VR annotation and roaming road network automatic generation method and device

Abstract

The invention discloses a VR annotation and roaming road network automatic generation method and device. The method comprises the following steps of S1, image preprocessing, S2, feature recognition training, S3, annotation generation and S4, roaming road network construction. The method comprises the following steps: S1, performing image enhancement and de-noising processing on an image acquired by image acquisition equipment, specifically comprising the following two steps; the invention relates to the technical field of VR road network planning methods and equipment, and provides an automatic generation method and equipment for a VR annotation and roaming road network, which constructs a VR panoramic annotation and road network automatic generation system by fusing deep learning and intelligent hardware design. The innovation of the method is embodied in that a dynamic linkage image enhancement, multi-modal feature recognition and closed-loop verification mechanism is adopted at an algorithm end, and a hardware end is integrated with an annular motion holder, a modular design and a quick release structure, so that the full-process automation from panoramic data acquisition and marking to navigation road network generation is realized, and the processing efficiency and the system reliability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to a VR road network planning method and device, in particular to a VR labeling and roaming road network automatic generation method and device. BACKGROUND

[0002] VR, namely virtual reality technology, is a development result based on computer technology combined with three-dimensional graphics technology, sensor technology and servo technology and the like, and the basic structure is determined by the development and application of computer technology. With the development of VR technology in recent years, more and more VR devices are applied to automobiles, processing, games and life and the like, and it is a very promising development direction.

[0003] In the application process of modern VR technology, an important development direction of the VR device is virtual navigation and navigation of space. Real map navigation, tourism navigation and space display are realized by using the VR device, which has a technical advantage that conventional devices do not have. In today's rapidly changing city planning, the existing VR device needs manual addition and configuration for panoramic map labeling and roaming road network. When the number of panoramic maps reaches a certain order of magnitude, not only the error probability is increased, but also the labor cost and time cost are doubled.

[0004] For example, the map fusion method and device based on multiple frames of images disclosed in CN114764847A disclose that the similarity between the key frames in the first initial map constructed by the first terminal and the key frames in the second initial map constructed by the second terminal can reduce the interference of the local similar regions in the current key frames in different initial maps, improve the accuracy of loop detection, and further improve the fusion precision of the map. Although the matching and orientation of the map can be performed by comparing the key frames, the unrecorded map information cannot be input and marked in the VR panoramic application process. With the development of artificial intelligence technology and deep learning technology at the present stage, if the development achievements of machine deep learning technology are used for VR panoramic map labeling, the labor cost and time input can be greatly reduced, and the efficiency of panoramic map labeling and road network generation can be improved.

[0005] At the same time, in the labeling process of the panoramic VR shooting device at the present stage, the VR picture materials are usually collected by manually holding the camera by the artificial, and the cooperation of the holder is needed if the shooting angle is not good in the use process, so that a better shooting effect can be obtained. The device demand is complex, the structure is relatively heavy, the user is difficult to hold for a long time, the shooting experience of the user is affected, and the material collection efficiency of the VR panoramic map is reduced. In view of this, in-depth research is conducted on the above problems, and the present case is generated. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a name, solving the problems of the prior art.

[0007] To achieve the above object, the present application is implemented by the following technical solutions: a VR labeling and roaming road network automatic generation method, comprising the following steps: step S1: image preprocessing, step S2: feature recognition training, step S3: label generation, and step S4: roaming road network construction.

[0008] Step S1: image enhancement and denoising processing are performed on the images collected by the image collection device respectively, specifically including the following two steps.

[0009] Step S1.1 image enhancement: histogram equalization and contrast stretching method are used to improve the overall visual quality of the panoramic image, making the image details clearer, and providing a good foundation for subsequent feature extraction.

[0010] Step S1.2 denoising processing: Gaussian filter and median filter algorithm are used to remove noise interference in the panoramic image, ensuring the accuracy and stability of the image data.

[0011] Step S2: training of image recognition model based on deep learning, detecting the target through the image recognition model, outputting the new type of target, and further refining the recognition of the scene by using the semantic segmentation model, specifically including the following two steps.

[0012] Step S2.1 target detection based on deep learning: using a pre-trained deep learning model to detect and recognize buildings, roads, landmarks, and traffic sign key elements in the panoramic image, outputting the class, position, and size information of the target.

[0013] Step S2.1 semantic segmentation: through the semantic segmentation model, different objects and regions in the panoramic image are classified at the pixel level, further refining the understanding of the scene, assisting in labeling and road network generation.

[0014] Step S3: according to the results of target detection and semantic segmentation, adding text labels to the recognized elements, the label content including name, attribute, introduction related information, the font, color and position of the label being optimized and designed to ensure clear and easy to read without affecting the beauty of the panoramic image and user browsing experience.

[0015] Step S4: according to the road labeling confidence and the machine deep learning results of the panoramic image, extracting the road, summarizing the information of the crossroads, T-shaped intersection and other road connection node positions to determine the nodes, combining the road navigation information with the panoramic map information to determine the road path, specifically including the following three steps.

[0016] Step S4.1 Road extraction and refinement: Extract the road region from the semantic segmentation result, and through morphological processing and edge detection algorithm, refine and vectorize the road to determine the center line and boundary of the road;

[0017] Step S4.2 Node determination: According to the intersection and branch point position information of the road, determine the nodes in the road network, and define the number and attributes of the nodes, such as intersection type and traffic rules;

[0018] Step S4.3 Path connection: According to the actual connection relationship of the road, connect each node to form a complete roaming road network, and assign length, width and slope attribute information to each road segment to realize path planning and navigation function.

[0019] A VR marking and roaming road network image acquisition device, comprising a helmet, the helmet is a semi-enclosed structure, the top of the helmet is provided with an image acquisition pan-tilt structure;

[0020] The image acquisition pan-tilt structure is equipped with at least one pair of image acquisition camera structures, and the pair of image acquisition camera structures are symmetrically arranged and connected with each other. The image acquisition pan-tilt structure is also integrated with a sensor structure, which is composed of an infrared sensor, an encoder and a wireless communication module.

[0021] The image acquisition pan-tilt structure comprises a ring-shaped guide rail, which is assembled on the helmet. The ring-shaped guide rail is provided with a ring motion controller. The ring-shaped guide rail is equipped with a ring motion sliding block. The top surface of the ring motion sliding rail is a slope structure. The ring motion sliding block is connected with the ring motion controller. The ring motion controller controls the ring motion sliding block to move in a ring direction. The ring motion sliding block is connected with the image acquisition camera structure. The ring-shaped sliding rail is connected with the image acquisition pan-tilt.

[0022] The image acquisition camera structure comprises a pitch control assembly, which is assembled on the ring motion sliding block. The pitch control assembly is connected with a rotation controller. The pitch control assembly is connected with a camera.

[0023] The top of the helmet is provided with a fixing seat, and the fixing seat is connected with a quick release fixing structure.

[0024] The quick release fixing structure is equipped with a battery module, which supplies power to the ring motion controller, the pitch control assembly and the camera.

[0025] The ring-shaped guide rail is a ring-shaped track of plastic PVC material. One side of the ring-shaped guide rail is provided with a plurality of buckles. The outside of the helmet is provided with a plurality of clamping grooves and a plurality of buckles.

[0026] The circumferential motion controller includes a circumferential motion control box, a control motor is provided on one side of the circumferential motion control box, a driving end of the control motor is connected to a gear set, the gear set is provided with a circumferential motion ring, a connecting block extends from one side of the circumferential motion ring, the connecting block is connected to the circumferential motion slider, the circumferential motion slider is matched with the circumferential motion guide rail, and the number of the circumferential motion sliders matches the number of image acquisition camera structures.

[0027] The gear set is connected to the driving end of the control motor through a reduction gear. An output gear is meshed on one side of the reduction gear. A first bevel gear is coaxially arranged on one side of the output gear. A second bevel gear is meshed on one side of the first bevel gear. A toggle gear is coaxially arranged on one side of the second bevel gear. The toggle gear is meshed with the annular ring.

[0028] The pitch control assembly includes a camera bracket, which is mounted on a circular slider. A pitch control motor is provided on the camera bracket. A driving end of the pitch control motor is connected to a pitch axis, which is connected to the camera.

[0029] An inflatable liner is provided inside the helmet, an inflation control valve extends from one side of the inflatable liner, the inflation control valve is used to inflate and deflate the inflatable liner, and a pressure valve is provided on the other side of the inflatable liner.

[0030] The camera is an automatic zoom camera, a protective cover is provided outside the camera, and an axle seat is provided on one side of the camera to be connected with the pitch axis.

[0031] The fixing seat is a rectangular plate. The outer ring of the fixing seat is provided with a plurality of connecting belts, which are connected to the circular movable guide rail. The plurality of connecting belts and the circular movable guide rail form an integrally formed structure.

[0032] The quick-release fixing structure includes a disassembly and assembly seat, a guide slide groove is provided on the disassembly and assembly seat, a disassembly and assembly guide rail is provided on the fixing seat, the guide slide groove is assembled on the disassembly and assembly guide rail, the disassembly and assembly seat is connected to the battery module, a warning light is provided on one side of the disassembly and assembly seat, at least one pair of elastic positioning rods are provided on the inner side of the guide slide groove, and a pair of positioning slots are provided on the disassembly and assembly guide rail corresponding to the pair of elastic positioning rods.

[0033] Beneficial effects

[0034] The application provides a VR labeling and roaming road network automatic generation method and device. The VR labeling and roaming road network automatic generation method and device have the following advantages: the VR panoramic labeling and road network automatic generation system is constructed by fusing deep learning and intelligent hardware design. The innovation is reflected in that the algorithm end adopts dynamic linkage image enhancement, multi-modal feature recognition and closed-loop verification mechanism, the hardware end integrates ring dynamic holder, modular design and quick release structure, realizes full-process automation of panoramic data acquisition, labeling and navigation road network generation, significantly improves processing efficiency and system reliability, and provides a high-precision and low-cost solution for VR geographic information construction. In addition to the above advantages, the application also has the following advantages.

[0035] The target detection and semantic segmentation model is combined with the camera pose data to realize three-dimensional space binding of the labeling information, and the labeling position is projected in real time through the helmet attitude sensor to eliminate visual drift error.

[0036] The image enhancement module is dynamically linked with the acquisition device, the parameters are adaptively adjusted in real time through real-time environment perception, the uniform enhancement of road texture under different light conditions is ensured, and the noise suppression effect is improved.

[0037] The road vectorization data is linked with the ring dynamic controller to guide automatic retaking, the path attribute is verified in reverse through the SLAM track, the cumulative error of mapping is reduced to centimeter level, and the road network topology accuracy is improved.

[0038] The ring guide rail and the pressure sensor cooperate to realize mechanical overload protection, the quick release structure integrates super capacitor to maintain power-off memory, and the hydrophobic protective cover and the thermal deformation compensation buckle are matched to improve the stability of the device.

[0039] The inflatable lining combined with the pearl cotton layer adapts to the head shape, reduces image blur caused by device shaking, reduces user fatigue, actively cools by using ring air flow, and guarantees long-time operation performance. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 FIG. 1 is a front view of a VR labeling and roaming road network image acquisition device according to the application.

[0041] Figure 2 FIG. 2 is a first perspective view of a VR labeling and roaming road network image acquisition device according to the application.

[0042] Figure 3 FIG. 3 is a second perspective view of a VR labeling and roaming road network image acquisition device according to the application.

[0043] Figure 4 FIG. 4 is a third perspective view of a VR labeling and roaming road network image acquisition device according to the application.

[0044] Figure 5 This is a fourth stereoscopic structural schematic diagram of a VR annotation and roaming road network image acquisition device described in the present invention.

[0045] Figure 6 This is a fifth stereoscopic structural diagram of a VR annotation and roaming road network image acquisition device described in the present invention.

[0046] Figure 7 This is a rear-view structural schematic diagram of a VR annotation and roaming road network image acquisition device described in the present invention.

[0047] Figure 8 This is a schematic diagram of the top view of the structure of a VR annotation and roaming road network image acquisition device described in the present invention.

[0048] Figure 9 This is a schematic diagram of the partial three-dimensional structure of a VR annotation and roaming road network image acquisition device described in the present invention.

[0049] Figure 10 This is a flow chart of a method for automatically generating VR annotation and roaming road networks according to the present invention.

[0050] Figure 1: Helmet; 2: Image acquisition gimbal structure; 3: Image acquisition camera structure; 4: Sensor structure; 5: Quick-release fixing structure; 6: Battery module; 7: Inflatable liner; 11: Fixing seat; 12: Slot; 13: Buckle; 14: Connecting belt; 21: Annular guide rail; 22: Annular motion controller; 23: Annular motion slider; 31: Pitch control assembly; 32: Camera; 51: Disassembly and assembly seat; 52: Guide slide; 53: Disassembly and assembly rail; 54: Warning light ; 55. Elastic positioning rod; 56. Positioning slot; 71. Inflation control valve; 72. Pressure valve; 221. Circular motion control box; 222. Control motor; 223. Gear set; 224. Circular motion ring; 225. Connecting block; 311. Camera bracket; 312. Pitch control motor; 313. Pitch axis; 314. Axle seat; 315. Protective cover; 2231. First bevel gear; 2232. Second bevel gear; 2233. Toggle gear. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] See also Figures 1-10 The present invention provides an implementation scheme: Example 1: According to the attached instructionsFigure 10 It can be seen that the present application discloses a method for automatically generating VR annotation and roaming road network, comprising the following steps: step S1: image preprocessing, step S2: feature recognition training, step S3: annotation generation, and step S4: roaming road network construction;

[0053] Step S1: performing image enhancement and denoising processing on the images captured by the image acquisition device, specifically including the following two steps;

[0054] Step S1.1 Image Enhancement: Utilize histogram equalization and contrast stretching to improve the overall visual quality of the panorama, making image details clearer and providing a good foundation for subsequent feature extraction. The image enhancement module is linked to the automatic exposure control of the image acquisition device. By analyzing the histogram data transmitted in real time by the camera, it dynamically adjusts the contrast stretching parameters to ensure uniform enhancement of road texture under different lighting conditions.

[0055] Step S1.2: Denoising: Gaussian filtering and median filtering algorithms are used to remove noise interference from the panoramic image to ensure the accuracy and stability of the image data. The denoising algorithm works in conjunction with the sensor noise reduction function of the image acquisition device. By reading the ambient light intensity data collected by the infrared sensor, the filter kernel size is adaptively selected. In low-light scenes, median filtering is preferentially enabled to protect edge details.

[0056] Step S2: Train an image recognition model based on deep learning, detect the target through the image recognition model, output a new model of the target, and then use the semantic segmentation model to further refine the recognition of the scene. Specifically, it includes the following two steps;

[0057] Step S2.1: Deep learning-based object detection: Using a pre-trained deep learning model, the module detects and identifies key elements in the panoramic image, including buildings, roads, landmarks, and traffic signs. It then outputs the target's category, location, and size. The target detection module uses an encoder to obtain camera pose data and binds the detection results to spatial coordinates, providing a 3D positioning benchmark for subsequent road network construction.

[0058] Step S2.1: Semantic Segmentation: Using a semantic segmentation model, different objects and regions in the panorama are classified at the pixel level, further refining the understanding of the scene and assisting with annotation and road network generation. The road mask data output by the semantic segmentation network is directly transmitted to the environmental controller, which dynamically adjusts the camera acquisition angle to capture additional images of occluded areas.

[0059] Step S3: According to the results of target detection and semantic segmentation, add text labels to the recognized elements of each type, including name, attribute, and introduction of related information. The font, color, and position of the labels are optimized to ensure clear readability and do not affect the aesthetics of the panoramic map and user browsing experience. The labeling engine is linked with the head pose sensor to project the label position in real time according to the user's head movement, maintaining the visual stability of the spatial label.

[0060] Step S4: According to the road labeling confidence and the machine deep learning results of the panoramic picture, extract the road, summarize the information of the crossroads, T-shaped intersections, and other road connection node positions to determine the nodes, combine the road navigation information with the panoramic map information, and determine the road path, which includes the following three steps.

[0061] Step S4.1 Road extraction and refinement: Extract the road area from the semantic segmentation results, refine and vectorize the road through morphological processing and edge detection algorithms, determine the center line and boundary of the road, and return the road vectorization data to the ring motion controller through the wireless communication module to guide the camera to automatically scan and shoot along the road center line trajectory.

[0062] Step S4.2 Node determination: Determine the nodes in the road network according to the intersection and branch point position information of the road, and define the node number and attribute, such as intersection type and traffic rules. The node attribute data is fused with the camera 32 GPS module (optional) to establish the topological relationship in the absolute geographic coordinate system.

[0063] Step S4.3 Path connection: Connect each node according to the actual connectivity of the road to form a complete roaming road network, and assign length, width, and slope attribute information to each road segment to realize path planning and navigation functions. The path attribute data is recorded through the encoder to verify the camera 32 motion trajectory in reverse, eliminating the cumulative error of SLAM mapping.

[0064] Embodiment 2: According to the description Figures 1-9 It can be seen that, in order to cooperate with the application of the above method, the present application discloses a VR labeling and roaming road network image acquisition device, which comprises a helmet 1. The helmet 1 is a semi-enclosed structure, and an image acquisition gimbal structure 2 is arranged around the top of the helmet 1.

[0065] According to the description Figures 1-9It can be seen that the image acquisition holder structure 2 is provided with at least one pair of image acquisition camera structures 3, and the pair of image acquisition camera structures 3 are symmetrically arranged and connected with each other, and the real-time depth map is generated through binocular parallax calculation, and the symmetric arrangement of the pair of image acquisition camera structures 3 effectively ensures that the helmet 1 counterweight is not tilted during the image acquisition process. The image acquisition holder structure 2 is also integrated with a sensor structure 4, which is composed of an infrared sensor, an encoder and a wireless communication module. When the distance of the detected obstacle is less than 1 meter, the infrared sensor and the camera 32 form a closed loop control, and the zooming and retracting protection mechanism of the camera 32 is automatically triggered;

[0066] According to the description attached Figures 1-9 It can be seen that the image acquisition holder structure 2 includes a ring-shaped guide rail 21, which is arranged on the helmet 1. The ring-shaped guide rail 21 is provided with a ring motion controller 22, and the ring-shaped guide rail 21 is provided with a ring motion sliding block 23. The top surface of the ring motion sliding block 23 is a slope structure, and the slope slope is matched with the field of view angle of the camera 32, so as to ensure that the adjacent camera 32 images have a 15% overlapping area. The ring motion sliding block 23 is connected with the ring motion controller 22, and the ring motion controller 22 controls the ring motion sliding block 23 to move in a ring direction. The ring motion sliding block 23 is connected with the image acquisition camera structure 3, and the ring-shaped sliding rail is connected with the image acquisition holder. The ring motion sliding block 23 is provided with a pressure sensor, which can monitor the contact stress of the guide rail in real time, and automatically reduce the speed when the pressure exceeds 2N for protection.

[0067] According to the description attached Figures 1-9 It can be seen that the image acquisition camera structure 3 includes a pitch control assembly 31, which is arranged on the ring motion sliding block 23. The pitch control assembly 31 is connected with a rotation controller, which receives the position instruction of the ring motion controller 22 through the CAN bus. The pitch control assembly 31 is connected with the camera 32, and the pitch shaft 313 is provided with a torque limiter. When the external impact exceeds 0.5N·m, the torque limiter is automatically tripped to prevent damage to the mechanical structure.

[0068] The top of the helmet 1 is provided with a fixing seat 11, and the fixing seat 11 is connected with a quick-release fixing structure 5. The fixing seat 11 is internally integrated with a heat dissipation air duct, which actively cools the camera 32 CMOS through the airflow generated by the movement of the ring motion sliding block 23.

[0069] According to the description attached Figures 1-9 It can be seen that the quick-release fixing structure 5 is provided with a battery module 6, which supplies power to the ring motion controller 22, the pitch control assembly 31 and the camera 32. The battery module 6 is connected with the system through the sliding rail contact, and automatically cuts off the power supply and enables the super capacitor to maintain the position memory when disassembled.

[0070] According to the description attached Figures 1-9It can be known that the annular guide rail 21 is a plastic PVC material annular structure track, and a plurality of buckles 13 are arranged on one side of the annular guide rail 21. The helmet 1 is externally annularly provided with a plurality of clamping grooves 12 and a plurality of buckles 13 for clamping connection. The buckle 13 is internally embedded with a shape memory alloy, which automatically increases the locking force when the temperature exceeds 60℃, preventing loosening caused by thermal deformation;

[0071] According to the description attached Figures 1-9 It can be known that the annular motion controller 22 comprises an annular motion control box 221, and a control motor 222 is arranged on one side of the annular motion control box 221. The driving end of the control motor 222 is connected with a gear set 223. The gear set 223 is provided with an annular motion sleeve 224. The annular motion sleeve 224 extends out a connecting block 225 on one side. The connecting block 225 is connected with the annular motion sliding block 23. The annular motion sliding block 23 is matched with the annular guide rail. The number of the annular motion sliding block 23 is matched with the number of the image acquisition camera structure 3. The control motor 222 is installed through the annular motion control box 221. The control motor 222 drives the gear set 223 to rotate. Then the gear set 223 drives the annular motion sleeve 224. The connecting block 225 and the annular motion sliding block 23 are synchronously rotated under the drive of the annular motion sleeve 224. Thus, the rotating direction of the image acquisition camera structure 3 is controlled. The purpose of controlling the camera direction is achieved.

[0072] According to the description attached Figures 1-9 It can be known that the gear set 223 is connected with the driving end of the control motor 222 through the first bevel gear 2231. The first bevel gear 2231 is meshingly provided with the second bevel gear 2232 on one side. The second bevel gear 2232 is coaxially provided with the dial gear 2233 on one side. The dial gear 2233 is meshed with the annular motion sleeve 224. The first bevel gear 2231 drives the second bevel gear 2232 to rotate. Then the second bevel gear 2232 drives the coaxial dial gear 2233 to rotate. The dial gear 2233 further drives the annular motion sleeve 224 to rotate in a circle. Since the annular motion sleeve 224 is connected with the annular motion sliding block 23, the annular motion sliding block 23 is rotated. The bevel gear pair adopts an oil mist lubrication system to improve the smoothness of operation.

[0073] According to the description attached Figures 1-9 It can be known that the pitch control assembly 31 comprises a camera support 311 installed on the annular motion sliding block 23. The camera support 311 is provided with a pitch control motor 312. The driving end of the pitch control motor 312 is connected with a pitch shaft 313. The pitch shaft 313 is connected with a camera. The pitch shaft 313 is controlled to move through the pitch control motor 312. Then the camera is controlled to move to adjust the pitch angle through the pitch shaft 313. The pitch shaft 313 is integrated with a slip ring assembly at the end. The slip ring assembly realizes unlimited and continuous rotation transmission of power supply and data signals.

[0074] The inside of the helmet 1 is provided with an inflatable liner 7, one side of the inflatable liner 7 extends out an inflation control valve 71, the inflation control valve 71 is used for inflating and deflating the inflatable liner 7, the other side of the inflatable liner 7 is provided with a pressure valve 72, in order to further ensure the accuracy and stability of AR image acquisition, the inflatable liner 7 is built in the helmet 1, the inflatable liner 7 is a plastic air bag wrapped with pearl wool layer, the inflatable liner 7 can be inflated and deflated, fastened according to the head shape of the user, ensures stable wearing, reduces the fatigue influence caused by the shaking of the head device, and improves the accuracy and stability of image acquisition;

[0075] According to the description Figures 1-9 It can be known that the camera 32 is an automatic zoom camera 32, a protective cover 315 is arranged outside the camera 32, an axis seat 314 is arranged on one side of the camera 32 and connected with a pitch shaft 313, the camera 32 moves synchronously with the pitch shaft 313 through the axis seat 314, in the angle adjusting process of the camera 32, the axis seat 314 drives the camera 32 to move, and then the surface of the protective cover 315 is plated with a hydrophobic nano coating, and the camera 32 is protected by the protective cover 315;

[0076] The fixing seat 11 is a plate with a rectangular structure, a plurality of connecting bands 14 are arranged on the outer side of the fixing seat 11, the plurality of connecting bands 14 are connected with the ring-shaped movable guide rail, the plurality of connecting bands 14 and the ring-shaped movable guide rail are in an integral molding structure, the fixing seat 11 and the ring-shaped movable guide rail are connected through the plurality of connecting bands 14, and the stability of the ring-shaped movable guide rail is improved during wearing.

[0077] According to the description Figures 1-9 It can be known that the quick-release fixing structure 5 includes a dismounting seat 51, a guide sliding groove 52 is arranged on the dismounting seat 51, a dismounting guide rail 53 is arranged on the fixing seat 11, the guide sliding groove 52 is assembled on the dismounting guide rail 53, the dismounting seat 51 is connected with the battery module 6, a warning lamp 54 is arranged on one side of the dismounting seat 51, at least one pair of elastic positioning rods 55 is arranged on the inner side of the guide sliding groove 52, a pair of positioning grooves 5612 is arranged on the dismounting guide rail 53 and corresponds to the pair of elastic positioning rods 55, a Hall sensor is arranged in the elastic positioning rod 55, the connection state is judged by detecting the change of magnetic flux, and the power supply contact reliability reaches 99.99%.

[0078] As can be known from the above, the VR labeling and roaming road network automatic generation method and device, by fusing deep learning and intelligent hardware design, a VR panoramic labeling and road network automatic generation system is constructed. The innovation lies in that the algorithm end adopts dynamic linkage image enhancement, multi-modal feature recognition and closed loop verification mechanism, the hardware end integrates ring-shaped movable holder, modular design and quick-release structure, realizes full-process automation from panoramic data acquisition, labeling to navigation road network generation, significantly improves processing efficiency and system reliability, and provides a high-precision and low-cost solution for VR geographic information construction.

[0079] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since numerous changes, modifications, substitutions and variations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A method for automatically generating VR annotation and roaming road networks, characterized in that: The steps include: Step S1: image preprocessing, step S2: feature recognition training, step S3: annotation generation, and step S4: roaming road network construction; Step S1: performing image enhancement and denoising processing on the images captured by the image acquisition device, specifically including the following two steps; Step S1.1 Image enhancement: Use histogram equalization and contrast stretching methods to improve the overall visual quality of the panorama, making the image details clearer and providing a good foundation for subsequent feature extraction; Step S1.2: Denoising: Use Gaussian filtering and median filtering algorithms to remove noise interference in the panoramic image to ensure the accuracy and stability of the image data; Step S2: Train an image recognition model based on deep learning, detect the target through the image recognition model, output a new model of the target, and then use the semantic segmentation model to further refine the recognition of the scene. Specifically, it includes the following two steps; Step S2.1: Deep learning-based object detection: Using a pre-trained deep learning model, detect and identify key elements such as buildings, roads, landmarks, and traffic signs in the panoramic image, and output the target's category, location, and size information. Step S2.1: Semantic Segmentation: Using a semantic segmentation model, different objects and regions in the panorama are classified at the pixel level to further refine the understanding of the scene and assist in annotation and road network generation. Step S3: Based on the results of object detection and semantic segmentation, text annotations are added to the identified elements. The annotations include names, attributes, and related information. The fonts, colors, and positions of the annotations are optimized to ensure clarity and readability without affecting the aesthetics of the panorama and the user's browsing experience. Step S4: Based on the road annotation confidence and the machine deep learning results of the panoramic image, roads are extracted, information about intersections, T-junctions, and other road connection nodes is aggregated to determine nodes, and the road navigation information is combined with the panoramic map information to determine the road path. This specifically includes the following three steps: Step S4.1 Road Extraction and Refinement: Extract the road area from the semantic segmentation results, and refine and vectorize the road through morphological processing and edge detection algorithms to determine the centerline and boundary of the road; Step S4.2 Node determination: Based on the location information of the intersections and branch points of the roads, the nodes in the road network are determined, and the nodes are numbered and their attributes are defined, such as the intersection type and traffic rules; Step S4.3 Path connection: According to the actual connectivity of the roads, connect each node to form a complete roaming road network, and assign length, width, and slope attribute information to each road section to realize path planning and navigation functions.

2. A VR annotation and roaming road network image acquisition device, applied to the VR annotation and roaming road network automatic generation method according to claim 1, characterized in that: The invention comprises a helmet (1), wherein the helmet (1) is a semi-enclosed structure, and an image acquisition platform structure (2) is provided around the top of the helmet (1); The image acquisition platform structure (2) is equipped with at least one pair of image acquisition camera structures (3), the pair of image acquisition camera structures (3) are symmetrically arranged and interconnected, and the image acquisition platform structure (2) is also integrated with a sensor structure (4); The image acquisition pan-tilt structure (2) comprises an annular guide rail (21), the annular guide rail (21) is mounted on the helmet (1), a circular motion controller (22) is provided on the annular guide rail (21), a circular motion slider (23) is mounted on the annular guide rail (21), the circular motion slider (23) is connected to the circular motion controller (22), the circular motion controller (22) controls the circular motion slider (23) to perform circular motion, the circular motion slider (23) is connected to the image acquisition camera structure (3), and the annular slide rail is connected to the image acquisition pan-tilt structure; The image acquisition and camera structure (3) includes a pitch control component (31), the pitch control component (31) is assembled on the circular slider (23), and the pitch control component (31) is connected to a camera (32); A fixing seat (11) is provided on the top of the helmet (1), and a quick-release fixing structure (5) is connected to the fixing seat (11); The quick-release fixing structure (5) is equipped with a battery module (6), and the battery module (6) supplies power to the circumferential motion controller (22), the pitch control assembly (31), and the camera (32).

3. The VR annotation and roaming road network image acquisition device according to claim 2, characterized in that: The annular guide rail (21) is an annular structural track made of plastic PVC material. A plurality of buckles (13) are provided on one side of the annular guide rail (21). The outer ring of the helmet (1) is provided with a plurality of slots (12) which are engaged with the plurality of buckles (13).

4. The VR annotation and roaming road network image acquisition device according to claim 3, characterized in that: The circumferential motion controller (22) comprises a circumferential motion control housing (221), a control motor (222) being provided on one side of the circumferential motion control housing (221), a driving end of the control motor (222) being connected to a gear set (223), the gear set (223) being provided with a circumferential motion ring (224), a connecting block (225) extending from one side of the circumferential motion ring (224), the connecting block (225) being connected to a circumferential motion slider (23), and the circumferential motion slider (23) being matched with a circumferential motion guide rail.

5. The VR annotation and roaming road network image acquisition device according to claim 4, characterized in that: The gear set (223) is connected to the driving end of the control motor (222) via a reduction gear; an output gear is meshed with one side of the reduction gear; a first bevel gear (2231) is coaxially arranged on one side of the output gear; a second bevel gear (2232) is meshed with one side of the first bevel gear (2231); a toggle gear (2233) is coaxially arranged on one side of the second bevel gear (2232); and the toggle gear (2233) is meshed with the annular ring (224).

6. The VR annotation and roaming road network image acquisition device according to claim 5, characterized in that: The pitch control assembly (31) comprises a camera bracket (311), the camera bracket (311) is mounted on the ring slider (23), a pitch control motor (312) is provided on the camera bracket (311), a driving end of the pitch control motor (312) is connected to a pitch shaft (313), and the pitch shaft (313) is connected to the camera.

7. The VR annotation and roaming road network image acquisition device according to claim 6, characterized in that: An inflatable liner (7) is provided inside the helmet (1), an inflatable liner (7) is extended from one side of the inflatable liner (7), and the inflatable control valve (71) is used to inflate and deflate the inflatable liner (7), and a pressure valve (72) is provided on the other side of the inflatable liner (7).

8. The VR annotation and roaming road network image acquisition device according to claim 7, characterized in that: The camera (32) is an automatic zoom camera (32), a protective cover (315) is provided outside the camera (32), and an axle seat (314) is provided on one side of the camera (32) and is connected to the pitch axis (313).

9. The VR annotation and roaming road network image acquisition device according to claim 8, characterized in that: The fixing seat (11) is a plate with a rectangular structure. A plurality of connecting belts (14) are provided on the outer ring of the fixing seat (11), and the plurality of connecting belts (14) are connected to the circular guide rail.

10. The VR annotation and roaming road network image acquisition device according to claim 9, characterized in that: The quick-release fixing structure (5) includes a disassembly seat (51), a guide slot (52) is provided on the disassembly seat (51), a disassembly guide rail (53) is provided on the fixing seat (11), the guide slot (52) is assembled on the disassembly guide rail (53), the disassembly seat (51) is connected to the battery module (6), a warning light (54) is provided on one side of the disassembly seat (51), at least one pair of elastic positioning rods (55) are provided on the inner side of the guide slot (52), and a pair of positioning slots (56) (12) are provided on the disassembly guide rail (53) corresponding to the pair of elastic positioning rods (55).

Images