Navigation method, device and electronic equipment

Abstract

The disclosed embodiment discloses a navigation method, device and electronic device. The method includes: based on the acquired historical step information set and the visual information of the current step, redundant information processing is performed on the historical step information set to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical visual information; based on the first historical step information set, the visual information of the current step and the text instruction information, the next movement information is determined. Since redundant information processing is performed on the historical information set, the redundant information in the historical step information involved in the next step decision is reduced, and the negative impact of the redundant information in the historical step on the next navigation decision can be reduced, thereby making the navigation information more accurate.

Description

Technical Field

[0001] The present disclosure relates to the field of autonomous navigation technology, and in particular to a navigation method, device and electronic equipment. Background Art

[0002] With the development of intelligent agent technology, intelligent agents have been used more and more widely, such as household intelligent agents, commercial intelligent agents, industrial intelligent agents, etc. A prerequisite for the use of intelligent agents is the need to navigate the route.

[0003] One of the current navigation technologies is visual language navigation technology, which combines given language instructions with collected visual image information to autonomously adjust the direction of travel so as to reach the designated destination. Summary of the invention

[0004] This disclosure section is provided to introduce concepts in a brief form, which will be described in detail in the detailed description section below. This disclosure section is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0005] Embodiments of the present disclosure provide a navigation method, device, and electronic device.

[0006] In a first aspect, an embodiment of the present disclosure provides a navigation method, the method comprising: performing redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step, to obtain a first historical step information set after the redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes the historical step visual information; determining the next step movement information based on the first historical step information set, the visual information of the current step and the text instruction information

[0007] In a second aspect, an embodiment of the present disclosure provides a navigation device, which includes: a redundant information processing unit, used to perform redundant information processing on a historical step information set based on an acquired historical step information set and visual information of a current step, to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; a movement information determination unit, used to determine the movement information of the next step based on the first historical step information set, the visual information of the current step and the text instruction information.

[0008] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the navigation method as described in the first aspect.

[0009] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the navigation method as described in the first aspect.

[0010] The navigation method, device and electronic device provided by the embodiments of the present disclosure perform redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; based on the first historical step information set, the visual information of the current step and the text instruction information, the next movement information is determined. Since the redundant information processing is performed on the historical information set, the redundant information is not included in the historical step information involved in the next step decision, which can reduce the negative impact of the redundant information in the historical step on the next navigation decision, thereby making the navigation information more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0012] Figure 1 is a flow chart of an embodiment of a navigation method according to the present disclosure;

[0013] Figure 2 is a schematic structural block diagram for redundant information processing provided by the present disclosure;

[0014] Figure 3 A schematic block diagram of the navigation method disclosed in the present invention

[0015] Figure 4 is a structural schematic diagram of an embodiment of a navigation device according to the present disclosure;

[0016] Figure 5 is an exemplary system architecture in which the navigation method of an embodiment of the present disclosure may be applied;

[0017] Figure 6 It is a schematic diagram of the basic structure of an electronic device provided according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0019] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0020] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0021] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0022] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0023] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0024] Please refer to Figure 1 , which shows a flow diagram of some embodiments of the navigation method provided by the present disclosure, such as Figure 1 As shown, the navigation method includes the following steps.

[0025] Step 101, based on the acquired historical step information set and the visual information of the current step, redundant information processing is performed on the historical step information set to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information.

[0026] The historical step information set may include at least one historical step information. The historical step information may include historical step visual information and reliability information. The reliability information is used to characterize the matching degree between the historical step visual information and the instruction text.

[0027] The historical step visual information includes the panoramic information corresponding to the historical step (the panoramic information of the historical step includes the visual features of each single perspective of the panorama and the heading angle and pitch angle of each single perspective of the panorama), the visual features of the agent's orientation perspective, and the agent's orientation rotation angle from the previous historical step to this historical step.

[0028] The visual information of the current step includes the panoramic information of the current step (the panoramic information of the current step includes the visual features of each single perspective of the panoramic view of the current step and the heading angle and pitch angle of each single perspective of the panoramic view), the visual features of the agent's orientation perspective, and the orientation rotation angle of the agent from the previous step to the current step.

[0029] As a schematic illustration, the agent has gone through t-1 historical steps before the current step (t-th step). The historical step information here may include information from step 1 to step t-1.

[0030] That is, at step t, redundant information processing can be performed on each of the historical step information from step 1 to step t-1. The processing process can be to perform redundant information processing on each historical step one by one according to the visual information of the current step.

[0031] For each step, the visual information corresponding to the step can be stored. The stored multiple historical step information constitutes a historical step information set. In addition, when each new step is completed, the visual information and reliability information corresponding to the step are stored in the historical step information set, and the historical step information set is updated.

[0032] When the intelligent agent reaches the t-th step, it can gradually extract the information of each historical step from the 1st step to the t-1th step from the historical step information set, and perform redundant information processing on the extracted information of each historical step.

[0033] When performing redundant information processing on historical step information, visual information of the current step can be used. The visual information of the current step is compared with the visual information of the historical step, and at least part of the redundant information in the visual information of the historical step is removed, thereby achieving the processing of redundant information of the visual information of the historical step. After removing the redundant information of the visual information of each historical step, a first historical step information set is obtained.

[0034] Step 102, determining the next step movement information based on the first historical step information set and the visual information and text instruction information of the current step.

[0035] After obtaining the first historical step information set, the movement information of the next step may be determined by combining the first historical step information set, the visual information of the current step, and the text instruction information.

[0036] The above text instruction information may be pre-set standard path information. The standard path information may include, for example, a topological structure map of a preset environment. The topological structure map of the preset environment here may be, for example, a topological structure map of a preset house, and movement information corresponding to each node. For example, "walk into the corridor and pass through the entrance of the kitchen area. Walk past the sink and stove, and stop between the refrigerator and the dining table."

[0037] For example, the historical step visual information in the first historical step information set includes visual information of a corridor and a kitchen entrance in a preset environment. If the current position indicated by the visual information of the current step is a sink, it can be determined that the next movement information includes moving toward the stove.

[0038] The navigation of the intelligent agent is a conditional decision-making process, that is, its decision will be affected by historical information, the current visual scene and instructions. In the related technology, a memory mechanism is used to assist the intelligent agent in navigation. The visual information and rotation angle information corresponding to each historical node are stored in the memory. When the intelligent agent makes a decision to select the node to be transmitted next, the visual information of the historical node in the memory, the visual information of the current perspectives and the text instruction information are referred to determine the target node for the next transmission. However, there is both beneficial and unhelpful information in the historical information for the current navigation step. In this way, the related technology will inevitably incorporate the noise brought by the unhelpful information into the navigation decision of the current step, thereby having a negative impact on the navigation of the next step.

[0039] The navigation method provided in this example performs redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; based on the first historical step information set and the visual information and text instruction information of the current step, the next movement information is determined. Since redundant information processing is performed on the historical information set, the redundant information in the historical step information involved in the next step decision is reduced, thereby reducing the negative impact of the redundant information in the historical step on the next navigation decision, thereby making the navigation information more accurate.

[0040] In some optional implementations, the historical step visual information includes historical step visual features; the historical step information also includes historical step reliability, and the redundant information processing is performed on the historical step information set based on the acquired historical step information set and the visual information of the current step to obtain the first historical step information set after redundant information processing, including: extracting the current step visual features from the current step visual information; performing redundant information processing on the historical step visual features according to the current step visual features and the historical step reliability to obtain the first historical step information set after redundant information processing.

[0041] In these optional implementations, the visual information of the current step t includes the panoramic information of the current step, the visual features of the agent's orientation angle, and the information of the agent's rotation angle features from the t-1th step to the tth step. The panoramic information of the current step is encoded, and the encoding result is O t The visual features of the current agent’s view angle are encoded, and the encoding result is represented as I t The rotation angle feature of the agent from the previous step to the current step is encoded, and the encoding result is represented as a t The encoding result of the above panoramic image information is O t , Visual feature encoding results of orientation perspective I t The encoding result of the rotation angle feature a t Fusion is performed to obtain the current step visual features.

[0042] In some application scenarios, the extracting of the current step visual features from the current step visual information includes: inputting the current step visual information into a preset visual feature extraction unit to obtain the current step visual features. Inputting the panoramic image information of the current step, the visual features of the viewing angle, and the rotation angle feature information of the agent from the previous step to the current step into the preset visual feature extraction unit to obtain the current step visual features.

[0043] The preset visual feature extraction unit may be, for example, a preset encoding unit. The preset visual feature extraction unit may encode the panoramic information of the current step, the visual features of the agent's orientation angle, and the features of the rotation angle of the agent from the previous step to the current step, to generate the visual features of the current step.

[0044] The visual features of each step can be obtained by encoding by the preset encoding unit.

[0045] The current step visual feature includes the encoding result of the panoramic image information O t , Visual features of the agent’s orientation perspective I t and the rotation angle feature a of the agent from the previous step to the current step t The visual feature of the current step t can be expressed as h(t).

[0046] The acquisition of the visual features of the historical steps can also refer to the above-mentioned method of acquiring the visual features of the current step.

[0047] The historical step information also includes historical step reliability, which is used to characterize the matching degree between historical visual features and text instruction information.

[0048] When the current step visual information is used to process the redundant information of the historical step visual information, the current step visual feature is used to process the redundant information of the historical step visual feature. That is, the features extracted from the historical step visual information are analyzed and processed according to the features extracted from the current step visual information, thereby removing at least part of the redundant information of the historical step visual information, which can simplify the amount of data that needs to be processed when determining the next moving operation.

[0049] In addition, the reliability corresponding to the historical step visual information is involved in the redundant information processing of the historical step visual information. For the historical step information with higher reliability, less redundant information can be removed, and for the historical step information with lower reliability, more redundant information can be removed.

[0050] Redundant information processing is performed on the historical step information with reference to the reliability of the historical step, so that more redundant information processing is performed on the historical step information with lower reliability, so that more useless historical information can be removed during the redundant information processing of the historical step, thereby further reducing the interference of the useless information of the historical step.

[0051] In some optional implementations, redundant information processing is performed on historical step visual features according to the current step visual features and the historical step reliability to obtain a first historical step information set after redundant information processing, including: inputting the current step visual features, the historical step reliability and the historical step visual features into a preset erasure unit to obtain first erasure information; and performing redundant information processing on the historical step visual features in the historical step information set using the first erasure information to obtain the first historical step information set.

[0052] The preset erasure unit here may include a machine learning model. The machine learning model here may be a neural network model, a deep learning model, etc. The preset erasure unit may be trained in advance to obtain a trained preset erasure unit. The preset erasure unit is used to output first erasure information based on the input current step visual features, historical step reliability, and historical step visual features. The first erasure information here may include, for example, an erasure vector. Erasure vector It is represented by the following formula (1):

[0053]

[0054] Among them, h i is the visual feature of the i-th historical step; ri is the reliability of the i-th historical step, h t is the visual feature of the current step t. Here, the erase vector may be a vector output by the first fully connected network, which is related to the visual feature of the i-th historical step, the reliability of the i-th historical step, and the visual feature of the current step t. For example, the first fully connected network takes the visual feature of the i-th historical step, the reliability of the i-th historical step, and the visual feature of the current step t as input variables.

[0055] Redundant information processing can be performed on the historical step visual features according to the erasure information. Specifically, the redundant information processing process can be represented by the following formula (2):

[0056]

[0057] in, It is the visual feature of the i-th historical step obtained after redundant information processing.

[0058] The i-th historical step visual feature obtained after redundant information processing is used to update the historical step information set. After the multiple historical step visual features are used to update the historical step information set, the first historical step information set is obtained. The historical step visual feature after redundant information processing can be regarded as the first historical step visual feature.

[0059] In some application scenarios, the above-mentioned step 102 includes: performing cross-modal feature information processing on each first historical step visual feature, current step visual feature and next step perspective feature and text instruction feature in the first historical step information set to obtain a second historical step visual information set, a first next step perspective feature and a first text instruction feature; determining the next step movement information based on the second historical step visual information set, the first next step perspective feature and the first text instruction feature.

[0060] The text instruction feature can be obtained by extracting text features from instructions recorded in text form. The process of extracting text features from text instructions can include segmenting the text instructions to obtain multiple segmentation results. Each segmentation result is searched from a preset encoding library to find the code corresponding to each segmentation result. The codes corresponding to each segmentation result are combined to obtain the text instruction feature code.

[0061] Specifically, each first historical step visual feature, current step visual feature, next step viewing feature and text instruction feature in the first historical step information set can be input into a pre-trained cross-modal conversion unit, and the cross-modal conversion unit outputs the second historical step visual information set, the first next step viewing feature and the first text instruction feature; the cross-modal conversion unit is used to cross-process the visual features and text features.

[0062] The cross-modal conversion unit includes a multi-level combination network. Each level of the combination network includes a cross-modal attention network, a self-attention network and a feedforward network. The historical step visual features, the current step visual features and the next step viewing angle features are spliced ​​with the text instruction features and input into the cross-modal conversion unit. Specifically, the spliced ​​historical step visual features, the current step visual features and the next step viewing angle features and the text instruction features are first input into the cross-modal attention network, and the cross-modal attention network performs cross-analysis processing on the fused historical step visual features, the current step visual features and the next step viewing angle features and the text instruction features. Then the output of the cross-modal attention network is input into the self-attention network, and the self-attention network performs self-attention processing on the above features after the cross-analysis processing. The output of the self-attention network is input into the feedforward network, and the feedforward network transmits the output of the self-attention network to the next level of the combination network. After being processed by the multi-level combination network, the second historical step visual information set, the first next step viewing angle feature and the first text instruction feature are output. The first next step viewing angle feature and the first text instruction feature can be input into the action prediction network to obtain the next step movement information.

[0063] The next move information here includes the agent's moving direction and moving distance.

[0064] In some application scenarios, the above navigation method further includes the following steps:

[0065] First, cross-modal feature information processing is performed on the first historical step visual features, current step visual features, next step viewing features and text instruction features in the first historical step information set to obtain the second historical step visual information set, the first next step viewing features and the first text instruction features.

[0066] Secondly, the second historical step visual feature, the reliability corresponding to the historical step and the first text instruction feature in the second historical step visual information set are input into a preset erasure unit to obtain second erasure information.

[0067] Specifically, the second historical step visual feature, the reliability corresponding to the historical step and the first text instruction feature may be spliced ​​and input into the preset erasing unit.

[0068] The preset erasing unit here may include a pre-trained machine learning model. The machine learning model here may include various neural network models, deep learning models, etc. The preset erasing unit may be trained in advance to obtain a trained preset erasing unit. The training of the erasing unit may be performed using a supervised training method.

[0069] Thirdly, redundant information processing is performed on each historical step visual feature in the second historical step visual feature set by using the second erasure information to obtain the third historical step visual information set.

[0070] Here, the second erasure information may include an erasure vector

[0071]

[0072] in, is the visual feature of the second historical step of the i-th historical step; r i is the reliability of the i-th historical step, is the first text instruction feature. Here, the erase vector may be a vector output by the second fully connected network, and the second fully connected network may be related to the reliability of the second historical step visual feature historical step and the first text instruction feature. For example, the second fully connected network takes the second historical step visual feature, the reliability of the historical step, and the first text instruction feature as input variables. In some application scenarios, the first fully connected network may be the same fully connected network as the second fully connected network.

[0073] The redundant information processing can be performed on the second historical step visual feature according to the second erasure information. Specifically, the redundant information processing process can be represented by the following formula (4):

[0074]

[0075] in, It is the visual feature of the third historical step of the ith step obtained after redundant information processing using the second erasure information.

[0076] The second historical step information set is updated using the i-th third historical step visual feature obtained after the redundant information processing. The second historical step information set is updated using multiple third historical step visual features to obtain a third historical step information set.

[0077] Finally, the reliability of the current step visual feature is determined according to the global visual feature in the third historical step visual information set and the first instruction text feature.

[0078] The reliability of the current step visual feature is used to characterize the matching degree between the current step visual feature and the instruction information.

[0079] The reliability of the current step visual feature can be determined by using the visual feature obtained by erasing redundant information from each historical step visual feature in the historical step visual feature set through the second erasure information and the first instruction text feature.

[0080] In some application scenarios, determining the reliability of the current step visual feature based on the third historical step visual feature set and the first instruction text feature includes: inputting the third historical step visual feature set and the first instruction text feature into a pre-trained reliability prediction unit, and the reliability prediction unit outputs the reliability of the current step visual feature.

[0081] The reliability prediction unit here may include various pre-trained machine learning models. The reliability prediction unit may determine the matching degree between the visual features and the text instruction features according to the inputted historical step visual features and text instruction features.

[0082] The training of the machine learning model in the reliability prediction unit can be performed using supervised training or unsupervised training.

[0083] In these optional implementations, the historical step features are erased a second time using second erasure information. Since the second erasure information is related to the text instruction information, redundant information irrelevant to the text instruction information is removed from the third historical step features obtained from the second erasure information. Then the reliability between the current step historical features and the text instructions is determined, so that the obtained reliability of the current step is less affected by noise, and a higher accuracy reliability can be obtained.

[0084] While obtaining the next step movement information, the user can also learn about the reliability of the match between the current visual features and the text instructions. The reliability of the current step can, to a certain extent, represent the accuracy of moving from the starting point through each step to the current step, and the accuracy of the next step movement information determined by the visual features of the current step, the visual features of the historical steps, the reliability of the historical steps and the text instructions can also be preliminarily estimated based on the reliability of the current step.

[0085] Please refer to Figure 2 , which shows a schematic structural block diagram for redundant information processing provided by the present disclosure. Figure 2 As shown, the structural block diagram for redundant information processing may include a visual erasing module and a text erasing module.

[0086] First, the visual erasure module is used to process the redundant visual information of historical steps. The processing flow is as follows:

[0087] (1) The preset encoding unit encodes the current step panoramic image information (for details, please refer to the previous content) to obtain the encoding result O t . Encode the visual features of the agent’s viewpoint and obtain the encoding result I t . Encode the rotation angle of the agent from the previous step to this step and get the encoding result a t . The encoding result O t,I t and a t Input to the fusion unit for fusion to obtain the latest historical feature h t .

[0088] (2) The latest historical step visual feature h t As the scene-related visual features, together with the historical step features h sampled from the historical step visual information i , and its corresponding reliability r i , is sent to the preset erasure unit (such as the erasure vector prediction network in the figure) to predict the visual erasure vector

[0089]

[0090] (3) Use the visual erase vector Perform noise erasure on the sampled historical visual features:

[0091]

[0092] (4) Use the erased historical features Update the historical step information set to obtain the first historical step information set.

[0093] (5) Sample historical features in sequence and repeat steps (2) to (4) until the redundant information of the visual features of all historical steps has been removed.

[0094] Second, for the text erasing module, its workflow is similar to that of the visual erasing module, and it can share the same preset erasing unit with the visual erasing module. The workflow of the text erasing module is as follows:

[0095] (1) The first historical step of visual information collection Follow the next step perspective feature Instruction text features {w 0 ,w 1 ,...,w L} are sent to the cross-modal conversion unit (cross-modal Transformer) together, and the visual-text modality features are fused through the mutual attention and self-attention modules to obtain the fused second historical step visual information set First step perspective feature and the first instruction text feature Extract the global representation of the instruction in the current scenario from the first instruction text And use it as scene-related text features. Among them, It is the global historical step feature in the historical step information set; is the next step global perspective feature; w 0, Is a global directive text feature.

[0096] (2) Scene-related text features Together with the i-th historical step feature and the corresponding reliability r i , fed into the erasure vector generation network to predict the text erasure vector:

[0097]

[0098] (3) Use the text erase vector Perform noise erasure on the sampled i-th historical step features:

[0099]

[0100] (4) Use the erased history step feature Update the visual feature set of the second historical step;

[0101] (5) Sample historical features in sequence and repeat steps (2) to (4) until all historical visual features in memory have been noise-erased.

[0102] Among them, the erasure vector generation network is implemented by a multi-layer perceptron.

[0103] Please refer to Figure 3 , which shows a principle framework diagram of the navigation method provided by the present disclosure. Figure 3 As shown, the navigation method provided by the present disclosure includes:

[0104] (1) The i-th historical step visual feature h i and reliability r i And the current step visual feature h t The visual erasure module includes a preset erasure unit and a first processing unit. The preset erasure unit is used to generate first erasure information, and the first processing unit is used to erase redundant information of the i-th historical step visual feature according to the first erasure information to obtain the first historical step visual feature.

[0105] (2) The first historical step visual information set consisting of the first historical step visual features after erasure Follow the next step perspective feature Instruction text features {w 0 ,w 1 ,...,w L} are sent to the cross-modal conversion unit (cross-modal Transformer) together, and the visual-text modality features are fused through the mutual attention and self-attention modules to obtain the fused second historical step visual information set First step perspective feature and the first instruction text feature in, is the historical step global visual feature in the first historical step visual information set. 0 is the global text feature in the instruction text, is the global visual feature of the historical step in the second historical step visual information set. is the global text feature in the first instruction text.

[0106] (3) The action prediction network selects the perspective with the highest similarity based on the similarity between the first next-step perspective features of the current surrounding navigable area and the global features in the first instruction text, and uses the navigation point within it as the next-step transmission node.

[0107] (4) Global feature representation in the first instruction text feature As the text feature related to the current step, along with the visual feature of the second historical step of the ith and its corresponding reliability r i , erase the noise in the history through the text erasure module.

[0108] The text erasing module includes a preset erasing unit and a second processing unit. The preset erasing unit is used to generate second erasure information, and the second processing unit is used to process the visual feature of the second historical step i according to the second erasure information. Erase redundant information and obtain the visual features of the third historical step i A plurality of third history step visual features constitute a third history step information set.

[0109] (5) Based on the global visual features in the visual information set of the third historical step after erasure Global features in the first instruction text features in the current scenario The reliability evaluation module is used to evaluate the matching degree of text vision in the current scene. Since the current scene is obtained after the execution of the previous action, and the decision quality of the previous action depends on the reliability of the corresponding state of the previous node, the matching degree of text vision in the current scene can be regarded as the reliability of the historical features encoded by the previous node.

[0110] like Figure 4As shown, the navigation device of this embodiment includes: a redundant information processing unit 401 and a movement information determination unit 402. The redundant information processing unit 401 is used to perform redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step, so as to obtain a first historical step information set after the redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; the movement information determination unit 402 is used to determine the movement information of the next step based on the first historical step information set, the visual information of the current step and the text instruction information.

[0111] In this embodiment, the specific processing of the redundant information processing unit 401 and the mobile information determining unit 402 of the navigation device and the technical effects thereof can be referred to respectively. Figure 1 The relevant descriptions of step 101 and step 102 in the corresponding embodiment are not repeated here.

[0112] In some optional implementations, the historical step visual information includes historical step visual features; the historical step information also includes historical step reliability, and the redundant information processing unit 401 is further used to: extract the current step visual features from the current step visual information; perform redundant information processing on the historical step visual features in the historical step information set according to the current step visual features and the historical step reliability, to obtain a first historical step information set after redundant information processing.

[0113] In some optional implementations, the redundant information processing unit 401 is further used to: input the current step visual feature, the historical step reliability and the historical step visual feature into a preset erasure unit to obtain first erasure information; use the first erasure information to perform redundant information processing on the historical step visual features in the historical step information set to obtain the first historical step information set.

[0114] In some optional implementations, the text instruction information includes instruction text features; the first historical step information set includes at least one first historical step visual feature after redundant information processing; the movement information determination unit 402 is further used to: perform cross-modal feature information processing on each first historical step visual feature, current step visual feature and next step viewing feature in the first historical step information set and the text instruction feature to obtain a second historical step visual feature set, a first next step viewing feature and a first text instruction feature; determine the next step movement information based on the second historical step visual information set, the first next step viewing feature and the first text instruction feature.

[0115] In some optional implementations, the movement information determination unit 402 is further used to: input each first historical step visual feature, current step visual feature, next step viewing feature and text instruction feature in the first historical step information set into a pre-trained cross-modal conversion unit, and the cross-modal conversion unit outputs the second historical step visual information set, the first next step viewing feature and the first text instruction feature; the cross-modal conversion unit is used to cross-process the visual features and text features.

[0116] In some optional implementations, the cross-modal conversion unit includes a multi-stage combination network, each combination network including: an attention network, a self-attention network and a feedforward network.

[0117] In some optional implementations, the navigation device also includes a reliability determination unit 403, and the reliability determination unit 403 is used to: perform cross-modal feature information processing on each first historical step visual feature, current step visual feature, and next step perspective feature and text instruction feature in the first historical step information set to obtain a second historical step visual information set, a first next step perspective feature, and a first text instruction feature; input the second historical step visual feature, the reliability corresponding to the historical step, and the first text instruction feature in the second historical step visual information set into a preset erasure unit to obtain second erasure information; use the second erasure information to perform redundant information processing on each historical step visual feature in the second historical step visual feature set to obtain the third historical step visual information set; determine the reliability of the current step visual feature based on the third visual global feature and the first instruction text feature in the third historical step visual information set.

[0118] In some optional implementations, the reliability determination unit 403 is further used to: input the third visual global feature and the first instruction text feature into a pre-trained reliability prediction unit, and the reliability prediction unit outputs the reliability of the current step visual feature.

[0119] Please refer to Figure 5 , Figure 5 An exemplary system architecture is shown in which the navigation method according to one embodiment of the present disclosure can be applied.

[0120] like Figure 5 As shown, the system architecture may include an intelligent terminal 501, a network 502, and a server 503. The network 502 is used to provide a medium for a communication link between the intelligent terminal 501 and the server 503. The network 502 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0121] The intelligent terminal 501 can interact with the server 503 through the network 502 to receive or send messages, etc. The intelligent terminal 501 can be an intelligent agent. The intelligent terminal can install an image setting application, a communication application, and a control application. The communication application in the intelligent terminal 501 can exchange information with the server through the network 502. The control application controls the action of the intelligent terminal through the instructions received from the server.

[0122] The intelligent terminal 501 can be hardware or software. When the intelligent terminal 501 is hardware, it can be an intelligent agent. When the intelligent terminal 501 is software, it can be installed in the intelligent agent listed above. It can be implemented as multiple software or software modules (for example, software or software modules for providing distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0123] The server 503 can provide various services, such as receiving image information sent by the intelligent terminal 501, analyzing and processing the image information, and sending the analysis and processing results (such as the next movement information of the intelligent body) to the intelligent terminal 501.

[0124] It should be noted that the navigation method provided in the embodiment of the present disclosure may be executed by a terminal, and accordingly, a navigation device may be disposed in the smart terminal 501 , or may be executed by a server 503 , and accordingly, the navigation device may be disposed in the server 503 .

[0125] It should be understood that Figure 5 The number of terminals, networks and servers in the embodiment is only for illustration. Any number of terminals, networks and servers may be provided according to the implementation requirements.

[0126] Reference below Figure 6 , which shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure. The electronic device here generally refers to a terminal or server in the form of hardware (for example Figure 5 The terminal device in the embodiments of the present disclosure may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0127] like Figure 6As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0128] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0129] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0130] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0131] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0132] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0133] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: performs redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; determines the next step movement information based on the first historical step information set, the visual information of the current step and the text instruction information.

[0134] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0135] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0136] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0137] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0138] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0139] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0140] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0141] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A navigation method, comprising: Based on the acquired historical step information set and the visual information of the current step, redundant information processing is performed on the historical step information set to obtain a first historical step information set after redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes historical step visual information; Based on the first historical step information set, the visual information of the current step and the text instruction information, the movement information of the next step is determined.

2. The method according to claim 1, wherein: The historical step visual information includes the historical step visual features, and based on the acquired historical step information set and the visual information of the current step, redundant information processing is performed on the historical step information in the historical step information set to obtain a first historical step information set after redundant information processing, including: Extracting current step visual features from current step visual information; According to the current step visual feature and the historical step reliability, redundant information processing is performed on the historical step visual features in the historical step information set to obtain a first historical step information set after redundant information processing.

3. The method according to claim 2, wherein: The redundant information processing is performed on the historical step visual features according to the current step visual features and the historical step reliability to obtain the first historical step information set after the redundant information processing, including: Inputting the current step visual feature, the historical step reliability and the historical step visual feature into a preset erasing unit to obtain first erasing information; The first erasure information is used to perform redundant information processing on the historical step visual features in the historical step information set to obtain a first historical step information set.

4. The method according to any one of claims 1 to 3, wherein: The text instruction information includes instruction text features; the first historical step information set includes at least one first historical step visual feature after redundant information processing; the step of determining the next step movement information based on the first historical step information set and the visual information and text instruction information of the current step includes: Performing cross-modal feature information processing on each first historical step visual feature, current step visual feature, next step viewing angle feature, and text instruction feature in the first historical step information set to obtain a second historical step visual information set, a first next step viewing angle feature, and a first text instruction feature; The movement information of the next step is determined according to the second historical step visual information set, the first next step viewing angle feature and the first text instruction feature.

5. The method according to claim 4, wherein: The cross-modal feature information processing is performed on each first historical step visual feature, current step visual feature, next step viewing angle feature and text instruction feature in the first historical step information set to obtain a second historical step visual information set, a first next step viewing angle feature and a first text instruction feature, including: The first historical step visual features, current step visual features, next step viewing features and text instruction features in the first historical step information set are input into a pre-trained cross-modal conversion unit, and the cross-modal conversion unit outputs the second historical step visual information set, the first next step viewing features and the first text instruction features; the cross-modal conversion unit is used to cross-process the visual features and the text features.

6. The method according to claim 5, wherein: The cross-modal conversion unit includes a multi-level combination network, each level of the combination network includes: a cross-modal attention network, a self-attention network and a feedforward network; wherein The cross-modal attention network is used to perform cross-analysis processing on the fused historical step visual features, current step visual features, next step viewing angle features and text instruction features; The self-attention network is used to perform self-attention processing on the historical step visual features, the current step visual features, the next step viewing angle features and the text instruction features after the cross-analysis processing output by the cross-modal attention network; The feedforward network is used to transmit the output of the self-attention network to the next-level combination network.

7. The method according to claim 4, wherein: The historical step information also includes historical step reliability; the method also includes: Performing cross-modal feature information processing on each first historical step visual feature, current step visual feature, next step viewing angle feature, and text instruction feature in the first historical step information set to obtain a second historical step visual information set, a first next step viewing angle feature, and a first text instruction feature; Inputting the second historical step visual features in the second historical step visual information set, the reliability corresponding to each historical step, and the first text instruction features into a preset erasure unit to obtain second erasure information; Using the second erasure information, redundant information processing is performed on each historical step visual feature in the second historical step visual information set to obtain a third historical step visual information set; The reliability of the current step visual feature is determined according to the third visual global feature and the first instruction text feature in the third historical step visual information set.

8. The method according to claim 7, wherein: The step of determining the reliability of the current step visual feature according to the third visual global feature and the first instruction text feature in the third historical step visual information set includes: The third visual global feature and the first instruction text feature are input into a pre-trained reliability prediction unit, and the reliability prediction unit outputs the reliability of the current step visual feature.

9. A navigation device, comprising: A redundant information processing unit, configured to perform redundant information processing on the historical step information set based on the acquired historical step information set and the visual information of the current step, to obtain a first historical step information set after the redundant information processing, wherein the historical step information set includes at least one historical step information, and the historical step information includes the historical step visual information; The movement information determination unit is used to determine the movement information of the next step based on the first historical step information set, the visual information of the current step and the text instruction information.

10. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

11. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images