Automatic natural language description method for live working field behaviors based on multiple models

Through multi-model combination technology, including personnel detection and posture recognition models, the problem of inability to correctly describe live operations in the prior art is solved, and a stable and comprehensive behavioral automated natural language description is achieved in a changing environment.

CN119007284AInactive Publication Date: 2024-11-22TONGHUA POWER SUPPLY COMPANY STATE GRID JILIN ELECTRIC POWER
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411039782.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the face of the changing environment of live work sites, the existing technology cannot correctly describe the behavior, resulting in the description results being meaningless. Relying on a specific single model and training data set, it is difficult to exhaust all violations, especially when new ones appear on the scene. The descriptions of the actions, objects, and directions are inconsistent.

Method used

Using a multi-model combination method, including personnel detection depth model, equipment description model, arm motion recognition model, leg motion recognition model and posture recognition model, through multiple steps, we can describe the equipment and personnel behavior at the live work site one by one, and realize automated natural language. describe.

Benefits of technology

It realizes a more stable and comprehensive behavior description ability, can correctly describe behavior when facing a changeable environment, and does not rely on a specific single model, which enhances the stability and comprehensiveness of the description.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007284A_ABST
    Figure CN119007284A_ABST
Patent Text Reader

Abstract

The invention discloses an automatic natural language description method for live working site behaviors based on multiple models, and belongs to the field of live working site description and analysis. According to the method, the problem that the behavior cannot be correctly described in a variable environment of a live working site in the prior art is solved. The description process of the method does not depend on a specific single model, but utilizes a plurality of models to describe behaviors of constructors in the live-line work scene from the aspects of equipment, personnel and actions of different parts, so that more stable and comprehensive description capability is realized, and the construction personnel can be effectively protected even in the changeable environment of the live-line work scene. The method disclosed by the invention can also correctly describe the behavior. The method can be applied to live working site description.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of description and analysis of live working sites, and in particular to a multi-model-based method for automatic natural language description of live working site behaviors. Background Art

[0002] The analysis and monitoring of live working sites are very important during the inspection and maintenance of power equipment. In recent years, with the emergence of large language models such as ChatGPT, the technical path of describing group behaviors based on text and natural language and then conducting analysis has shown powerful and flexible new features. Converting live working site behaviors into natural language descriptions is the basis of this type of technology. Therefore, it is of great significance to automatically describe live working site behaviors in natural language.

[0003] For the description and analysis of live working behaviors in a picture, the current traditional methods include: 1) scene classification technology, using CNN and other models to analyze the scene to obtain category information, and then convert the category information into text description; this type of method is good for typical positions, objects and related relationships, but has poor description capabilities for detailed actions and postures; 2) using RNN neural network for image caption description, this type of method inputs an image and automatically outputs a descriptive text; the disadvantage of this type of method is that it requires a huge amount of training data. In the case of insufficient data, slight changes in angles and directions can lead to the loss of output text content. 3) Using action recognition to detect illegal operation behaviors. The typical problem of this type of method is that it is difficult to enumerate all illegal behaviors in the laboratory stage, so it performs well in the laboratory, but in the field, due to the emergence of new actions, objects, and directions, the description content is often inconsistent with the facts.

[0004] In summary, existing technologies usually rely on a specific model, which in turn relies on a specific training data set. When faced with a changing on-site environment, this single dependency makes it impossible to correctly describe the behavior, resulting in the description results losing their reference significance. Therefore, a method is needed to automatically describe the behavior of live working on-site in natural language. This description process does not rely on a specific single model but multiple groups of models, making the description more stable and comprehensive. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the existing methods cannot correctly describe the behavior when facing the changing environment of the live working site, and propose a method for automatic natural language description of the behavior of the live working site based on multiple models. By introducing multiple intelligent models, the equipment and personnel at the live working site, as well as the key actions of the personnel, are described step by step, so as to realize the automatic natural language description of the behavior of the live working site.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a multi-model-based automatic natural language description method for on-site behavior of live working, the method specifically comprising the following steps:

[0007] Step S1, inputting a personnel detection depth model ModelR, a device description model ModelM, an arm action recognition model ModelH, a leg action recognition model ModelL, and a posture recognition model ModelZ; establishing a multi-model list ModelList;

[0008] Step S2, establishing a single person description operator OptRY, where the input of OptRY is the person image OptRYInput, and the output is the person's behavior language description OptRYOutput;

[0009] Step S3, establishing an operator OptQT for describing the behavior of a group of people with distance, wherein the input of OptQT is the scene image OptQTInput, and the output of the OptQT operator is the language description of the behavior of a group of people with distance OptQTOutput;

[0010] Step S4, establishing a multi-model live working site automation natural language description operator OptDD, the input of OptDD is the picture OptDDInput of the live working site, and the output of OptDD is the natural language description OptDDOutput of the live working site;

[0011] Step S5, input the electrical work site picture DDImage, and use OptDD to obtain the automatic natural language description result ZDDesc of the live work site behavior.

[0012] Furthermore, the specific process of step S1 is as follows:

[0013] S101, inputting a person detection deep model ModelR, where ModelR is a recognition model based on the Yolo model, and uses a box to mark the area where a person is in a picture or video;

[0014] S102, inputting a device description model ModelM, where ModelM is a recognition model based on a Yolo model, and marking a device area and a device name with a box;

[0015] S103, inputting the arm action recognition model ModelH, where ModelH is a two-stream convolutional neural network used to recognize the arm posture of a person;

[0016] S104, inputting a leg action recognition model ModelL, where ModelL is a two-stream convolutional neural network used to recognize the posture of a person's legs;

[0017] S105, inputting a posture recognition model ModelZ, where ModelZ is a two-stream convolutional neural network used to recognize the body posture of a specific person;

[0018] S106, establish a multi-model list ModelList = [ModelR, ModelM, ModelH, ModelL, ModelZ].

[0019] Furthermore, the specific process of step S2 is as follows:

[0020] S201, establishing a single person description operator OptRY, where the OptRY input is the person image OptRYInput;

[0021] S202, create a behavior language description of personnel OptRYOutput = character string "construction personnel";

[0022] S203, using the three elements of ModelList to identify OptRYInput, and storing the identification result in the first temporary identification tag OptRYTemp1; converting OptRYTemp1 into a character string describing the arm action;

[0023] S204, using the four elements of ModelList to identify OptRYInput, and storing the identification result in the second temporary identification tag OptRYTemp2; converting OptRYTemp2 into a character string describing the leg action;

[0024] S205, using the five elements of ModelList to identify OptRYInput, and storing the identification result in the third temporary identification tag OptRYTemp3; converting OptRYTemp3 into a character string describing the posture action;

[0025] S206, OptRYOutput = OptRYOutput + "Take OptRYTemp3 as posture" + "Arm movement is OptRYTemp1" + "Leg movement is OptRYTemp2"; where "+" is used to combine character strings;

[0026] S207, return OptRYOutput as the result of OptRY.

[0027] Furthermore, the specific process of step S3 is as follows:

[0028] S301, establishing a distance group personnel behavior description operator OptQT, where the input of OptQT is the scene image OptQTInput;

[0029] S302, establish a language description of the behavior of the group of people with distance OptRYOutput = string "The behavior of all personnel at the live working site is as follows;";

[0030] S303, identifying the personnel list OptQTList = using one model in ModelList to identify all personnel in OptQTInput, and obtaining the identification results of all personnel on site;

[0031] S304, with distance group personnel behavior description operator counter OptQTCounter = 1;

[0032] S305, the first temporary variable OptQTTemp1 of the distance group personnel behavior description operator = taking out the OptQTCounterth element of OptQTList;

[0033] S306, the second temporary variable OptQTTemp2 with distance group personnel behavior description operator = intercepting a sub-image from OptQTInput in the area described by OptQTTemp1;

[0034] S307, the third temporary variable OptQTTemp3 of the distance group personnel behavior description operator = the Y coordinate of the area described by OptQTTemp1 / the height of OptQTInput;

[0035] S308, the fourth temporary variable OptQTTemp4 of the distance group personnel behavior description operator is calculated by using OptRY, and the value of OptQTInput of OptRY is OptQTTemp2;

[0036] S309, setting the value of the fifth temporary variable OptQTTemp5 of the distance group personnel behavior description operator, if OptQTTemp3 is greater than 1 / 2, then OptQTTemp5 = "close distance", if OptQTTemp3 is less than or equal to 1 / 2, then OptQTTemp5 = "far distance";

[0037] S310, OptRYOutput = OptRYOutput + "OptQTCounterth person" + OptQTTemp5 + OptQTTemp4 + ";"; wherein "+" indicates character string combination;

[0038] S311, set OptQTCounter=OptQTCounter+1;

[0039] S312, if OptQTCounter is less than or equal to the number of elements in OptQTList, go to S305, otherwise go to S313;

[0040] S313, return OptRYOutput as the result of OptQT.

[0041] Furthermore, the specific process of step S4 is as follows:

[0042] S401, establishing a multi-model live working site automation natural language description operator OptDD, where the input of OptDD is a picture of the live working site OptDDInput;

[0043] S402, establish a natural language description of the live working site OptDDOutput = string "The description of the live working site is:";

[0044] S403, the first temporary variable OptDDTemp1 of the multi-model live working field automation natural language description operator = calculation is performed using OptQT, and OptQTInput of OptQT is OptDDInput;

[0045] S404, the second temporary variable OptDDTemp2 of the multi-model live working field automation natural language description operator = using the second element of ModelList to identify all devices in OptDDInput, and converting all identified device names into a list;

[0046] S405, the third temporary variable OptDDTemp3 of the multi-model live working site automation natural language description operator = "The live working site includes the following equipment: " + tostr (OptDDTemp2), where "+" is string combination, and tostr (·) is to convert each element of a list into a string separated by ",";

[0047] S406, OptDDOutput=OptDDOutput+OptDDTemp3+OptDDTemp1.

[0048] Furthermore, the specific process of step S5 is as follows:

[0049] S501, input the electrical work site picture DDImage;

[0050] S502, the result of the automatic natural language description of the on-site behavior of live working ZDDesc = calculation using OptDD, and the OptDDInput of OptDD is DDImage;

[0051] S503, outputting ZDDesc as a result.

[0052] The beneficial effects of the present invention are:

[0053] The description process of the method of the present invention does not rely on a specific single model, but uses multiple models to describe the behavior of construction workers in live working scenarios from the aspects of equipment, personnel, and actions of different parts, thereby achieving a more stable and comprehensive description capability. Even when facing the changing environment of the live working site, the method of the present invention can correctly describe the behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a multi-model-based natural language description method for live working on-site behavior automation of the present invention. DETAILED DESCRIPTION

[0055] Specific implementation method 1: Combination Figure 1 This embodiment describes a multi-model-based automatic natural language description method for on-site behaviors of live working, and the method specifically includes the following steps:

[0056] Step S1, inputting a personnel detection depth model ModelR, a device description model ModelM, an arm action recognition model ModelH, a leg action recognition model ModelL, and a posture recognition model ModelZ; establishing a multi-model list ModelList;

[0057] Step S2, establishing a single person description operator OptRY, where the input of OptRY is the person image OptRYInput, and the output is the person's behavior language description OptRYOutput;

[0058] Step S3, establishing an operator OptQT for describing the behavior of a group of people with distance, wherein the input of OptQT is the scene image OptQTInput, and the output of the OptQT operator is the language description of the behavior of a group of people with distance OptQTOutput;

[0059] Step S4, establishing a multi-model live working site automation natural language description operator OptDD, the input of OptDD is the picture OptDDInput of the live working site, and the output of OptDD is the natural language description OptDDOutput of the live working site;

[0060] Step S5, input the electrical work site picture DDImage, and use OptDD to obtain the automatic natural language description result ZDDesc of the live work site behavior.

[0061] Specific implementation method 2: This implementation method is different from the specific implementation method 1 in that the specific process of step S1 is as follows:

[0062] S101, inputting a person detection deep model ModelR, which is a recognition model based on the Yolo model and can mark the area of ​​a person in a picture or video with a box;

[0063] S102, inputting a device description model ModelM, where ModelM is a recognition model based on a Yolo model, and a device area and a device name can be marked with a box;

[0064] S103, inputting the arm action recognition model ModelH, where ModelH is a two-stream convolutional neural network that can be used to recognize the arm posture of a specific person;

[0065] S104, inputting a leg action recognition model ModelL, where ModelL is a two-stream convolutional neural network that can be used to recognize the leg posture of a specific person;

[0066] S105, inputting a posture recognition model ModelZ, where ModelZ is a two-stream convolutional neural network that can be used to recognize the body posture of a specific person;

[0067] S106, establish a multi-model list ModelList = [ModelR, ModelM, ModelH, ModelL, ModelZ].

[0068] The other steps and parameters are the same as those in the first embodiment.

[0069] Specific implementation method three: This implementation method is different from specific implementation methods one or two in that the specific process of step S2 is as follows:

[0070] S201, establishing a single person description operator OptRY, where the OptRY input is the person image OptRYInput;

[0071] S202, create a behavior language description of personnel OptRYOutput = character string "construction personnel";

[0072] S203, using the three elements of ModelList to identify OptRYInput, and storing the identification result in the first temporary identification tag OptRYTemp1; converting OptRYTemp1 into a character string describing the arm action;

[0073] S204, using the four elements of ModelList to identify OptRYInput, and storing the identification result in the second temporary identification tag OptRYTemp2; converting OptRYTemp2 into a character string describing the leg action;

[0074] S205, using the five elements of ModelList to identify OptRYInput, and storing the identification result in the third temporary identification tag OptRYTemp3; converting OptRYTemp3 into a character string describing the posture action;

[0075] S206, OptRYOutput = OptRYOutput + "with OptRYTemp3 as posture" + "its arm movement is OptRYTemp1" + "its leg movement is OptRYTemp2"; wherein "+" is used to combine character strings;

[0076] S207, return OptRYOutput as the result of OptRY.

[0077] The other steps and parameters are the same as those in the first or second embodiment.

[0078] Specific implementation method 4: This implementation method is different from any one of the specific implementation methods 1 to 3 in that the specific process of step S3 is as follows:

[0079] S301, establishing a distance group personnel behavior description operator OptQT, where the input of OptQT is the scene image OptQTInput;

[0080] S302, establish a language description of the behavior of the group of people with distance OptRYOutput = string "The behavior of all personnel at the live working site is as follows;";

[0081] S303, identifying the personnel list OptQTList = using one model in ModelList to identify all personnel in OptQTInput, and obtaining the identification results of all personnel on site;

[0082] S304, with distance group personnel behavior description operator counter OptQTCounter = 1;

[0083] S305, the first temporary variable OptQTTemp1 of the distance group personnel behavior description operator = taking out the OptQTCounterth element of OptQTList;

[0084] S306, the second temporary variable OptQTTemp2 with distance group personnel behavior description operator = intercepting a sub-image from OptQTInput in the area described by OptQTTemp1;

[0085] S307, the third temporary variable OptQTTemp3 of the distance group personnel behavior description operator = the Y coordinate of the area described by OptQTTemp1 / the height of OptQTInput;

[0086] S308, the fourth temporary variable OptQTTemp4 of the distance group personnel behavior description operator is calculated by using OptRY, and the value (input) of OptQTInput of OptRY is OptQTTemp2;

[0087] S309, setting the value of the fifth temporary variable OptQTTemp5 of the distance group personnel behavior description operator, if OptQTTemp3 is greater than 1 / 2, then OptQTTemp5 = "close distance", if OptQTTemp3 is less than or equal to 1 / 2, then OptQTTemp5 = "far distance";

[0088] S310, OptRYOutput = OptRYOutput + "OptQTCounterth person" + OptQTTemp5 + OptQTTemp4 + ";"; wherein "+" indicates character string combination;

[0089] S311, set OptQTCounter=OptQTCounter+1;

[0090] S312, if OptQTCounter is less than or equal to the number of elements in OptQTList, go to S305, otherwise go to S313;

[0091] S313, return OptRYOutput as the result of OptQT.

[0092] The other steps and parameters are the same as those in Specific Embodiments 1 to 3.

[0093] Specific implementation mode 5: This implementation mode is different from the specific implementation modes 1 to 4 in that the specific process of step S4 is as follows:

[0094] S401, establishing a multi-model live working site automation natural language description operator OptDD, where the input of OptDD is a picture of the live working site OptDDInput;

[0095] S402, establish a natural language description of the live working site OptDDOutput = string "The description of the live working site is:";

[0096] S403, the first temporary variable OptDDTemp1 of the multi-model live working field automation natural language description operator = calculation is performed using OptQT, and OptQTInput (input) of OptQT is OptDDInput;

[0097] S404, the second temporary variable OptDDTemp2 of the multi-model live working field automation natural language description operator = using the second element of ModelList to identify all devices in OptDDInput, and converting all identified device names into a list;

[0098] S405, the third temporary variable OptDDTemp3 of the multi-model live working site automation natural language description operator = "The live working site includes the following equipment: " + tostr (OptDDTemp2), where "+" is string combination, and tostr (·) is to convert each element of a list into a string separated by ",";

[0099] S406, OptDDOutput=OptDDOutput+OptDDTemp3+OptDDTemp1.

[0100] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0101] Specific implementation method 6: This implementation method is different from the specific implementation methods 1 to 5 in that the specific process of step S5 is as follows:

[0102] S501, input the electrical work site picture DDImage;

[0103] S502, live working on-site behavior automatic natural language description result ZDDesc=calculated using OptDD, OptDDInput (input) of OptDD is DDImage;

[0104] S503, outputting ZDDesc as a result.

[0105] The other steps and parameters are the same as those in Specific Implementation Methods 1 to 5.

[0106] The above calculation examples of the present invention are only used to explain the calculation model and calculation process of the present invention in detail, and are not intended to limit the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-model-based method for automatic natural language description of on-site behaviors of live working, characterized in that: The method specifically comprises the following steps: Step S1, inputting a personnel detection depth model ModelR, a device description model ModelM, an arm action recognition model ModelH, a leg action recognition model ModelL, and a posture recognition model ModelZ; establishing a multi-model list ModelList; Step S2, establishing a single person description operator OptRY, where the input of OptRY is the person image OptRYInput, and the output is the person's behavior language description OptRYOutput; Step S3, establishing an operator OptQT for describing the behavior of a group of people with distance, wherein the input of OptQT is the scene image OptQTInput, and the output of the OptQT operator is the language description of the behavior of a group of people with distance OptQTOutput; Step S4, establishing a multi-model live working site automation natural language description operator OptDD, the input of OptDD is the picture OptDDInput of the live working site, and the output of OptDD is the natural language description OptDDOutput of the live working site; Step S5, input the electrical work site picture DDImage, and use OptDD to obtain the automatic natural language description result ZDDesc of the live work site behavior.

2. The method for automatic natural language description of on-site behavior of live working based on multiple models according to claim 1 is characterized in that: The specific process of step S1 is as follows: S101, inputting a person detection deep model ModelR, where ModelR is a recognition model based on the Yolo model, and uses a box to mark the area where a person is in a picture or video; S102, inputting a device description model ModelM, where ModelM is a recognition model based on a Yolo model, and marking a device area and a device name with a box; S103, inputting the arm action recognition model ModelH, where ModelH is a two-stream convolutional neural network used to recognize the arm posture of a person; S104, inputting a leg action recognition model ModelL, where ModelL is a two-stream convolutional neural network used to recognize the posture of a person's legs; S105, inputting a posture recognition model ModelZ, where ModelZ is a two-stream convolutional neural network used to recognize the body posture of a specific person; S106, establish a multi-model list ModelList = [ModelR, ModelM, ModelH, ModelL, ModelZ].

3. The method for automatic natural language description of on-site behavior of live working based on multiple models according to claim 2 is characterized in that: The specific process of step S2 is: S201, establishing a single person description operator OptRY, where the OptRY input is the person image OptRYInput; S202, create a behavior language description of personnel OptRYOutput = character string "construction personnel"; S203, using the three elements of ModelList to identify OptRYInput, and storing the identification result in the first temporary identification tag OptRYTemp1; converting OptRYTemp1 into a character string describing the arm action; S204, using the four elements of ModelList to identify OptRYInput, and storing the identification result in the second temporary identification tag OptRYTemp2; converting OptRYTemp2 into a character string describing the leg action; S205, using the five elements of ModelList to identify OptRYInput, and storing the identification result in the third temporary identification tag OptRYTemp3; converting OptRYTemp3 into a character string describing the posture action; S206, OptRYOutput = OptRYOutput + "Take OptRYTemp3 as posture" + "Arm movement is OptRYTemp1" + "Leg movement is OptRYTemp2"; where "+" is for character string combination; S207, return OptRYOutput as the result of OptRY.

4. The method for automatic natural language description of on-site behavior of live working based on multiple models according to claim 3 is characterized in that: The specific process of step S3 is: S301, establishing a distance group personnel behavior description operator OptQT, where the input of OptQT is the scene image OptQTInput; S302, establish a language description of the behavior of the group of people with distance OptRYOutput = string "The behavior of all personnel at the live working site is as follows;"; S303, identifying the personnel list OptQTList = using one model in ModelList to identify all personnel in OptQTInput, and obtaining the identification results of all personnel on site; S304, with distance group personnel behavior description operator counter OptQTCounter = 1; S305, the first temporary variable OptQTTemp1 of the distance group personnel behavior description operator = taking out the OptQTCounterth element of OptQTList; S306, the second temporary variable OptQTTemp2 with distance group personnel behavior description operator = intercepting a sub-image from OptQTInput in the area described by OptQTTemp1; S307, the third temporary variable OptQTTemp3 of the distance group personnel behavior description operator = the Y coordinate of the area described by OptQTTemp1 / the height of OptQTInput; S308, the fourth temporary variable OptQTTemp4 of the distance group personnel behavior description operator is calculated by using OptRY, and the value of OptQTInput of OptRY is OptQTTemp2; S309, setting the value of the fifth temporary variable OptQTTemp5 of the distance group personnel behavior description operator, if OptQTTemp3 is greater than 1 / 2, then OptQTTemp5 = "close distance", if OptQTTemp3 is less than or equal to 1 / 2, then OptQTTemp5 = "far distance"; S310, OptRYOutput = OptRYOutput + "OptQTCounterth person" + OptQTTemp5 + OptQTTemp4 + ";"; "+" indicates character string combination; S311, set OptQTCounter=OptQTCounter+1; S312, if OptQTCounter is less than or equal to the number of elements in OptQTList, go to S305, otherwise go to S313; S313, return OptRYOutput as the result of OptQT.

5. The method for automatic natural language description of on-site behavior of live working based on multiple models according to claim 4 is characterized in that: The specific process of step S4 is as follows: S401, establishing a multi-model live working site automation natural language description operator OptDD, where the input of OptDD is a picture of the live working site OptDDInput; S402, establish a natural language description of the live working site OptDDOutput = string "the description of the live working site is:"; S403, the first temporary variable OptDDTemp1 of the multi-model live working field automation natural language description operator = calculation is performed using OptQT, and OptQTInput of OptQT is OptDDInput; S404, the second temporary variable OptDDTemp2 of the multi-model live working field automation natural language description operator = using the second element of ModelList to identify all devices in OptDDInput, and converting all identified device names into a list; S405, the third temporary variable OptDDTemp3 of the multi-model live working site automation natural language description operator = "The live working site includes the following equipment: " + tostr (OptDDTemp2), where "+" is string combination, and tostr (·) is to convert each element of a list into a string separated by ", "; S406, OptDDOutput=OptDDOutput+OptDDTemp3+OptDDTemp1.

6. The method for automatic natural language description of live working site behavior based on multiple models according to claim 5 is characterized in that: The specific process of step S5 is as follows: S501, input the electrical work site picture DDImage; S502, the result of the automatic natural language description of the on-site behavior of live working ZDDesc = calculation using OptDD, and the OptDDInput of OptDD is DDImage; S503, outputting ZDDesc as a result.

Citation Information

Patent Citations

  • Vehicle device and method for determining vehicle-distance

    CN101337536A

  • Information processing device, movement device, method, and program

    CN112119282A

  • Multi-model comprehensive remote sensing image scene description method

    CN113610025A

  • Depth measurement method, depth measurement equipment and storage medium

    CN114419143A

  • Video description method and system based on multi-modal interaction and time semantic aggregation

    CN115512265A