edge color sequence and edge orientation sequence information.

[0009] Step S5, the same-order matching detection of the binocular object edge information is performed to output stereo information. The binocular visual depth calculation system further constructs a sequence similarity model. The overall system receives input from two visual receiver models, and both channels form corresponding edge sequences. The same two-channel sequences have the same sequence but are not synchronized due to independent processing. Under the action of two edge sequences with the same sequence but not synchronized, the sequence similarity model runs, forms the excitation of a new pattern, and outputs the original sequence. Under the action of two edge sequences with different sequences, the sequence similarity model has no output. Under the action of two edge sequences with approximately the same sequence, the sequence similarity model outputs a short sequence. The output original sequence represents the foreground depth disparity, and the output short sequence represents the background disparity. DETAILED DESCRIPTION

[0010] In order to make the technical solution of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings.

[0011] Furthermore, the visual receptor model described in step S1 of FIG1 receives external light stimulation, forms a physical object boundary, and forms object boundary color and direction information; this needs to be achieved through the visual receptor model shown in FIG2 and FIG3 , and the characteristics of the visual receptor model are:

[0012] The binocular visual depth calculation system of the present invention utilizes a pair of visual receptor models that synchronously perform irregular, small-amplitude reciprocating motions. The pair of visual receptor models simultaneously outputs a sequence of color information about the objective world's colors and boundaries. The individual structures of this model are shown in Figures 2 and 3:

[0013] Figure 2 is a side view of the visual receptor model. The visual receptor model is a special hemispherical structure with light receiving units evenly arranged on the inner surface of the structure, as shown by the number 1 in Figure 2. The light receiving units in the figure receive external visible light and electromagnetic wave stimuli and convert them into electrical signals.

[0014] The visual receptor model performs irregular small-amplitude reciprocating motion, causing the background in the visual receptor model to constantly change. The color difference between the foreground and background forms a logical change boundary, which the present invention calls the object boundary. As shown in Figure 3:

[0015] Figure 3 is a top view of the visual receptor model. In Figure 3, 4 denotes the curve representing the image of an external straight line on the visual receptor model. In Figure 3, 2 denotes the center of the visual receptor model. In Figure 3, 3 denotes the evenly distributed light receiving units of the visual receptor model.

[0016] When the visual receptor model performs random small-amplitude reciprocating motion synchronously, the light receiving units at each point on the boundary formed by the foreground object and the background are continuously and intermittently excited in sequence, forming a color change sequence of the boundary.

[0017] At the same time, the straight line imaged by the boundary of an objective object appears as a curve on the visual receptor model. When the visual receptor model undergoes simultaneous, irregular, small-amplitude back-and-forth motion, the color change rates of the light receiving units at each point on the curve imaged on the visual receptor model differ. Specifically, the closer the boundary point is to the center of the visual receptor, the closer its motion speed is to the actual boundary speed; the farther the boundary point is from the center of the visual receptor, the greater the difference in its speed from the actual boundary speed. Therefore, the differences in the color change rates of the boundary formed on the visual receptor model contain directional information about the boundary of the objective object. Furthermore, the back-and-forth motion imparts motion components to each boundary point in both the radial and latitudinal directions, causing each boundary point to be excited sequentially.

[0018] Furthermore, the "generation of visual sequence pulses by the visual sequence model under the small-amplitude reciprocating irregular motion of the visual receptor model" described in step S2 of Figure 1 requires processing by the visual sequence model shown in Figure 4. The color stimulus generated by the visual receptor model is processed by the visual sequence model composed of the horizontal neuron model to form a neuronal color information pulse sequence.

[0019] The horizontal inhibitory neuron model has at least two pathways for receiving input signals from the visual receptor light receiving unit. It also has a horizontal convergence unit, with all signal pathways connected to the horizontal convergence unit. When the horizontal convergence unit receives a pulsed electrical signal from any input pathway, it suppresses the electrical signals from all other input pathways. After a certain interval, the horizontal inhibitory neuron suppresses the current input signal, allowing it to receive signals from other pathways. This reciprocating process allows each input signal to be fed into the horizontal inhibitory neuron in turn.

[0020] The visual sequence model consists of a horizontal inhibitory neuron linked to multiple local light-receiving units in the visual receptor model. The visual sequence model receives external visual signals, processes them, and generates output signals from multiple light-receiving units, which are then released in sequence, forming a logical sequence. The specific model operation process is shown in Figure 4:

[0021] Figure 4 Visual Sequence Model: The symbol "Shu" in the figure represents the horizontal inhibitory neuron convergence unit, the symbol A in Figure 4 represents the visual receiver model, the symbol B in the figure represents the light receiving unit of the visual receiver model, and the symbols 1, 2, 3, 4, and 5 in the figure represent the link paths sent by the light receiving unit of the visual receiver model to the horizontal inhibitory neuron convergence unit. When the horizontal inhibitory neuron convergence unit receives an excitation signal from any of the paths 1, 2, 3, 4, and 5, it sends an inhibitory signal to the remaining paths through the original path. The inhibitory signal has an inhibitory effect on all paths except the path that first receives the excitation signal. After the current excited path stops being excited, the remaining paths randomly become excited and inhibit other paths. This logically forms a sequence output of the electrical signal of the light receiving unit.

[0022] Furthermore, as described in step S3 of FIG1 , “separating the color of visual information from the edge information” requires the coordinated processing of the neuron model, the neural network model, and the basic neuron model shown in FIG5 .

[0023] The "neuron model" is divided into excitatory neuron model and inhibitory neuron model; its characteristics are:

[0024] The excitatory neuron model receives excitatory input and outputs excitatory signals. The excitatory neuron model has at least three input signal pathways, of which at least two are excitatory pathways and at least one is an inhibitory pathway. The excitatory neuron model has at least two output signal pathways, both of which are excitatory pathways. The pulse signal carried by the input pathway is either in a high-frequency state or the connected target neuron convergence unit is in a continuously high-excitation state, which improves the efficiency of the input pathway and maintains it for a certain period of time. The convergence unit of the excitatory neuron model transmits excitatory pulse signals to all output pathways.

[0025] The excitatory neuron model features a convergence unit. This unit aggregates signal levels and delays output pulses. The convergence unit receives pulses from all input pathways, aggregates them to a certain level, and then sends pulses to all output pathways. When the convergence unit receives signals from high-efficiency input pathways, the delay is reduced or even eliminated, allowing pulses to be sent quickly to the output pathways.

[0026] The inhibitory neuron model receives excitatory inputs and outputs inhibitory signals. The inhibitory neuron model has at least two input signal pathways, of which at least one is an excitatory pathway and one is an inhibitory pathway. The inhibitory neuron model has at least one output signal pathway, both of which are inhibitory pathways. When the pulse signal carried by the input pathway or the excitation action is in a high-frequency state, or the connected target neuron convergence unit is in a continuously high-excitation state, the efficiency of the input pathway is improved and maintained for a certain period of time. The output pathway of the inhibitory neuron model can sense the electrical level of the connected neuron convergence unit, and the inhibitory efficiency of the inhibitory output pathway is positively linearly related to the electrical level of the connected neuron convergence unit.

[0027] The inhibitory neuron model features a convergence unit. This unit aggregates signal levels and delays output pulses. It receives pulses from all input pathways and, upon reaching a certain level, transmits pulses to all output pathways. When the convergence unit receives signals from high-efficiency input pathways, the delay is reduced or even eliminated, allowing pulses to be rapidly transmitted to the input and output pathways. The inhibitory neuron model's convergence unit transmits inhibitory pulses to all output pathways.

[0028] When the aggregation unit of the excitatory and inhibitory neuron model receives the inhibitory pulse signal from the input pathway, the inhibitory pulse signal is aggregated in the aggregation unit, resulting in a reduced possibility of the aggregation unit outputting a pulse signal.

[0029] The excitatory and inhibitory neuron model aggregation units do not output pulse signals when there is no external signal input.

[0030] When the converging unit of the excitatory and inhibitory neuron model receives excitatory signals from the input pathway, the excitatory signals converge at the converging unit, which increases the probability of the converging unit outputting signals or outputting signals through the output pathway.

[0031] Furthermore, when the input path is an initial path, there is a time delay between the input signal and the corresponding signal generated by the convergence unit; when the input path is an efficient path, the time delay between the input signal and the corresponding signal generated by the convergence unit is shortened or even disappears.

[0032] When the convergence unit of the excitatory and inhibitory neuron model receives excitatory signals from the input pathway, the cumulative effect of long-sequence signals and signals from high-efficiency pathways is stronger, making it easier to stimulate the output pathway to output signals.

[0033] The convergence unit of the excitatory and inhibitory neuron model can simultaneously receive two or more excitatory signals from the input pathway. If the excitatory signals from multiple inputs are synchronized, each input has the same effect as any single excitatory input, and the pulse signals generated by the convergence unit are not summed up or amplified. If the excitatory signals from multiple inputs are asynchronous, the signals generated by each input at the convergence unit are amplified.

[0034] Furthermore, the aforementioned “neural network model” has the following characteristics:

[0035] A neural network model consists of excitatory and inhibitory neuron models connected through input and output pathways. Each excitatory neuron model is linked to at least one surrounding excitatory and one inhibitory neuron model. Each inhibitory neuron model has at least one output link to an excitatory neuron model and at least one input link to an excitatory neuron model. In a neural network model, all neuron models have the potential to connect to other neurons.

[0036] Furthermore, under the influence of external stimuli, the neural network model autonomously forms a basic neuron model (perception model), a general sequence model, a sequence similarity model, an effector perception model, and a logical perception model at the local mesoscopic level.

[0037] The basic neuron model is also called the perception model, and the present invention calls it the basic neuron model (perception model). It is characterized in that the basic neuron model (perception model) is composed of an excitatory source neuron, an excitatory target neuron and an inhibitory neuron in the local part of the neural network model, which are linked to each other through the initial state. When the basic neuron model (perception model) receives an external stimulation signal, it forms a mutually connected loop. Under the action of the loop, the signal realizes the high frequency of the signal carried by the pathway between the source neuron and the target neuron, and the continuous accumulation of the target neuron signal, thereby forming an enhanced transmission efficiency; at the same time, the model realizes signal separation and output under the action of the inhibitory loop. The specific model operation process is shown in Figure 5:

[0038] In Figure 5, the basic neuron model (perception model): Sink 1 represents the sink unit of excitatory source neurons, Sink 2 represents the sink unit of excitatory target neurons, and Sink 3 represents the sink unit of inhibitory neurons. Arrows represent pathways between neurons, and 1, 2, 3, 4, 5, and 6 are the pathway numbers. Pathways 1, 2, 3, 4, and 6 transmit excitatory signals, while pathway 5 transmits inhibitory signals.

[0039] Under the condition that the neuron models of the neural network model are interconnected, an excitatory source neuron, an excitatory target neuron, and an inhibitory neuron in the local neural network model are linked together through the initial state to form a basic neuron model (perception model). The combined link form is shown in Figure 5. At this time, the basic neuron model (perception model) is in the initial state of the first stage. Except for pathway 1, all other pathways are in the initial effect stage. There is no signal transmission in each pathway, and the pathway transmission efficiency does not change.

[0040] Furthermore, the basic neuron model (perception model) enters the second stage of the basic neuron model (perception model) preliminary excitation state. The input pulse signal of path 1 directly passes through the convergence unit 1 of the source neuron, and is accumulated in the convergence unit 2 of the target neuron through path 2. At the same time, the input pulse signal of path 1 directly passes through the convergence unit 1, and is accumulated in the convergence unit 3 through path 3.

[0041] Furthermore, the basic neuron model (perception model) enters the third stage of the basic neuron model (perception model) cycle formation state; after the 2nd channel is accumulated to a certain extent in the convergence unit 2, it generates a pulse signal in the 4th and 6th channels. After the 3rd channel is accumulated to a certain extent in the convergence unit 3, it generates an inhibitory pulse signal in the 5th channel. The inhibitory pulse signal generated by the 5th channel suppresses the excitatory pulse signals of the 4th and 6th channels. Due to the accumulation effect of the convergence unit 3, the inhibitory pulse signal generated by the 5th channel is delayed after the excitatory signal of the 2nd channel generated by the convergence unit 1. At the beginning of the cycle, the inhibitory effect of the 5th channel is smaller than the excitatory effect of the 2nd channel.

[0042] Furthermore, the basic neuron model (perception model) enters the fourth stage of the basic neuron model (perception model) feedback high-frequency state. The efficiency of pathway 4 is improved and under the joint action of the pulse signal and the input pulse signal of pathway 1, the pulse signal frequency of pathways 2 and 3 is accelerated, and the cumulative effect of convergence units 2 and 3 is accelerated. The high-frequency signal of pathway 2 causes the convergence accumulation effect of convergence unit 2 to accelerate, causing the signal frequency of pathway 4 to accelerate. Pathways 2 and 4 form a positive feedback loop pathway, and the signal frequency of the positive feedback loop pathway continues to increase. At the same time, the transmission signal frequency of pathway 3 is accelerated. Due to the accumulation of convergence unit 2, the signal of pathway 4 is delayed behind the signals of pathways 2 and 3.

[0043] Furthermore, the basic neuron model (perception model) enters the fifth stage, the basic neuron model (perception model) pathway gain state. A positive feedback loop is formed between pathways 2 and 4, and the overall signal frequency of the loop continues to increase. Under the influence of high signal frequencies, pathways 2 and 3 continue to increase the cumulative effect of convergence units 2 and 3, gradually forming a highly efficient excitation pathway. The accumulated delay of signal transmission in convergence units 2 and 3 on this pathway is shortened or even nearly eliminated.

[0044] At the same time, the basic neuron model (perception model) enters a state of pathway inhibition enhancement. The inhibitory pathway of pathway 5 has an electronic channel with a synchronous potential effect. The stimulation level of pathway 2 causes the potential level of convergence unit 2 to increase, and under the effect of potential synchronization, the inhibitory effect of pathway 5 gradually increases; but there is a time delay difference between the improvement of the efficiency of pathway 5 and pathway 2. Therefore, in the stage of rising frequency of the overall circulatory system signal, the inhibitory effect of pathway 5 is less than the excitatory stimulation effect of pathway 2; after the overall circulatory system signal frequency reaches a high frequency state and stabilizes for a certain period of time, the inhibitory effect of pathway 5 and the excitatory effect of pathway 2 can form an excitation-inhibition balance in convergence unit 2;

[0045] Furthermore, the basic neuron model (perception model) enters the sixth stage of the basic neuron model (perception model) model formation state. After reaching a balance between the inhibitory effect of pathway 5 in the high-frequency state and the excitation inhibition of pathway 2, the inhibitory effect of pathway 5 quickly suppresses the signal transmission of the circulatory system; pathways 2, 3, and 4 reach an efficient communication state, and the model enters a stable state.

[0046] Furthermore, the basic neuron model (perception model) enters the seventh stage of the basic neuron model (perception model) operating state. After the model establishes efficient transmission states for pathways 2 and 3, the inhibitory effect of pathway 5 subsequently enters a highly efficient state. However, compared to the excitation of pathway 2, the inhibition of pathway 5 still has a delay in the initial stages of the overall system signal transmission. Under the stimulation of pathway 1, the system transmits the original sequence carried by pathway 1. Later in the continuous transmission process, the signal of pathway 1 cannot be transmitted through pathways 4 and 6.

[0047] The basic neuron model (perception model) serves as the foundation for establishing efficient connections between source and target neurons in other brain-inspired computing models. This basic model encompasses different excitation conditions for convergence unit 2. These excitation conditions include, but are not limited to: 1. Excitation by other efficient neuronal pathways produces a transient high-level potential, placing convergence unit 2 in an easily excitable state. At this point, stimulation from other common link pathways can cause excitation. 2. After a common link pathway is connected to convergence unit 2, long or repeated sequences of stimulation along that pathway can cause excitation of convergence unit 2 due to the cumulative effect. 3. Two common link pathways are connected to convergence unit 2, acting simultaneously on convergence unit 2, but the two stimulation sequences are asynchronous. Alternatively, the two stimulation sequences are nearly asynchronous, resulting in excitation of convergence unit 2. 4. Sustained stimulation of other neurons with which convergence unit 2 has established an efficient pathway creates an easily excitable state, leading to excitation of convergence unit 2. 5. Convergence unit 2 is in an easily excitable state during high-frequency excitation in a signal-looping state.

[0048] Furthermore, as described in step S3 of Figure 1, "separating visual color information from edge information" involves the neural network model receiving electrical signals from the visual receptor model, which causes the basic neuron model (perception model) to form and operate. During this period, the basic neuron model (perception model) within the contiguous area receives a stable stimulus sequence, which is suppressed by inhibitory neurons for a certain period of time, remaining in an inhibited state. The basic neuron model (perception model) at the edge or boundary receives a continuously changing stimulus sequence, resulting in excitation. This creates a temporal distinction, separating the color pattern from the edge pattern.

[0049] Furthermore, step S4 in Figure 1, "processing and extracting color and orientation information from object edges and integrating it to form local boundary neural spike train information," requires the collaborative processing of the neuron model, the neural network model, and the neuronal mesoscopic model. The neuron model and the neural network model are the same as those described in step S3; the neuronal mesoscopic model uses the general sequence model and the sequence identity model to process information.

[0050] The general sequence model is characterized by being composed of multiple sequentially excited excitatory neurons linked in sequence, which output sequence signals after processing. The specific model operation process is shown in Figure 6:

[0051] In Figure 6, the symbol "Hub 1" represents the first excited excitatory neuron's convergence unit, the symbol "Hub 2" represents the second excited excitatory neuron's convergence unit, and the symbol "Hub 3" represents the third excited excitatory neuron's convergence unit. Arrows connect the pathways between neurons, and symbols 1, 2, 3, 4, and 5 are the pathway numbers. 1, 2, 3, 4, and 5 are all excitatory pathways transmitting excitatory signals, and symbols A, B, and C represent the sequence of stimuli received by each neuron. The figure only lists three neurons that are excited in sequence, but in actual situations, more neuron combinations can be inferred.

[0052] In the first stage of the general sequential model, the first neuron activates, receiving a stimulus sequence A and emitting a pulse signal A through pathway 1. Meanwhile, the converging unit is in the ready-to-fire state. In the second stage, the second neuron activates, connecting to the first neuron converging unit 1, which is currently firing or has fired, through the initial state pathway 2. Because neuron converging unit 1 is in the ready-to-fire state, neuron 1 fires again. Continued sequential stimulation of pathway 2 forms feedback pathway 3, creating a feedback loop between pathways 2 and 3. In the third stage, pathways 2 and 3 in the general sequential model circulate high-frequency stimulation, making pathway 2 more efficient. The mechanism for efficient pathway transmission in this model is similar to that of the basic neuron model (perception model), and the excitation and inhibition cycles are not described in detail. Each neuron link in Figure 6 has a corresponding inhibitory link pathway. Its operation is similar to that of the inhibitory pathway in the basic neuron model, but is not depicted in this figure. During the next round of A, B, and C sequence excitation, pathway 1 transmits the A+B sequence.

[0053] In the fourth stage of the general sequence model, the third neuron activates, connecting to the second activated neuron, Unit 2, through Pathway 4. Because Unit 2 is at a high potential, Neuron 2 fires again, and Pathway 4 continues the sequential stimulation, forming Feedback Pathway 5. Pathways 4 and 5 form a feedback loop. In the fifth stage of the general sequence model, Pathways 4 and 5 cycle through high-frequency stimulation, making Pathway 4 more efficient. When the next round of A, B, C excitation sequence occurs, Pathway 1 transmits the A+B+C sequence. Similarly, more neurons firing sequentially through the general sequence model can produce corresponding stimulation sequences.

[0054] The sequence similarity model is characterized by: it consists of two excitatory neuron sequence chains, which output the original sequence signal after processing. The specific model operation process is shown in Figure 7:

[0055] In Figure 7, the symbol "Sink 1" represents the first excitatory neuron aggregation unit of the first general sequence model's excitatory neuron sequence chain, the symbol "Sink 2" represents the second excitatory neuron aggregation unit of the first general sequence model's excitatory neuron sequence chain, and the symbol "Sink 3" represents the third excitatory neuron aggregation unit of the first general sequence model's excitatory neuron sequence chain. Arrows represent pathways between neurons, and the symbols 1, 2, 3, 4, 5, and 6 are the pathway numbers. Pathways 1, 2, 3, 4, 5, and 6 are all excitatory pathways that transmit excitatory signals, and the symbols A, B, and C represent the sequence stimuli received by each neuron. Pathways 3 and 5 are high-efficiency pathways, and neuron aggregation units 1, 2, and 3 have formed the general sequence model's excitatory neuron sequence chain through pathways 3 and 5. The symbol "Sink 7" represents the sequence-similarity model's consensus neuron aggregation unit, and the symbol "h" represents the sequence-similarity model's output pathway.

[0056] In Figure 7 , the symbol "4" represents the first excitatory neuron converging unit of the excitatory neuron sequence chain of the second general sequential model; the symbol "5" represents the second excitatory neuron converging unit of the excitatory neuron sequence chain of the second general sequential model; and the symbol "6" represents the third excitatory neuron converging unit of the excitatory neuron sequence chain of the second general sequential model. Arrows represent pathways between neurons, and the symbols "a, b, c, d, e, and f" are the pathway numbers. Each of these pathways transmits excitatory signals, and the symbols "A, B, and C" represent the sequence of input stimuli received by each neuron. Paths a and c are efficient pathways, and neuron converging units 4, 5, and 6 have formed the excitatory neuron sequence chain of the general sequential model through pathways a and c. Each neuron link in Figure 7 has a corresponding inhibitory pathway. Its operation is similar to that of the inhibitory pathway in the basic neuron model, but is not depicted or described in this figure. The mechanism for efficient pathway transmission in this model is the same as that in the basic neuron model (perception model), and the cycles of excitation and inhibition are not described in detail.

[0057] In the first stage, the sequence identity model is in its initial state. Convergence units 1, 2, and 3 form the output sequence A+B+C. Convergence units 4, 5, and 6 are stimulated by the same stimulus and also form the output sequence A+B+C. The two sequences are identical, but due to the differences in the converging units, the sequences are out of sync when they reach converging unit 7.

[0058] In the second stage, the sequence identity model converges its output states. When the two sequences act on convergence unit 7, they form a simultaneous detection effect and generate sequence accumulation. Loops are formed in channels 1 and 6 and channels e and g, respectively. Under high-frequency stimulation of their respective feedback loops, the channel effect gains 1 and e, respectively, enter a state of efficient transmission. However, loops 1 and 6, and loops e and f, respectively, reach the efficient state sequentially. Constrained by the loop that reached the efficient state first and the inhibitory pathway, the loop that reached the efficient state later is inhibited, resulting in no signal output at convergence unit 7. This results in the signal output from channel h being completely consistent with the sequence carried by the loop that reached the efficient state first.

[0059] Furthermore, as described in step S4 of Figure 1, "the color information and direction information of the edge of the object are processed and extracted, and integrated to form local edge neural pulse sequence information" means that the stimulus sequence received by the basic neuron model (perception model) at the edge or boundary continuously stimulates the operation of the general sequence model, and each point at the edge or boundary is stimulated in turn through the general sequence model to form an edge sequence combination pulse. The pulse at this time contains the color sequence and edge direction sequence information of the edge. In addition, at the same time as the fourth step in the above step S4, the sequence formed by the basic neuron model (perception model) in the patch area also stimulates the sequence similarity model and the general sequence model to run, forming a unified color information sequence in the region. At this time, all color sequences in the region are output as original color sequences under the integration of the general sequence model and the sequence similarity model. This is the color vision mode output.

[0060] Furthermore, the "same-order matching detection of binocular object edge information to output stereo information" in step S5 of Figure 1 needs to be processed by a sequence similarity model. The sequence similarity model of this step is consistent with the sequence similarity model of the previous step.

[0061] Furthermore, the system receives input from two visual receiver models, each of which generates corresponding edge sequences. The two identical sequences are identical but out of sync due to independent processing. The sequence similarity model operates under the influence of two identical but out of sync edge sequences, generating new pattern excitations and outputting the original sequence. Under the influence of two edge sequences with different sequences, the sequence similarity model has no output. Under the influence of two edge sequences with approximately the same sequence, the sequence similarity model outputs a short sequence. The original sequence output represents the foreground depth disparity, and the short sequence output represents the background disparity.

[0062] The above principle of neuron collaboration can also form a model for learning how neural networks use effectors and logical concepts. The following is the corresponding working principle of the model.

[0063] The effector perception model is characterized by the formation of implicit neural action patterns or memories through the interaction of external effectors (such as the wrist joint formed by muscles and bones) and internal neural networks, that is, the establishment of efficient connections between sensory neurons and action planning neurons. External effectors form a loop connection with sensory neurons and action planning neurons, and the pulse signal or excitation signal generated under stimulus drive is transmitted at high frequency within the loop. The neuronal connection between sensory neurons and action planning neurons is made more efficient under the action of high-frequency pulse signals or excitation signals. The specific model formation and calculation process is shown in Figure 8:

[0064] In Figure 8, Sink 1 represents the sensory neuron aggregation unit of the effector perception model, Sink 2 represents the action planning neuron aggregation unit, and Sink 3 represents the inhibitory neuron aggregation unit. A represents the external effector action portion. B represents the external effector perception portion. The arrowed line from A to B represents the driving connection between external effectors. The remaining arrowed lines represent pathways between neurons. 1, 2, 3, 4, 5, and 6 are the pathway numbers. Pathways 1, 2, 3, 4, and 5 transmit excitatory signals, while pathway 6 transmits inhibitory signals. Pathways 1, 2, and 3 are established, efficient pathways, while pathways 4 and 5 are initial state pathways. Each neuronal link in Figure 8 has a corresponding inhibitory link pathway. With the exception of the inhibitory bypass between Sink 1 and Sink 2, the accompanying inhibitory bypasses of the excitatory pathways are not depicted in this figure.

[0065] In the first stage of the effector-perception model, the action-planning neuron (Sink 2 in the figure) is stimulated and activates Pathway 2. Pathway 2 drives the action component A of the external effector, which in turn modifies the perception component B. The perception component B of the effector outputs a stimulus, which triggers the firing of sensory neurons in the neural network. Sink 2 remains in an excitable state due to the continuous stimulation of Pathway 1.

[0066] In the second stage of the effector-perception model, sensory neurons in the neural network excite and emit signals, which activate the high-potential action-planning neurons (convergence 2 in Figure 8). This loop of sensory neurons, action-planning neurons, and effectors forms, and the signals within the loop self-excite, generating high-frequency signals. This makes the pathway between sensory neurons and action-planning neurons more efficient. Simultaneously, as inhibitory neurons bypass the loop, the signaling gradually reaches equilibrium, eventually reaching a resting state.

[0067] The logical perception model is characterized by the interaction between external logic and the internal neural network, forming implicit logical neuronal connections. External logic operations form a loop connection with the neural network. Signals generated under stimulus drive are transmitted at high frequency within the loop. Simultaneously, logical causality is mapped within the neural network to form separate neuronal patterns, which also form high-frequency signal connections. The mapping of logical causality within the neural network to form separate interneuronal connections achieves high efficiency under the action of high-frequency signals. The specific model formation and operation process is shown in Figure 9:

[0068] In Figure 9, sink 1 represents the neural network's aggregation of neurons that map the "result" of the external logic model; sink 2 represents the neural network's aggregation of neurons that map the "cause" of the external logic model; sink 3 represents the aggregation of neurons that map the "consequence" of the external logic model; sink 4 represents the aggregation of neurons that map the "intention" of the external logic model; and sinks 5 and 6 represent the aggregation of inhibitory neurons. A represents the cause portion of the external logic, and B represents the result portion of the logic. The arrowed line from A to B represents the logic of the external logic, and the remaining arrowed lines represent pathways between neurons. Paths 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, and 11 are the pathway numbers. Paths 1, 2, 3, 4, 5, 6, 7, 8, and 10 are excitatory pathways that transmit excitatory signals, while pathways 9 and 11 are inhibitory pathways that transmit inhibitory signals. Paths 2, 4, 5, 6, and 7 are established, efficient pathways, and pathways 1, 3, 8, and 10 are initial state pathways. In Figure 9, each segment of neuronal connections has a corresponding inhibitory link pathway. Except for the inhibitory bypass between sink 2 and sink 1 and the inhibitory bypass between sink 3 and sink 4, the accompanying inhibitory bypasses of other excitatory pathways are not drawn and described in this figure.

[0069] In the first stage of the logical perception model, the cause A and result B of the external logical model are mapped in the neural network to form separate neuron patterns, that is, mapping connections with neuron aggregation unit 1 and aggregation unit 2 are formed in the neural network.

[0070] In the second stage of the logical perception model, the intentional model neuron convergence unit 4 drives the cause side of the external logical model, which in turn drives the effect side. The effect side then activates the consequence mode of the internal neural network, thereby activating neuron convergence unit 3. Under the continuous stimulation of the intentional stimulation pathway of Pathway 7, convergence unit 4 is in a readily excitable state. Convergence unit 3 activates convergence unit 4, which is also in a readily excitable state, forming an excitation loop. This logical application loop sends high-frequency signals, resulting in a highly efficient pathway in Pathway 1.

[0071] In the third stage of the logical perception model, the logic application loop sends high-frequency signals, causing cause A and result B to repeatedly stimulate corresponding convergence points 2 and 1. This generates a high-frequency signal from convergence point 2 to convergence point 1. Ultimately, pathway 3 between convergence point 2 and convergence point 1 forms an efficient pathway, forming the internal logic model of the neural network.

[0072] The aforementioned visual receptor model, horizontal inhibitory neuron model, visual sequence model, basic neuron model (perception model), general sequence model, sequence consensus model, effector perception model, and logical perception model are all fundamental models for brain-inspired computing, playing a role not only in this example but also in various brain-inspired computing functions. Functions implemented through the collaboration and combination of these models in other brain-inspired and neural computing applications are also encompassed within the scope of this invention.

[0073] The technical solutions of the present invention have been described in conjunction with the embodiments illustrated in the accompanying drawings. However, those skilled in the art will readily appreciate that the scope of protection of the neuron model of the present invention is clearly not limited to the specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0074] Although the basic neuron model of the present invention is described through specific embodiments, those skilled in the art should understand that various changes and equivalent substitutions can be made to the present invention without departing from the scope of the present invention. The models and method steps described in the embodiments disclosed herein are not limited to those that can be implemented by electronic hardware, computer software, or a combination of the two. Various modifications to the present invention in various implementations do not depart from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed, but should include all implementation methods that fall within the scope of the claims of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In the drawings of the specification:

[0076] FIG1 is a flow chart of a binocular vision depth calculation system and a brain-like computing model according to the present invention;

[0077] FIG2 is a side view of a visual receiver model structure of a binocular vision depth calculation system and a brain-like computing model of the present invention;

[0078] FIG3 is a top view of a visual receiver model structure of a binocular visual depth calculation system and a brain-like computing model of the present invention;

[0079] FIG4 is a schematic diagram of the structure and construction steps of a visual sequence model of a binocular vision depth calculation system and a brain-inspired computing model of the present invention;

[0080] FIG5 is a schematic diagram of the structure and construction steps of a basic neuron model (perception model) of a binocular vision depth calculation system and a brain-like computing model of the present invention;

[0081] FIG6 is a schematic diagram of a general sequence model structure and construction steps of a binocular vision depth calculation system and a brain-inspired computing model of the present invention;

[0082] FIG7 is a schematic diagram of the structure and construction steps of a sequence similarity model of a binocular vision depth calculation system and a brain-inspired computing model of the present invention;

[0083] FIG8 is a schematic diagram of the structure and construction steps of an effector perception model of a binocular vision depth calculation system and a brain-inspired computing model of the present invention;

[0084] FIG9 is a schematic diagram of the structure and construction steps of a logical perception model of a binocular vision depth calculation system and a brain-like computing model of the present invention.

Claims

1. A binocular vision depth calculation system and a brain-like computing model are characterized by: The system and construction method include the following steps: Step S1, the visual receptor model receives external light stimulation, forms a physical object boundary, and forms object boundary color and direction information; Step S2, the visual receiver model generates visual sequence pulses by the visual sequence model under the condition of small reciprocating irregular motion; Step S3, separation of visual color information and edge information; Step S4, processing, extracting and integrating the color information and direction information of the edge of the object to form local edge neural pulse sequence information; Step S5: performing in-order matching detection on binocular object edge information to output stereo information.

2. A binocular vision depth calculation system and brain-like computing model according to claim 1, characterized in that: The step S1 comprises the following steps: A visual receiver model is constructed, and the visual receiver model is characterized in that the visual receiver model is a special hemispherical light receiver structure, and light receiving units are evenly arranged on the inner surface of the structure to convert light signals into electrical signals. The binocular visual depth calculation system adopts a pair of visual receiver models. A pair of visual receiver models simultaneously outputs sequence information of objective world colors and boundaries. The color difference between foreground and background stimuli forms a logical change boundary, which is called the object boundary in the present invention. The pair of visual receiver models used synchronously perform irregular small reciprocating motions, causing the backgrounds in the two visual receiver models to change continuously. When a pair of visual receiver models synchronously perform irregular small reciprocating motions, the light receivers of the receivers at each point on the boundary formed by the foreground object and the background are continuously intermittently excited in sequence to form the change sequence information of the boundary. At the same time, the objective world straight line image is a curve on the visual receiver model. When the visual receiver model synchronously performs irregular small reciprocating motions, the color change rates of each point of the curve imaged on the visual receiver model are different. The different rates of color change formed by the boundary on the visual receiver model contain the directional information of the boundary of the objective world object.

3. A binocular vision depth calculation system and brain-like computing model according to claim 1, characterized in that: The step S2 comprises the following steps: A horizontal inhibitory neuron model and a visual sequence model are constructed, characterized by: After the horizontal convergence unit of the horizontal inhibitory neuron model receives the pulse electrical signal from any input path, it will inhibit the electrical signals of other input paths. After a certain period of time, the horizontal inhibitory neuron inhibits the current input signal, and the horizontal inhibitory neuron can receive signals from other paths, thus forming a cycle of input electrical signals from various paths. The visual sequence model is characterized by: it is composed of a horizontal inhibitory neuron and multiple visual input links. The visual sequence model receives external visual signals and outputs visual sequence signals after processing.

4. The binocular vision depth calculation system and brain-like computing model according to claim 1, characterized in that: The step S3 comprises the following steps: Construct an excitatory neuron model, characterized in that the excitatory neuron model receives excitatory input and outputs an excitatory signal. Construct an inhibitory neuron model, characterized in that the inhibitory neuron model receives excitatory input and outputs an inhibitory signal. Construct a neural network model, characterized in that the neural network model is composed of an excitatory neuron model and an inhibitory neuron model interconnected through an input path and an output path. In the neural network model, all neurons have the possibility of being connected to other neurons. Construct a basic neuron model (perception model), which is also called a local mesoscopic basic model in the present invention, characterized in that the model is composed of an excitatory source neuron, an excitatory target neuron and an inhibitory neuron interconnected. When the basic neuron model (perception model) receives an external stimulus signal, a mutually connected loop is formed. Under the action of the loop, the signal realizes the high frequency of the signal carried by the path between the source neuron and the target neuron, and the target neuron signal continues to accumulate, thereby forming an enhanced transmission efficiency; at the same time, the model realizes signal separation and output under the action of the inhibitory loop.

5. The binocular vision depth calculation system and brain-like computing model according to claim 1, characterized in that: The step S4 comprises the following steps: A general sequence model is constructed, characterized in that the general sequence model is composed of a plurality of sequentially excited excitatory neurons connected in sequence, and a sequence signal is output after processing. Further, the basic neuron model (perception model) at the edge or boundary receives stimulation and continuously stimulates the operation of the general sequence model. Each point at the edge or boundary sequentially stimulates the general sequence model to form an edge sequence combination pulse, and the pulse at this time contains the color sequence of the edge and the edge direction sequence information. In addition, at the same time as the above step S4, the basic neuron model (perception model) in the patch area receives stimulation and also stimulates the sequence similarity model and the general sequence model to run, forming a unified color information sequence in the region. At this time, under the integration of the general sequence model and the sequence similarity model, all color sequences in the region are integrated into the original color sequence output, which is the color vision mode output.

6. A binocular vision depth calculation system and brain-like computing model according to claim 1, characterized in that : The step S5 comprises the following steps: A sequence similarity model is constructed, which is characterized by: the sequence similarity model is composed of two general sequence models composed of two excitatory neuron sequence chains connected together, and the original sequence signal is output after processing. The same-order matching detection of binocular object edge information and the output of stereo information need to be processed by the sequence similarity model. Furthermore, the system receives inputs from two visual receiver models, and both of them form corresponding edge sequences. The same two sequences have the same sequence but are not synchronized due to independent processing. Under the action of two edge sequences of the same order but not synchronized, the sequence similarity model runs, forms the excitation of a new pattern, and outputs the original sequence. Under the action of two edge sequences of different orders, the sequence similarity model has no output. Under the action of two edge sequences of approximately the same order, the sequence similarity model outputs a short sequence. The output original sequence represents the foreground depth disparity, and the output short sequence represents the background disparity.

7. This patent proposes an effector perception model, which is characterized by: Through the interaction between external effectors (such as the wrist joint formed by muscles and bones) and the internal neural network, implicit neural action patterns or memories are formed, that is, efficient connections are established between sensory neurons and action planning neurons. External effectors form a loop connection with sensory neurons and action planning neurons, and the signals formed under stimulus drive are transmitted at high frequency within the loop. The neuronal connection between sensory neurons and action planning neurons is made efficient under the action of high-frequency signals.

8. This patent proposes a logical perception model, which is characterized by: Through the interaction between external logic and internal neural network, implicit logical connection is formed. External logic operation forms a loop connection with the neural network. The signal formed under the stimulus drive is transmitted at high frequency in the loop. At the same time, the logical cause and effect is mapped in the neural network to form a separate neuron pattern, and a high-frequency signal connection is also formed in between. The logical cause and effect is mapped in the neural network to form a separate neuron connection, which is highly efficient under the action of high-frequency signals.

9. A binocular vision depth calculation system and a brain-like computing model, characterized in that: A pair of visual receptors are used to input stimulation to the neural network under small irregular reciprocating motion, and a disparity signal is output through collaborative processing including a visual receptor model, a horizontal inhibitory neuron model, a visual sequence model, a basic neuron model (perception model), a general sequence model, and a sequence similarity model; The construction method described in the patent includes a visual receptor model, a horizontal inhibitory neuron model, a visual sequence model, a basic neuron model (perception model), a general sequence model, a sequence similarity model, an effector perception model, and a logical perception model. These models are basic models of brain-like computing, and they play a role not only in this example, but also in various functions of brain-like computing. The functions achieved through the coordination and combination of the above models in other brain-like computing and neural computing should be included in the protection scope of the present invention.

Images