Method, device, equipment and medium for simulating human vision to realize target positioning

By simulating the dorsal and ventral visual pathways of human eye vision, the feature information of the target image is extracted and fused, and the problem of difficulty in feature extraction in complex scenarios is solved based on reinforcement learning method, and the accuracy and speed of target positioning are improved.

CN118135018BActive Publication Date: 2024-12-06XIAN TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410408693.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-07
Publication Date
2024-12-06
Estimated Expiration
2044-04-07

AI Technical Summary

Technical Problem

The target positioning method based on reinforcement learning is difficult to extract feature in complex scenarios, resulting in poor recognition performance and unable to meet the needs of target recognition.

Method used

By simulating the dorsal and ventral visual pathways of human eye vision, feature information of the target image is extracted, and the distance between the center point of the feature information and the center point of the human eye vision is reduced based on the geometric relationship of these pathways, so that it overlaps , thereby achieving target positioning.

Benefits of technology

It improves the accuracy and speed of target positioning, reduces the number of searches, avoids the problem of difficulty in feature extraction in complex scenarios, and can extract more comprehensive and rich feature information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118135018B_ABST
    Figure CN118135018B_ABST
Patent Text Reader

Abstract

The embodiments of the present application relate to the field of target positioning technology, and disclose a method, device, equipment and medium for simulating human vision to achieve target positioning. The above-mentioned method for simulating human vision to achieve target positioning includes: obtaining a target image to be identified; inputting the target image into a pre-trained target positioning model, simulating the dorsal visual pathway of human vision through the target positioning model, extracting feature information of the target image, and simulating the ventral visual pathway of human vision, obtaining the geometric relationship between the center point of the feature information and the center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reducing the distance between the center point of the feature information and the center point of human vision, so that the center point of the feature information coincides with the center point of human vision, and obtaining the target positioning result of the target image, which can solve the technical problem of low target positioning accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of target positioning technology, and in particular to a method, device, equipment and medium for simulating human vision to achieve target positioning. Background Art

[0002] Target localization refers to determining whether a given target exists in an image and its specific location. Target localization requires the following conditions to be met: (1) the target is within the field of view; (2) there is a visual path between the human eye and the target; (3) the feature information contained in the image is sufficient for target recognition. Therefore, target localization requires the construction of a geometric equation that contains the relative position relationship between the target point and the human eye's visual point, and this equation needs to have an optimal unique solution under the existing measurement conditions.

[0003] The target localization method based on reinforcement learning can speed up the target search and localization process to a certain extent, and the target localization method mainly uses the policy gradient algorithm (such as the REINFORCE algorithm) and the value function algorithm (such as the DQN algorithm) of reinforcement learning for learning. Compared with the target recognition method of convolutional neural network, the above reinforcement learning-based algorithm can find the target in less than 10 steps while reducing the computational cost.

[0004] However, the target localization method based on reinforcement learning has difficulty in feature extraction in complex scenes, and the extracted local information cannot meet the needs of target recognition, so there is a problem of poor recognition performance. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method, device, equipment and medium for simulating human vision to achieve target positioning, which can improve the accuracy of target positioning.

[0006] To solve the above technical problems, an embodiment of the present application provides a method for simulating human vision to achieve target positioning, comprising the following steps:

[0007] Acquire a target image to be identified;

[0008] Inputting the target image into a pre-trained target localization model;

[0009] The target positioning model simulates the dorsal visual pathway of human vision to extract feature information of the target image, and simulates the ventral visual pathway of human vision to obtain the geometric relationship between the center point of the feature information and the center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reduces the distance between the center point of the feature information and the center point of human vision, so that the center point of the feature information coincides with the center point of human vision, and obtains the target positioning result of the target image.

[0010] In some optional embodiments, before the ventral visual pathway simulating human vision obtains the geometric relationship between the center point of the feature information and the center point of human vision, the method further includes:

[0011] Weak stimulation signals in the characteristic information of the target image are eliminated.

[0012] In some optional embodiments, the feature information of the target image includes spatial features and semantic features of the target image, and obtaining the geometric relationship between the center point of the feature information and the center point of human vision through the ventral visual pathway simulating human vision includes:

[0013] fusing the spatial features and semantic features of the target image;

[0014] The geometric relationship between the center point of the fused feature information and the center point of human vision is obtained.

[0015] In some optional embodiments, the reducing the distance between the center point of the feature information and the center point of the human vision based on the geometric relationship between the center point of the feature information and the center point of the human vision so that the center point of the feature information coincides with the center point of the human vision includes:

[0016] Based on the geometric relationship between the center point of the feature information and the center point of the human vision, the center point of the human vision is moved in several different directions until the center point of the feature information coincides with the center point of the human vision.

[0017] In some optional embodiments, the moving direction of the center point of the human eye vision is obtained by the following steps:

[0018] The pre-built coordinate axis is divided into 8 vectors with each movement of 45°, and the direction of each vector is used to represent the moving direction of the center point of human vision.

[0019] In some optional embodiments, the method further includes:

[0020] After extracting the feature information of the target image, the geometric relationship between the center point of the feature information, the center point of human vision and the boundary point of the feature information is obtained;

[0021] The method of reducing the distance between the center point of the feature information and the center point of the human vision based on the geometric relationship between the center point of the feature information and the center point of the human vision so that the center point of the feature information coincides with the center point of the human vision includes:

[0022] Based on the geometric relationship between the center point of the feature information, the center point of human vision and the boundary points of the feature information, the distance between the center point of the feature information and the center point of human vision is reduced so that the center point of the feature information coincides with the center point of human vision.

[0023] In some optional embodiments, the target positioning model simulates the dorsal visual pathway of human vision through a plurality of series-connected feature extraction layers to extract feature information of the target image;

[0024] Each feature extraction layer is used to extract hierarchical feature information of the target image and pass the hierarchical feature information to the next feature extraction layer connected thereto.

[0025] In some optional embodiments,

[0026] The embodiment of the present application also provides a device for simulating human vision to achieve target positioning, including:

[0027] An image acquisition module, used to acquire a target image to be identified;

[0028] An image input module, used for inputting the target image into a pre-trained target positioning model;

[0029] The image positioning module is used to simulate the dorsal visual pathway of human vision through the target positioning model to extract feature information of the target image, and simulate the ventral visual pathway of human vision to obtain the geometric relationship between the center point of the feature information and the center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reduce the distance between the center point of the feature information and the center point of human vision so that the center point of the feature information coincides with the center point of human vision, thereby obtaining the target positioning result of the target image.

[0030] An embodiment of the present application also provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method of simulating human eye vision to achieve target positioning.

[0031] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method of simulating human eye vision to achieve target positioning.

[0032] The method for simulating human vision to realize target positioning provided by the embodiment of the present application has at least the following features:

[0033] Beneficial effects:

[0034] The present application simulates the process of human vision to achieve target positioning, and the human vision pathway of the human eye when performing target positioning includes the ventral visual pathway and the dorsal visual pathway. By simulating the dorsal visual pathway of the human eye, the feature information of the target image can be extracted. By simulating the ventral visual pathway of the human eye, the geometric relationship between the center point of the feature information and the center point of the human eye vision can be obtained. Finally, based on the geometric relationship between the center point of the feature information and the center point of the human eye vision, the distance between the center point of the feature information and the center point of the human eye vision is reduced, so that the center point of the feature information and the center point of the human eye vision coincide, and the target positioning result of the target image can be obtained. Compared with the traditional search mode, the human eye search and positioning method does not have the limitation of the difficulty of extracting complex scene features, and can extract more comprehensive and rich feature information of the target image, thereby improving the accuracy of target positioning; and the present application starts the search and positioning from the center point, rather than searching aimlessly, which greatly reduces the number of target searches, thereby improving the speed of target positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] One or more embodiments are exemplarily described by the pictures in the corresponding drawings, and these exemplary descriptions do not constitute limitations on the embodiments.

[0036] Figure 1 is a schematic diagram of a head-eye nerve pathway provided according to an embodiment of the present application;

[0037] Figure 2 It is a flow chart of a method for simulating human vision to realize target positioning according to an embodiment of the present application;

[0038] Figure 3 is a schematic diagram of a human eye visual pathway provided according to an embodiment of the present application;

[0039] Figure 4 is a schematic diagram of a recurrent neural network provided according to an embodiment of the present application;

[0040] Figure 5 is a schematic diagram of a target positioning model provided according to an embodiment of the present application;

[0041] Figure 6 is a schematic diagram of a device for simulating human vision to achieve target positioning according to an embodiment of the present application;

[0042] Figure 7 It is a structural diagram of a computer device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0044] The human eye's search method is simpler and more efficient than the search method based on reinforcement learning. The research on human eye physiology is constantly improving. Therefore, when referring to the structure of the human eye's visual pathway, it is also necessary to refer to the latest research on human eye physiology. Limited by the range of eye movement, the line of sight shift in target tracking often relies on the coordination of the head and eyes, and the head rotation and eye movement are completed by different neural pathways. The existing "two streams" hypothesis provides a good explanation for this research: when the human brain receives knowledge, it can be divided into "what stream" and "where stream", that is, memory can be spatial and non-spatial. The dorsal visual stream (DVS) preferentially transmits spatial sensory information to the postrhinal cortex (POR), and the ventral visual stream mainly transmits the identity / feature information of the stimulus to the PRC (perirhinal cortex). Based on anatomical evidence, spatial information is preferentially relayed from the POR to the medial entorhinal cortex (MEC), and non-spatial information is transmitted from the PRC to the lateral entorhinal cortex (LEC). MEC and LEC are the two main subregions of the entorhinal cortex. MEC processes "where", that is, spatial information, and LEC processes "what", that is, non-spatial information. These two pathways finally converge in the hippocampus, combining spatial and non-spatial information to form the representation of episodic memory. The CA1 to CA3 regions in the hippocampus also play a key role in processing spatial and non-spatial information. This pathway not only involves the human brain's processing of vision, but also involves the brain's decision-making behavior. According to the above research, Figure 1 , which summarizes the neural structures and connection buses in the brain that are related to positioning and decision-making.

[0045] This embodiment is based on the latest dual-stream hypothesis theory, simulates the ventral visual flow and dorsal visual flow pathways of the human eye, constructs the latest head-eye neural pathway network model, and then combines the sequential decision-making strategy of reinforcement learning to achieve target positioning.

[0046] An embodiment of the present application relates to a method for simulating human eye vision to achieve target positioning. The implementation details of the method for simulating human eye vision to achieve target positioning in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0047] The specific process of the method of simulating human eye vision to realize target positioning in this embodiment can be as follows: Figure 2 As shown, including:

[0048] Step 201: Acquire a target image to be identified.

[0049] Specifically, the target image to be identified is an image that needs to be positioned, and the image includes the target that needs to be positioned.

[0050] Step 202: input the target image into a pre-trained target localization model.

[0051] Step 203, simulating the dorsal visual pathway of human vision through the target positioning model to extract the feature information of the target image, and simulating the ventral visual pathway of human vision to obtain the geometric relationship between the center point of the feature information and the center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reducing the distance between the center point of the feature information and the center point of human vision, so that the center point of the feature information coincides with the center point of human vision, and obtaining the target positioning result of the target image.

[0052] Specifically, when a person's head remains still and observes an object, eye movement behaviors include two types: saccades and fixations. Therefore, the process of locating a target by the human eye can be seen as two steps: first, scanning the overall picture, and second, processing the details of the visual image. However, both saccades and fixations are inseparable from the foveal visual field of the eyeball. Saccades can be seen as a rapid eye movement that moves the foveal visual field from one point to another, while fixations are the process of keeping the foveal visual field on the target for a certain period of time to obtain sufficient visual image details. The real visual system has a strong foveal bias, and a more realistic treatment of uneven receptive field density may improve the model's fit to neural data. And the reason why the center position is keen on visual perception may also be related to color and light perception.

[0053] like Figure 3As shown in the figure, in the process of visual processing, the neuronal activities of almost all brain structures have a regulatory effect, and the most important of these are the ventral visual pathway and the dorsal visual pathway. The ventral visual pathway is from the primary visual cortex V1 through the extrastriate cortex (V2-V4) to the inferotemporal cortex; the dorsal pathway is responsible for the processing of spatial information in the medial temporal lobe, the medial superior temporal lobe, and the parietal cortex. V1 can essentially be regarded as the earliest area in the visual cortex to process visual information. It receives the original input of the retina and memory information, and performs feature extraction. V2 is located after V1. It receives information from V1, but it participates in more advanced feature extraction and combination, such as boundaries, directions, etc. V4 is located after V2 and is also an important area in the visual cortex. V4 plays a key role in visual cognition and object recognition. It has a high sensitivity to the details of movement and participates in high-level cognitive processes such as image attention, memory, and decision-making.

[0054] Inspired by the above-mentioned human eye search mode, a target positioning model including a feature extraction module, a foveal positioning module and a human eye movement decision module is established in this embodiment. Among them, the feature extraction module is specifically used to simulate the dorsal visual pathway of human eye vision to extract the feature information of the target image; the foveal positioning module is used to simulate the ventral visual pathway of human eye vision to obtain the geometric relationship between the center point of the feature information and the center point of human eye vision; the human eye movement decision module is specifically used to reduce the distance between the center point of the feature information and the center point of human eye vision based on the geometric relationship between the center point of the feature information and the center point of human eye vision, so that the center point of the feature information coincides with the center point of human eye vision, and obtain the target positioning result of the target image.

[0055] In the specific implementation, the feature extraction module includes several series-connected feature extraction layers, and the target positioning model simulates the dorsal visual pathway of the human eye through several series-connected feature extraction layers to extract the feature information of the target image. Each feature extraction layer is used to extract the hierarchical feature information of the target image and pass the hierarchical feature information to the next feature extraction layer connected to it.

[0056] In one example, the feature extraction module is also used to remove weak stimulus signals from the feature information of the target image. The feature extraction module is used to simulate the cerebral cortical tissue area, and the function of the dorsal visual cells is essentially a local spatiotemporal energy filter, so we also set unnecessary neural stimulation of the filter structure. At this time, each feature extraction layer of the feature extraction module is used to extract the hierarchical feature information of the target image and filter out the weak stimulus signals in the hierarchical feature information to obtain the feature information of the target image.

[0057] That is, the feature extraction module is a layered structure, each layer contains a simple neural circuit in the sensory input, and then these layers are connected in series. The deep network composed of these layers calculates complex transformations of the input data similar to the transformations produced in the dorsal stream. Each single layer of the network contains linear and nonlinear structures, in which the added filter for screening weak stimulus signals is a linear operation, followed by FN, which stands for activation, which is a nonlinear structure. Each group of FN contains a local maximum threshold, pooling and regularization. Similar to the neural receptive field, all network operations are applied locally, on a fixed-size input area, which is usually smaller than the full spatial range of the input.

[0058] In a specific implementation, the feature information of the target image includes the spatial features and semantic features of the target image. Therefore, in this embodiment, after obtaining the feature information of the target image, the spatial features and semantic features of the target image are fused, and the geometric relationship between the center point of the fused feature information and the center point of human eye vision is obtained.

[0059] Specifically, this embodiment specifically fuses the spatial features and semantic features of the target image through the foveal positioning module, and obtains the geometric relationship between the center point of the fused feature information and the center point of the human eye vision. Among them, the foveal positioning module uses a recurrent neural network to achieve the above-mentioned feature fusion. Because for the convolutional network, the high-resolution shallow features have spatial information for positioning, and the low-resolution deep features have strong semantic information. Top-down feature fusion usually transfers the strong semantic information of the deep layer to the shallow layer, which only enhances the semantic features and does not transfer the spatial information, which is obviously unfriendly for target positioning. And the significant response of a certain spatial position is not only the current stimulus response of the receptive field at that position, but also should be affected by the contextual situation. Therefore, the recurrent neural network of this embodiment aims to further enhance the feature representation of the target through the transmission of spatial information.

[0060] The backbone network used in the recurrent neural network of this embodiment is VGG16, which includes five maxpooling layers, and four of them are used for fusion, such as Figure 4As shown in Figure 1, bidirectional feature fusion is divided into three steps. The first step is top-down transfer, which transfers the strong semantic information of the deep layer to the shallow layer (the features of S4, S3 and S2 are integrated into S1); the second step is to obtain a series of nodes with spatial features through the spatial convolutional feature extraction network. The third step is to fuse the features in two directions to extract the feature matrix (i.e. the fused feature information).

[0061] In a specific implementation, after obtaining the geometric relationship between the center point of the feature information and the center point of the human eye vision, the center point of the human eye vision is moved in several different directions based on the geometric relationship between the center point of the feature information and the center point of the human eye vision, until the center point of the feature information coincides with the center point of the human eye vision, and this process is still implemented by the human eye movement decision module. Among them, the moving direction of the center point of the human eye vision is obtained by the following steps: the pre-built coordinate axis is divided into 8 vectors with each movement of 45°, and the direction of each vector is used to characterize the moving direction of the center point of the human eye vision.

[0062] In one example, after extracting the feature information of the target image, the foveal positioning module obtains the geometric relationship between the center point of the feature information, the center point of the human eye vision, and the boundary point of the feature information. Then, based on the geometric relationship between the center point of the feature information, the center point of the human eye vision, and the boundary point of the feature information, the human eye movement decision module reduces the distance between the center point of the feature information and the center point of the human eye vision, so that the center point of the feature information coincides with the center point of the human eye vision, and obtains the target positioning result of the target image. The eight directions of movement designed by the human eye movement decision module are related to reinforcement learning, which is intended to simulate the process of human eye positioning of the target, comprehensively consider the neural response between the eyeball and the brain, and realize the eye saccade and fixation mechanism based on the fovea. This method emphasizes the use of geometric operations to reduce the distance between the target point and the visual center point. When the two points (i.e., the center point of the feature information and the center point of the human eye vision) overlap, it is the best state, and when the two points are too close to the boundary point, a penalty term is set to achieve rapid positioning of the target.

[0063] In one embodiment, the structure of the target positioning model can be as follows: Figure 5As shown in the figure, it includes: spatial convolution feature extraction module (i.e. feature extraction module), foveal positioning module and human eye movement decision module, which are referred to as spatial extraction module, positioning module and decision module respectively. Among them, the spatial extraction module simulates the extraction process of spatial information of the image by the dorsal visual stream through a hierarchical structure, and sets the filter to slightly under-respond to the stimulus; the positioning module contains two network structures, namely, the recurrent context feature fusion network and the center positioning model. The former is based on the recurrent neural network to integrate historical information and spatial information, and the latter simulates the foveal positioning mechanism of the human eye, designs a geometric calculation model based on the target point, the center point, and the boundary point (i.e. the center point of the feature information, the center point of the human eye vision, and the boundary point of the feature information) to obtain ZX, which represents a series of ω values, where ZX={ω_1,ω_2,ω_3,…ω_n}, to help complete the rapid positioning process of the target; finally, the motion decision network, which divides the coordinate axis into 8 vectors with each movement of 〖45〗^° according to the mechanism of human eye movement, and realizes the action decision of simulating the human eye based on the fusion information and ZX combined with reinforcement learning.

[0064] In order to verify the target localization model, this embodiment conducted experiments on the PASCAL VOC and MS COCO datasets and compared them with the baseline methods.

[0065] In the verification results, the impact of spatial feature extraction on recognition results: For the spatial feature extraction network, this embodiment conducted two sets of comparative experiments, one with spatial feature extraction and the other without spatial feature extraction. Overall, the network with spatial feature extraction is better than the one without AP extraction. S The improvement of 5 percentage points shows that spatial feature extraction is an important means to enhance target feature representation.

[0066] The impact of the cyclic context model on the recognition results: This embodiment conducted five experiments under the condition of embedding spatial feature extraction. The first three groups used 0.5, 0.6, and 0.8 respectively. The results were similar and the positioning results of the target were not ideal. The fourth group of experiments mixed the three ratios and conducted experiments. AP and AP S The results showed that the aspect ratios of the original image and the target in the COCO dataset are mostly opposite. Therefore, in the fifth set of experiments, the rectangular boxes of the three ratios were rotated 180°. The results showed that AP and AP S Both have increased by at least 4 percentage points, which shows that fusing context and spatial information can improve the recognition accuracy of targets.

[0067] The impact of the central positioning network on the recognition results: Compared with the previous methods, the search and positioning mode of the central positioning network for the target has been greatly improved in terms of speed and accuracy. In order to verify this result, this embodiment conducted two sets of comparative experiments, one with the central positioning network added and the other without the central positioning network added. The results show that AP and AP S The improvements were 4.3% and 3.9% respectively. The experimental results prove that the positioning mode of this geometric calculation model is indeed effective.

[0068] In this embodiment, the process of achieving target positioning by human vision is simulated, and the human visual pathway of the human eye when performing target positioning includes a ventral visual pathway and a dorsal visual pathway. By simulating the dorsal visual pathway of the human eye, the feature information of the target image can be extracted. By simulating the ventral visual pathway of the human eye, the geometric relationship between the center point of the feature information and the center point of the human eye vision can be obtained. Finally, based on the geometric relationship between the center point of the feature information and the center point of the human eye vision, the distance between the center point of the feature information and the center point of the human eye vision is reduced, so that the center point of the feature information and the center point of the human eye vision coincide, and the target positioning result of the target image can be obtained. Compared with the traditional search mode, the human eye search and positioning method does not have the limitation of the difficulty of extracting complex scene features, and can extract more comprehensive and rich feature information of the target image, thereby improving the accuracy of target positioning; and the present application starts the search and positioning from the center point, rather than a completely purposeless search, which greatly reduces the number of target searches, thereby improving the speed of target positioning.

[0069] The step division of the above methods is only for the purpose of clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this patent; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this patent.

[0070] Another embodiment of the present application relates to a device for simulating human eye vision to achieve target positioning. The implementation details of the device for simulating human eye vision to achieve target positioning in this embodiment are specifically described below. The following content is only for the convenience of understanding the implementation details provided, and is not necessary for the implementation of this solution. The schematic diagram of the device for simulating human eye vision to achieve target positioning in this embodiment can be as follows Figure 6 As shown, it includes: an image acquisition module 601, an image input module 602 and an image positioning module 603.

[0071] Specifically, the image acquisition module 601 is used to acquire a target image to be identified;

[0072] The image input module 602 is used to input the target image into the pre-trained target positioning model.

[0073] The image positioning module 603 is used to simulate the dorsal visual pathway of the human vision through the target positioning model to extract the feature information of the target image, and simulate the ventral visual pathway of the human vision to obtain the geometric relationship between the center point of the feature information and the center point of the human vision, and based on the geometric relationship between the center point of the feature information and the center point of the human vision, reduce the distance between the center point of the feature information and the center point of the human vision, so that the center point of the feature information coincides with the center point of the human vision, and obtain the target positioning result of the target image.

[0074] It is not difficult to find that this embodiment is a device embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiment are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiment.

[0075] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0076] Another embodiment of the present application relates to a computer device, such as Figure 7 As shown, it includes: at least one processor 701; and a memory 602 that is communicatively connected to the at least one processor 701; wherein the memory 702 stores instructions that can be executed by the at least one processor 701, and the instructions are executed by the at least one processor 701 so that the at least one processor 701 can execute the method of simulating human eye vision to achieve target positioning in the above-mentioned embodiments.

[0077] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0078] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0079] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0080] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as: ROM), random access memory (Random Access Memory, referred to as: RAM), disk or optical disk and other media that can store program codes.

[0081] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A method for simulating human vision to achieve target positioning, characterized in that: The method comprises: Acquire a target image to be identified; Inputting the target image into a pre-trained target localization model; The target positioning model simulates the dorsal visual pathway of human vision to extract feature information of the target image, and simulates the ventral visual pathway of human vision to obtain a geometric relationship between a center point of the feature information and a center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reduces the distance between the center point of the feature information and the center point of human vision, so that the center point of the feature information coincides with the center point of human vision, and obtains a target positioning result of the target image; The method of reducing the distance between the center point of the feature information and the center point of the human vision based on the geometric relationship between the center point of the feature information and the center point of the human vision so that the center point of the feature information coincides with the center point of the human vision includes: Based on the geometric relationship between the center point of the feature information and the center point of the human vision, the center point of the human vision is moved in several different directions until the center point of the feature information coincides with the center point of the human vision.

2. The method for simulating human vision to achieve target positioning according to claim 1, characterized in that: Before the ventral visual pathway simulating human vision obtains the geometric relationship between the center point of the feature information and the center point of human vision, the method further includes: Weak stimulation signals in the characteristic information of the target image are eliminated.

3. The method for simulating human vision to realize target positioning according to claim 2, characterized in that: The feature information of the target image includes spatial features and semantic features of the target image, and the geometric relationship between the center point of the feature information and the center point of human vision is obtained by simulating the ventral visual pathway of human vision, including: fusing the spatial features and semantic features of the target image; The geometric relationship between the center point of the fused feature information and the center point of human vision is obtained.

4. The method for simulating human vision to achieve target positioning according to claim 1, characterized in that: The moving direction of the center point of the human eye vision is obtained by the following steps: The pre-built coordinate axis is divided into 8 vectors with each movement of 45°, and the direction of each vector is used to represent the moving direction of the center point of human vision.

5. The method for simulating human vision to realize target positioning according to claim 4, characterized in that: The method further comprises: After extracting the feature information of the target image, the geometric relationship between the center point of the feature information, the center point of human vision and the boundary point of the feature information is obtained; The method of reducing the distance between the center point of the feature information and the center point of the human vision based on the geometric relationship between the center point of the feature information and the center point of the human vision so that the center point of the feature information coincides with the center point of the human vision includes: Based on the geometric relationship between the center point of the feature information, the center point of human vision and the boundary points of the feature information, the distance between the center point of the feature information and the center point of human vision is reduced so that the center point of the feature information coincides with the center point of human vision.

6. The method for simulating human vision to realize target positioning according to any one of claims 1 to 5, characterized in that: The target positioning model simulates the dorsal visual pathway of human vision through a number of series-connected feature extraction layers to extract feature information of the target image; Each feature extraction layer is used to extract hierarchical feature information of the target image and pass the hierarchical feature information to the next feature extraction layer connected thereto.

7. A device for simulating human vision to achieve target positioning, characterized in that: The device comprises: An image acquisition module, used to acquire a target image to be identified; An image input module, used for inputting the target image into a pre-trained target positioning model; An image positioning module, used to simulate the dorsal visual pathway of human vision through the target positioning model to extract feature information of the target image, and simulate the ventral visual pathway of human vision to obtain a geometric relationship between a center point of the feature information and a center point of human vision, and based on the geometric relationship between the center point of the feature information and the center point of human vision, reduce the distance between the center point of the feature information and the center point of human vision so that the center point of the feature information coincides with the center point of human vision, thereby obtaining a target positioning result of the target image; Among them, the image positioning module is used to move the center point of human vision in several different directions based on the geometric relationship between the center point of the feature information and the center point of human vision until the center point of the feature information coincides with the center point of human vision.

8. A computer device, characterized in that: include: at least one processor; And, a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for simulating human eye vision to achieve target positioning as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for simulating human eye vision to realize target positioning as described in any one of claims 1 to 6 is implemented.