Environmental map construction method based on dynamic vision

Abstract

The invention provides an environment map construction method based on dynamic vision, which comprises the following steps: acquiring an environment image sequence based on a robot vehicle-mounted camera, and extracting feature points of each environment image in the environment image sequence based on an attention mechanism of YOLOv8n; carrying out dynamic point and static point distinguishing on the feature points based on a sparse optical flow method, rejecting the dynamic points based on a distinguishing result, and meanwhile, carrying out channel feature enhancement on the static points; and performing self-positioning and environment map construction on the robot based on the channel feature enhancement result of the static point, and dynamically updating the self-positioning and environment map based on the running state of the robot. The interference of a dynamic object on map construction and positioning identification is reduced, the accuracy and reliability of environment map construction are ensured, and a reliable basis is provided for the operation of a robot.

Description

Technical Field

[0001] The present invention relates to the fields of computer vision and robotics technology, and in particular to a method for constructing an environment map based on dynamic vision. Background Art

[0002] At present, Simultaneous Localization And Mapping (SLAM) technology has attracted widespread attention from all walks of life as a high-tech technology. SLAM technology can provide location information for mobile robots in general scenarios and simultaneously build environmental maps in real time to escort the robot's movement.

[0003] However, with the deepening of research, the application scenarios have also changed from highly static scenes to dynamic scenes with certain interference. In the process of feature point matching, it is very easy to cause the problem of feature point mismatch for dynamic target objects, which will cause certain interference to the accuracy of the robot's posture and the accuracy of map construction, resulting in a large error between the positioning trajectory effect and the actual trajectory;

[0004] Therefore, in order to overcome the above defects, the present invention provides an environment map construction method based on dynamic vision. Summary of the invention

[0005] The present invention provides an environmental map construction method based on dynamic vision, which is used for collecting environmental image sequences through a robot vehicle-mounted camera, and extracting features from the collected environmental image sequences, so as to achieve accurate and effective determination of feature points of each environmental image. Secondly, a sparse optical flow method is used to distinguish dynamic points from static points of the feature points, and the determined dynamic points are eliminated and channel features of the static points are enhanced, so as to reduce the interference of dynamic objects on map construction and positioning recognition, and provide traversal and guarantee for the robot to perform self-positioning and environmental map construction. Finally, the channel feature enhancement results of the static points are analyzed, so as to achieve accurate and reliable self-positioning of the robot and effective construction of the environmental map, and the self-positioning and environmental map are dynamically updated according to the running state of the robot, so as to ensure the accuracy and reliability of the environmental map construction, and provide a reliable basis for the operation of the robot.

[0006] The present invention provides a method for constructing an environment map based on dynamic vision, comprising:

[0007] Step 1: Collect the environment image sequence based on the robot's onboard camera, and extract the feature points of each environment image in the environment image sequence based on the attention mechanism of YOLOv8n;

[0008] Step 2: Based on the sparse optical flow method, the feature points are distinguished into dynamic points and static points, and the dynamic points are eliminated based on the distinction results. At the same time, the channel features of the static points are enhanced;

[0009] Step 3: Based on the channel feature enhancement results of static points, the robot is positioned and the environment map is constructed, and the self-positioning and environment map are dynamically updated based on the robot's operating status.

[0010] Preferably, a method for constructing an environment map based on dynamic vision, in step 1, collecting an environment image sequence based on a robot vehicle-mounted camera, comprises:

[0011] Obtain the viewing angle parameters of the robot and use the viewing angle parameters as the first camera configuration parameters;

[0012] Obtaining image acquisition requirements based on the management terminal, parsing the image acquisition requirements, obtaining image acquisition frequency and basic working parameters of the camera, and using the image acquisition frequency and basic working parameters as second camera configuration parameters;

[0013] The robot's onboard camera is configured based on the first camera configuration parameter and the second camera configuration parameter, and an interactive link between the management terminal and the robot's onboard camera is established according to the robot's main control center based on the configuration result;

[0014] Based on the management terminal, an image acquisition instruction is sent to the robot's onboard camera according to the interactive link, and based on the sending result, the robot's onboard camera is started to acquire the environment image sequence.

[0015] Preferably, a method for constructing an environment map based on dynamic vision, based on the sending result, starts the robot vehicle-mounted camera to collect the environment image sequence, including:

[0016] Obtaining the collected environment image sequence, and randomly selecting a preset sampling amount of environment image samples from the environment image sequence;

[0017] Perform pixel traversal on each environment image sample respectively to obtain corresponding pixel features, and analyze the pixel features based on a preset quality evaluation system to obtain the image quality of each environment image sample;

[0018] Determine the image acquisition pass rate of the robot's onboard camera based on the image quality of each environmental image sample, and optimize the robot's onboard camera when the pass rate does not meet a preset threshold;

[0019] Based on the optimization results, the environmental image sequence captured by the robot's onboard camera is transmitted back to the management terminal for storage.

[0020] Preferably, a method for constructing an environment map based on dynamic vision, in step 1, extracting feature points of each environment image in the environment image sequence based on the attention mechanism of YOLOv8n, including:

[0021] Acquire the obtained environment image sequence, and perform pixel traversal on each environment image in the environment image sequence based on a preset window size to obtain each pixel and a corresponding set of surrounding pixels;

[0022] Performing sum and average processing on each pixel and the corresponding surrounding pixel set to obtain a new pixel value corresponding to each pixel, and performing a first replacement on the original pixel value of each pixel based on the new pixel value to obtain a denoised environment image of the environment image;

[0023] At the same time, a brightness requirement interval of the environment image is determined based on the image processing requirement, and a brightness adjustment coefficient for the environment image is determined based on the brightness requirement interval;

[0024] Traversing each pixel in the denoised environment image in a row-by-row order, and obtaining a pixel value of each pixel based on the pixel traversal result;

[0025] Performing a linear operation on the pixel value of each pixel and the brightness adjustment coefficient, and obtaining an adjusted pixel value of each pixel based on the linear operation result;

[0026] Performing a second replacement on the pixel value of each pixel in the denoised environment image based on the adjusted pixel value, and performing a smoothing process on the adjusted pixel value of each pixel point based on the second replacement result to obtain an environment image with uniform brightness;

[0027] Based on the attention mechanism of YOLOv8n, the environment image with uniform brightness is processed to extract the corresponding feature points.

[0028] Preferably, a method for constructing an environment map based on dynamic vision processes a contrast-enhanced environment image based on the attention mechanism of YOLOv8n to extract corresponding feature points, including:

[0029] Obtaining the obtained uniform brightness environment image, and inputting the uniform brightness environment image into a convolutional neural network for processing;

[0030] Based on the convolutional neural network, the uniform brightness environment image is analyzed, the sensitive image area in the uniform brightness environment image is determined, and the environment detection is performed on the sensitive image area to obtain the preliminary image features of the uniform brightness environment image;

[0031] Determine key information based on feature extraction requirements, and adapt parameters of YOLOv8n's attention mechanism based on key information;

[0032] Performing attention analysis on the preliminary image features based on the parameter adaptation results, obtaining the attention weights corresponding to different feature positions in each preliminary image feature, and multiplying the attention weights with different feature positions in the preliminary image features to obtain weighted image features;

[0033] Key feature position points are selected based on weighted image features, and the key feature position points are determined as the final selected feature points.

[0034] Preferably, in a method for constructing an environment map based on dynamic vision, in step 2, the dynamic points and static points of the feature points are distinguished based on the sparse optical flow method, and the dynamic points are eliminated based on the distinction result, including:

[0035] Acquire the obtained environment image sequence, and split the environment image sequence to obtain each frame of the environment image;

[0036] Select two adjacent frames of environmental images from the environmental image sequence, and locate the same feature point in the two adjacent frames of environmental images. At the same time, based on the positioning result, perform constant processing on the pixel value brightness of the same feature point in the two adjacent frames of environmental images;

[0037] Based on the locking result, the time t of the previous frame environment image and the time t+Δt of the next frame environment image are extracted. At the same time, the pixel point of the previous frame environment image is set to I, and the pixel coordinates of I are (x, y, t), and the pixel point position of the next frame is (x+Δx, y+Δy, t+Δt);

[0038] Based on the constant brightness of pixel values, we get:

[0039] I(x,y,t)=I(x+Δx,y+Δy,t+Δt);

[0040] Perform Taylor series expansion on the right side of the above formula:

[0041]

[0042] Among them, α is the higher-order infinitesimal of Taylor series expansion;

[0043] Combining the above formulas, we can get:

[0044]

[0045] Assume that at time , the speed of the pixels on the x-axis and y-axis is:

[0046]

[0047] At the same time, the pixel gradient I of the two adjacent frames of environmental images x and I y Determine, at the same time, the brightness value I at time tt to determine;

[0048] Based on I x ,I y and I t Transform the above formula:

[0049] I x V x +I y V y =-I t ;

[0050] According to the principle of spatial consistency, take the extracted feature point as the center, take a square window with a side length of h = 3, use a 3*3 window, take 9 pixels, and establish 9 equations in the following form;

[0051]

[0052] The least square method is used to solve the above formula, and the average streamer vector is calculated using the solution obtained. The average streamer vector is:

[0053]

[0054] Among them, v x and v y is the average streamer value in the horizontal and vertical directions, then the equation for filtering the dynamic points to be removed is as follows:

[0055]

[0056] Wherein, k is the limit value of the dynamic feature point. If it exceeds the limit value, it is a dynamic feature point, otherwise it is a static feature point.

[0057] Distinguishing the feature points into dynamic points and static points based on the above formula, and locking the first dynamic point set and the static point set based on the differentiation result;

[0058] At the same time, continuous image frames are retrieved from a preset image library, and tracking annotations of different moving objects in the continuous image frames are extracted;

[0059] Based on the tracking and annotation results, the appearance features of different moving objects in different image frames are extracted respectively, and the appearance features of the moving objects in different image frames are subjected to correlation analysis from the starting point to the ending point based on the time series to determine the change features of the appearance features of the moving objects, wherein the change features include the displacement of the moving objects relative to the surrounding environment and the change amount of their own morphology;

[0060] Based on the deep learning model, the changing features of the appearance features of the moving object are learned, and a dynamic point detection model is obtained based on the learning results;

[0061] Extracting structural parameters of the dynamic point detection model, dividing the structural parameters into branches based on the business process of the dynamic point detection model, and determining model weights of different parameters in each branch based on the branch division;

[0062] Determine the prunable parameters based on the model weights, and prune the prunable parameters. At the same time, compress the required number of bits of the parameters based on the pruning results to obtain a lightweight dynamic point detection model.

[0063] Analyzing the static point set based on the lightweight dynamic point detection model, and obtaining a second dynamic point set based on the analysis result;

[0064] The first dynamic point set and the second dynamic point set are aggregated, and a elimination operation is performed on the aggregated dynamic point set.

[0065] Preferably, in a method for constructing an environment map based on dynamic vision, in step 2, channel feature enhancement is performed on static points, including:

[0066] Obtain the static point, and perform feature extraction on the static point and surrounding pixels in the environment image to obtain multi-channel features corresponding to the static point;

[0067] Analyze each channel feature to obtain the distribution and weight of multi-channel features, and determine the difference enhancement strategy for multi-channel features based on the distribution and weight;

[0068] Based on the difference enhancement strategy, personalized channel enhancement processing is performed on multi-channel features, and the personalized channel enhancement processing results of different channel features are associated and summarized to complete the channel feature enhancement processing of static points.

[0069] Preferably, a method for constructing an environment map based on dynamic vision, in step 3, the robot is positioned and the environment map is constructed based on the channel feature enhancement result of the static point, including:

[0070] Obtain the channel feature enhancement results of the static points and extract the original position information of the static points in the environment image sequence;

[0071] Determine the distribution characteristics of each static point based on the original position information, and perform co-location mapping on the channel feature enhancement results of each static point based on the distribution characteristics;

[0072] Based on the same-position mapping results, boundary nested features of channel feature enhancement results of adjacent static points are obtained, and based on the boundary nested feature pairs, the joint seams of channel feature enhancement results of adjacent static points are determined;

[0073] The channel feature enhancement results of adjacent static points are spliced ​​based on the splicing seams, and the depth of field information corresponding to the channel enhancement results of different static points is extracted based on the splicing results;

[0074] Obtain the configuration parameters of the robot's onboard camera, and parse the depth of field information based on the configuration parameters to obtain the position and orientation of the robot's onboard camera relative to different static points;

[0075] The robot is positioned based on the position and orientation of the robot's onboard camera relative to different static points to obtain the robot's position coordinates in the current environment;

[0076] The splicing results of the channel feature enhancement results of adjacent static points are visualized in a visualization coordinate system, and the relative directions and distance parameters between different static points are determined based on the visualization results;

[0077] At the same time, based on the splicing result of the channel feature enhancement results of adjacent static points, feature selection is performed on the channel feature enhancement result of each static point to obtain the feature representation of each static point;

[0078] The relative directions and distance parameters between different static points and the characteristic representation of each static point are used to construct an environmental map to obtain a map frame. The position of the robot is marked in the map frame according to the position coordinates of the robot in the current environment to obtain the final environmental map.

[0079] Preferably, a method for constructing an environment map based on dynamic vision, in step 3, dynamically updating the robot's own positioning and environment map based on the robot's operating state, includes:

[0080] Obtain the running status of the robot, and when the position of the robot is determined to have changed based on the running status, synchronously trigger the robot's onboard camera to reacquire the environment image sequence;

[0081] Recheck the reacquired environment image sequence, and determine the change parameters of the robot's own positioning and the channel characteristics of the static point based on the recheck results;

[0082] Determine the update nodes for the robot's own positioning and the environment map and the update parameters corresponding to the update nodes based on the change parameters of the channel characteristics of the robot's own positioning and the static points;

[0083] The robot's own positioning and environment map are dynamically updated based on the update nodes and the update parameters corresponding to the update nodes.

[0084] Preferably, a method for constructing an environment map based on dynamic vision, in step 3, dynamically updating the robot's own positioning and environment map based on the robot's operating state, includes:

[0085] Obtain a dynamically updated real-time environment map, and determine the robot's direction of travel based on the real-time environment map and the environment map at the previous moment;

[0086] At the same time, the distribution of obstacles in the direction of travel of the robot is determined based on the real-time environment map, and a guidance route for the robot in the direction of travel is generated based on the obstacle distribution;

[0087] A travel control instruction is generated based on the guidance route, and the robot is controlled to travel according to the guidance route based on the travel control instruction.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] The robot's on-board camera collects environmental image sequences, and features are extracted from the collected environmental image sequences to achieve accurate and effective determination of the feature points of each environmental image. Secondly, the sparse optical flow method is used to distinguish dynamic points from static points, and the determined dynamic points are eliminated and the channel features of static points are enhanced, which reduces the interference of dynamic objects on map construction and positioning recognition, and provides traversal and guarantee for the robot's self-positioning and environmental map construction. Finally, the channel feature enhancement results of static points are analyzed to achieve accurate and reliable self-positioning of the robot and effective construction of the environmental map, and dynamically update the self-positioning and environmental map according to the robot's operating status, ensuring the accuracy and reliability of environmental map construction, and providing a reliable basis for the robot's operation.

[0090] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0091] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0093] Figure 1 A flowchart of a method for constructing an environment map based on dynamic vision in an embodiment of the present invention;

[0094] Figure 2 This is a flowchart of step 1 in a method for constructing an environment map based on dynamic vision in an embodiment of the present invention;

[0095] Figure 3This is a flowchart of step 2 in a method for constructing an environment map based on dynamic vision in an embodiment of the present invention. DETAILED DESCRIPTION

[0096] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0097] Embodiment 1:

[0098] This embodiment provides a method for constructing an environment map based on dynamic vision. Figure 1 As shown, including:

[0099] Step 1: Collect the environment image sequence based on the robot's onboard camera, and extract the feature points of each environment image in the environment image sequence based on the attention mechanism of YOLOv8n;

[0100] Step 2: Based on the sparse optical flow method, the feature points are distinguished into dynamic points and static points, and the dynamic points are eliminated based on the distinction results. At the same time, the channel features of the static points are enhanced;

[0101] Step 3: Based on the channel feature enhancement results of static points, the robot is positioned and the environment map is constructed, and the self-positioning and environment map are dynamically updated based on the robot's operating status.

[0102] In this embodiment, YOLOv8n is a target detection algorithm. Compared with other versions, YOLOv8n has a smaller model size and computational complexity, and is more suitable for application in some resource-constrained devices or scenarios with high real-time requirements. At the same time, it achieves higher efficiency while maintaining a certain detection accuracy.

[0103] In this embodiment, the visual SLAM algorithm can provide location information for the mobile robot in general scenarios and simultaneously build an environmental map in real time to safeguard the robot's movement.

[0104] In this embodiment, the environmental image sequence refers to the result obtained by collecting images of the environment where the robot is located at different angles through the robot's onboard camera.

[0105] In this embodiment, the attention mechanism is a scheme used to quickly locate key image regions or image features from a sequence of environmental images.

[0106] In this embodiment, the feature point refers to a position and image parameter in each environment image that can characterize the characteristics or specific conditions of the environment image.

[0107] In this embodiment, the sparse optical flow method calculates the offset of a set of feature points in the image, can perform image registration accurate to the pixel level, and track a specified set of points (this set of points has certain similar features). The tracking is relatively stable and reliable, and is used to distinguish between dynamic points (such as running people or cars) and static points (such as trees in the environment, etc.) in the environmental image.

[0108] In this embodiment, channel feature enhancement refers to enhancing the image features of static points to facilitate construction of an environment map.

[0109] In this embodiment, the operating state refers to the operating status of the robot, that is, whether the robot is currently moving or in a stationary state.

[0110] The working principle and beneficial effects of the above technical solution are: collecting environmental image sequences through the robot's on-board camera, and extracting features from the collected environmental image sequences to achieve accurate and effective determination of the feature points of each environmental image; secondly, distinguishing dynamic points from static points through the sparse optical flow method, eliminating the determined dynamic points, and enhancing the channel features of the static points, thereby reducing the interference of dynamic objects on map construction and positioning recognition, and providing traversal and guarantee for the robot's self-positioning and environmental map construction; finally, analyzing the channel feature enhancement results of the static points, achieving accurate and reliable self-positioning of the robot and effective construction of the environmental map, and dynamically updating the self-positioning and environmental map according to the robot's operating status, thereby ensuring the accuracy and reliability of environmental map construction, and providing a reliable basis for the robot's operation.

[0111] Embodiment 2:

[0112] Based on Example 1, this example provides a method for constructing an environment map based on dynamic vision, such as Figure 2 As shown, in step 1, the environment image sequence is collected based on the robot's onboard camera, including:

[0113] Step 101: Obtain the viewing angle parameters of the robot, and use the viewing angle parameters as the first camera configuration parameters;

[0114] Step 102: obtaining image acquisition requirements based on the management terminal, parsing the image acquisition requirements, obtaining image acquisition frequency and basic working parameters of the camera, and using the image acquisition frequency and basic working parameters as second camera configuration parameters;

[0115] Step 103: configuring the robot's onboard camera based on the first camera configuration parameter and the second camera configuration parameter, and building an interactive link between the management terminal and the robot's onboard camera according to the robot's main control center based on the configuration result;

[0116] Step 104: Based on the management terminal, an image acquisition instruction is sent to the robot vehicle-mounted camera according to the interactive link, and based on the sending result, the robot vehicle-mounted camera is started to acquire the environment image sequence.

[0117] In this embodiment, the viewing angle parameter refers to the viewing angle range and viewing angle of the robot's ability to view the external environment. The purpose is to configure the robot's on-board camera with the same viewing angle according to the viewing angle parameter, thereby ensuring that the robot's on-board camera can capture corresponding environmental images according to the robot's travel requirements.

[0118] In this embodiment, the image acquisition requirement is obtained from the management terminal and is used to represent the frequency of image acquisition and the requirements such as focal length and resolution that the camera needs to meet when performing image acquisition.

[0119] In this embodiment, the basic working parameters refer to requirements that the camera needs to meet when performing image acquisition, such as specific numerical requirements for resolution.

[0120] The working principle and beneficial effects of the above technical solution are: the robot's on-board camera is configured through the robot's viewing angle parameters and the basic working parameters of the camera determined by the image acquisition requirements obtained from the management terminal, ensuring that the robot's on-board camera can effectively perform image acquisition operations. At the same time, after the configuration is completed, an interactive link between the management terminal and the robot's on-board camera is established, thereby realizing the control of the robot's on-board camera through the management terminal to perform corresponding image acquisition, thereby ensuring the reliability of image acquisition.

[0121] Embodiment 3:

[0122] Based on Example 2, this embodiment provides a method for constructing an environment map based on dynamic vision, which starts the robot vehicle-mounted camera to collect an environment image sequence based on the sending result, including:

[0123] Obtaining the collected environment image sequence, and randomly selecting a preset sampling amount of environment image samples from the environment image sequence;

[0124] Perform pixel traversal on each environment image sample respectively to obtain corresponding pixel features, and analyze the pixel features based on a preset quality evaluation system to obtain the image quality of each environment image sample;

[0125] Determine the image acquisition pass rate of the robot's onboard camera based on the image quality of each environmental image sample, and optimize the robot's onboard camera when the pass rate does not meet a preset threshold;

[0126] Based on the optimization results, the environmental image sequence captured by the robot's onboard camera is transmitted back to the management terminal for storage.

[0127] In this embodiment, the preset collection amount is set in advance, for example, it may be five or ten pictures.

[0128] In this embodiment, the pixel feature refers to information such as the pixel value and color value of each pixel obtained after checking the pixels of each environment image sample.

[0129] In this embodiment, the preset quality assessment system is trained in advance and is used to assess the image quality of the collected environment image samples, including assessing information such as the clarity and posture of the environment image samples.

[0130] In this embodiment, the preset threshold is set in advance and is used to measure whether the image acquisition pass rate meets the required standard and can be adjusted.

[0131] The working principle and beneficial effects of the above technical solution are: by retrieving a preset amount of environmental image samples from the collected environmental image sequence and determining the image quality of the environmental image samples, the image acquisition pass rate of the robot's on-board camera is locked according to the image quality, thereby facilitating the judgment of whether the configuration of the robot's on-board camera is qualified according to the image acquisition pass rate. Finally, after the configuration is qualified, the collected environmental image sequence is transmitted back to the management terminal for storage, thereby ensuring the accuracy and reliability of image acquisition of the robot's surrounding environment.

[0132] Embodiment 4:

[0133] On the basis of Example 1, this embodiment provides a method for constructing an environment map based on dynamic vision. In step 1, feature points of each environment image in the environment image sequence are extracted based on the attention mechanism of YOLOv8n, including:

[0134] Acquire the obtained environment image sequence, and perform pixel traversal on each environment image in the environment image sequence based on a preset window size to obtain each pixel and a corresponding set of surrounding pixels;

[0135] Performing sum and average processing on each pixel and the corresponding surrounding pixel set to obtain a new pixel value corresponding to each pixel, and performing a first replacement on the original pixel value of each pixel based on the new pixel value to obtain a denoised environment image of the environment image;

[0136] At the same time, a brightness requirement interval of the environment image is determined based on the image processing requirement, and a brightness adjustment coefficient for the environment image is determined based on the brightness requirement interval;

[0137] Traversing each pixel in the denoised environment image in a row-by-row order, and obtaining a pixel value of each pixel based on the pixel traversal result;

[0138] Performing a linear operation on the pixel value of each pixel and the brightness adjustment coefficient, and obtaining an adjusted pixel value of each pixel based on the linear operation result;

[0139] Performing a second replacement on the pixel value of each pixel in the denoised environment image based on the adjusted pixel value, and performing a smoothing process on the adjusted pixel value of each pixel point based on the second replacement result to obtain an environment image with uniform brightness;

[0140] Based on the attention mechanism of YOLOv8n, the environment image with uniform brightness is processed to extract the corresponding feature points.

[0141] In this embodiment, the preset window size is set in advance and is used to delimit the area of ​​each pixel and the surrounding pixels, so as to facilitate denoising of the environmental image, for example, it can be a 3*3 pixel point window.

[0142] In this embodiment, the surrounding pixel set refers to all pixels adjacent to each pixel, and is determined according to a preset window size.

[0143] In this embodiment, the first replacement refers to replacing the pixel value of the corresponding pixel point by performing a summing and averaging process on each pixel and a corresponding set of surrounding pixels, in order to perform a denoising process on the image.

[0144] In this embodiment, the required brightness interval refers to the brightness range that the environmental image needs to reach, in order to ensure that the brightness of the environmental images collected at night and during the day meets the analysis requirements.

[0145] In this embodiment, the brightness adjustment coefficient is determined according to the brightness requirement range, and is used to represent a specific value of the brightness adjustment of the environment image.

[0146] In this embodiment, performing a linear operation on the pixel value of each pixel and the brightness adjustment coefficient refers to recalculating the pixel value of each pixel using the brightness adjustment coefficient to obtain a new pixel value that meets the brightness range requirement.

[0147] In this embodiment, adjusting the pixel value refers to a result obtained by re-adjusting the pixel value of each pixel using the brightness adjustment coefficient.

[0148] In this embodiment, the second replacement refers to replacing the pixel value of the corresponding pixel point by obtaining the adjusted pixel value of each pixel through the linear operation result, in order to adjust the brightness value of the image.

[0149] In this embodiment, the smoothing process refers to performing an over-processing on the brightness values ​​of different pixels, in order to improve the overall effect of the image brightness.

[0150] In this embodiment, the brightness uniform environment image refers to a result obtained by adjusting the brightness of the denoised environment image.

[0151] The working principle and beneficial effects of the above technical solution are: by traversing and processing the pixels of each environmental image in the environmental image sequence, the noise in the environmental image and the brightness in the environmental image are adjusted, which provides convenience for extracting feature points in the environmental image. Finally, the processed environmental image is processed through the attention mechanism of the YOLOv8n model to achieve effective extraction of feature points, ensuring the reliability and accuracy of feature point extraction.

[0152] Embodiment 5:

[0153] Based on Example 4, this embodiment provides a method for constructing an environment map based on dynamic vision, which processes the contrast-enhanced environment image based on the attention mechanism of YOLOv8n and extracts corresponding feature points, including:

[0154] Obtaining the obtained uniform brightness environment image, and inputting the uniform brightness environment image into a convolutional neural network for processing;

[0155] Based on the convolutional neural network, the uniform brightness environment image is analyzed, the sensitive image area in the uniform brightness environment image is determined, and the environment detection is performed on the sensitive image area to obtain the preliminary image features of the uniform brightness environment image;

[0156] Determine key information based on feature extraction requirements, and adapt parameters of YOLOv8n's attention mechanism based on key information;

[0157] Performing attention analysis on the preliminary image features based on the parameter adaptation results, obtaining the attention weights corresponding to different feature positions in each preliminary image feature, and multiplying the attention weights with different feature positions in the preliminary image features to obtain weighted image features;

[0158] Key feature position points are selected based on weighted image features, and the key feature position points are determined as the final selected feature points.

[0159] In this embodiment, the sensitive image area refers to an image area related to static point and dynamic point detection obtained after analyzing the environment image with uniform brightness, and is a part of the environment image with uniform brightness.

[0160] In this embodiment, the preliminary image features refer to object information obtained after performing environmental detection on the sensitive image area, including information such as the shape and position of the recorded object.

[0161] In this embodiment, the feature extraction requirements are set in advance.

[0162] In this embodiment, key information refers to image parameters that need to be extracted emphatically when performing feature extraction, including key information such as the type, position, and shape of the object recorded in the image.

[0163] In this embodiment, the attention weight refers to the importance of different feature positions under the attention mechanism analysis.

[0164] In this embodiment, the weighted image feature refers to the result obtained by multiplying the obtained attention weight with different feature positions in the preliminary image feature, which is used to characterize the importance of feature correspondence at different positions.

[0165] In this embodiment, the key feature position point refers to the position corresponding to the final valid feature point that needs to be extracted determined according to the attention mechanism.

[0166] The working principle and beneficial effects of the above technical solution are: by inputting the environment image with uniform brightness into the convolutional neural network for analysis, the sensitive image area in the environment image with uniform brightness is preliminarily locked, and then the environment is detected in the sensitive image area to determine the preliminary image features. Secondly, the preliminary image features are analyzed by the attention mechanism of YOLOv8n to lock the key feature positions from the preliminary image features, and finally the effective selection of feature points is realized, which ensures the reliability and accuracy of the feature point selection and provides convenience and guarantee for the construction of environmental maps.

[0167] Embodiment 6:

[0168] Based on Example 1, this embodiment provides a method for constructing an environment map based on dynamic vision. In step 2, dynamic points and static points are distinguished from each other based on the sparse optical flow method, and dynamic points are eliminated based on the distinction results, including:

[0169] Acquire the obtained environment image sequence, and split the environment image sequence to obtain each frame of the environment image;

[0170] Select two adjacent frames of environmental images from the environmental image sequence, and locate the same feature point in the two adjacent frames of environmental images. At the same time, based on the positioning result, perform constant processing on the pixel value brightness of the same feature point in the two adjacent frames of environmental images;

[0171] Based on the locking result, the time t of the previous frame environment image and the time t+Δt of the next frame environment image are extracted. At the same time, the pixel point of the previous frame environment image is set to I, and the pixel coordinates of I are (x, y, t), and the pixel point position of the next frame is (x+Δx, y+Δy, t+Δt);

[0172] Based on the constant brightness of pixel values, we get:

[0173] I(x,y,t)=I(x+Δx,y+Δy,t+Δt);

[0174] Perform Taylor series expansion on the right side of the above formula:

[0175]

[0176] Among them, α is the higher-order infinitesimal of Taylor series expansion;

[0177] Combining the above formulas, we can get:

[0178]

[0179] Assume that at time , the speed of the pixels on the x-axis and y-axis is:

[0180]

[0181] At the same time, the pixel gradient I of the two adjacent frames of environmental images x and I y Determine, at the same time, the brightness value I at time t t to determine;

[0182] Based on I x ,I y and I t Transform the above formula:

[0183] I x V x +I y V y =-I t ;

[0184] According to the principle of spatial consistency, take the extracted feature point as the center, take a square window with a side length of h = 3, use a 3*3 window, take 9 pixels, and establish 9 equations in the following form;

[0185]

[0186] The least square method is used to solve the above formula, and the average streamer vector is calculated using the solution obtained. The average streamer vector is:

[0187]

[0188] Among them, v x and v y is the average streamer value in the horizontal and vertical directions, then the equation for filtering the dynamic points to be removed is as follows:

[0189]

[0190] Wherein, k is the limit value of the dynamic feature point. If it exceeds the limit value, it is a dynamic feature point, otherwise it is a static feature point.

[0191] Distinguishing the feature points into dynamic points and static points based on the above formula, and locking the first dynamic point set and the static point set based on the differentiation result;

[0192] At the same time, continuous image frames are retrieved from a preset image library, and tracking annotations of different moving objects in the continuous image frames are extracted;

[0193] Based on the tracking and annotation results, the appearance features of different moving objects in different image frames are extracted respectively, and the appearance features of the moving objects in different image frames are subjected to correlation analysis from the starting point to the ending point based on the time series to determine the change features of the appearance features of the moving objects, wherein the change features include the displacement of the moving objects relative to the surrounding environment and the change amount of their own morphology;

[0194] Based on the deep learning model, the changing features of the appearance features of the moving object are learned, and a dynamic point detection model is obtained based on the learning results;

[0195] Extracting structural parameters of the dynamic point detection model, dividing the structural parameters into branches based on the business process of the dynamic point detection model, and determining model weights of different parameters in each branch based on the branch division;

[0196] Determine the prunable parameters based on the model weights, and prune the prunable parameters. At the same time, compress the required number of bits of the parameters based on the pruning results to obtain a lightweight dynamic point detection model.

[0197] Analyzing the static point set based on the lightweight dynamic point detection model, and obtaining a second dynamic point set based on the analysis result;

[0198] The first dynamic point set and the second dynamic point set are aggregated, and a elimination operation is performed on the aggregated dynamic point set.

[0199] In this embodiment, locating the same feature point in two adjacent frames of environmental images refers to locking and tracking the unified feature point in two adjacent frames of environmental images. The purpose is to analyze the state of the unified feature point in different frame images, so as to facilitate determining whether the current feature point is a dynamic point.

[0200] In this embodiment, the first dynamic point set refers to the dynamic points existing in the feature points determined by the calculation formula.

[0201] In this embodiment, the preset image library is known in advance, and stores a series of continuous images of known static objects and dynamic objects.

[0202] In this embodiment, tracking and labeling refers to annotating different moving objects in continuous image frames to distinguish dynamic objects from static objects.

[0203] In this embodiment, the appearance feature refers to information such as the position and shape of different moving objects in different image frames.

[0204] In this embodiment, the dynamic point detection model refers to a model capable of determining dynamic points that is obtained by learning the changing characteristics of the appearance characteristics of a moving object through a deep learning model.

[0205] In this embodiment, the structural parameters refer to the levels included in the dynamic point detection model and the specific parameter structure of each level.

[0206] In this embodiment, the business process refers to the corresponding specific steps and the collaborative relationship between different structural parameters when the dynamic point detection model performs corresponding business analysis.

[0207] In this embodiment, branch division refers to splitting the structural parameters into independent structural parameters corresponding to different services, so as to facilitate determination of which structural parameters are important and which structural parameters are not important.

[0208] In this embodiment, the model weight is used to characterize the importance of different parameters in each branch.

[0209] In this embodiment, the prunable parameters refer to parameters that can be deleted.

[0210] In this embodiment, the number of bits required for the parameter refers to the length corresponding to the structure parameter.

[0211] In this embodiment, the lightweight dynamic point detection model refers to a final model obtained by deleting unimportant parameters in the dynamic point detection model and compressing data.

[0212] In this embodiment, the second dynamic point set refers to the result obtained by re-analyzing the static point set determined by the formula using the obtained lightweight dynamic point detection model, that is, determining the dynamic points existing in the static points.

[0213] The working principle and beneficial effects of the above technical solution are: each frame of the environmental image is processed by the obtained formula to achieve effective distinction and determination of the first dynamic point set and the static point set in each frame of the environmental image; secondly, the determined static point set is analyzed again by the constructed lightweight dynamic point detection model to achieve effective re-inspection of the static points, thereby locking the second dynamic point set existing in the static points; finally, the obtained first dynamic point set and the second dynamic point set are summarized, and all the summarized dynamic points are eliminated, ensuring the accuracy and reliability of the dynamic point determination, and also providing a guarantee for the construction of the environmental map.

[0214] Embodiment 7:

[0215] Based on Example 1, this example provides a method for constructing an environment map based on dynamic vision, such as Figure 3 As shown, in step 2, channel feature enhancement is performed on static points, including:

[0216] Step 201: Obtain the static point, and perform feature extraction on the static point and surrounding pixels in the environment image to obtain multi-channel features corresponding to the static point;

[0217] Step 202: Analyze each channel feature to obtain the distribution and weight of multi-channel features, and determine a difference enhancement strategy for the multi-channel features based on the distribution and weight;

[0218] Step 203: Perform personalized channel enhancement processing on the multi-channel features based on the difference enhancement strategy, and associate and summarize the personalized channel enhancement processing results of different channel features to complete the channel feature enhancement processing of the static point.

[0219] In this embodiment, the multi-channel feature refers to the result obtained after feature extraction of the static point and the surrounding pixels, which represents the characteristics presented by the static point and the surrounding pixels, such as the object type and the current posture presented.

[0220] In this embodiment, the difference enhancement strategy refers to a specific scheme for enhancing different channel features in the multi-channel features.

[0221] The working principle and beneficial effects of the above technical solution are: by extracting features from the determined static points and surrounding pixels, the multi-channel features corresponding to the static points can be effectively determined; secondly, personalized channel enhancement processing is performed on the obtained multi-channel features to effectively determine the details of different channel features, providing reliable data support for constructing an environmental map and ensuring the accuracy and reliability of the ultimately constructed environmental map.

[0222] Embodiment 8:

[0223] On the basis of Example 1, this embodiment provides a method for constructing an environment map based on dynamic vision. In step 3, the robot is positioned and the environment map is constructed based on the channel feature enhancement result of the static point, including:

[0224] Obtain the channel feature enhancement results of the static points and extract the original position information of the static points in the environment image sequence;

[0225] Determine the distribution characteristics of each static point based on the original position information, and perform co-location mapping on the channel feature enhancement results of each static point based on the distribution characteristics;

[0226] Based on the same-position mapping results, boundary nested features of channel feature enhancement results of adjacent static points are obtained, and based on the boundary nested feature pairs, the joint seams of channel feature enhancement results of adjacent static points are determined;

[0227] The channel feature enhancement results of adjacent static points are spliced ​​based on the splicing seams, and the depth of field information corresponding to the channel enhancement results of different static points is extracted based on the splicing results;

[0228] Obtain the configuration parameters of the robot's onboard camera, and parse the depth of field information based on the configuration parameters to obtain the position and orientation of the robot's onboard camera relative to different static points;

[0229] The robot is positioned based on the position and orientation of the robot's onboard camera relative to different static points to obtain the robot's position coordinates in the current environment;

[0230] The splicing results of the channel feature enhancement results of adjacent static points are visualized in a visualization coordinate system, and the relative directions and distance parameters between different static points are determined based on the visualization results;

[0231] At the same time, based on the splicing result of the channel feature enhancement results of adjacent static points, feature selection is performed on the channel feature enhancement result of each static point to obtain the feature representation of each static point;

[0232] The relative directions and distance parameters between different static points and the characteristic representation of each static point are used to construct an environmental map to obtain a map frame. The position of the robot is marked in the map frame according to the position coordinates of the robot in the current environment to obtain the final environmental map.

[0233] In this embodiment, the original position information refers to the specific position of the static point in the environment image sequence.

[0234] In this embodiment, the distribution feature refers to the specific distribution of each static point in the environment.

[0235] In this embodiment, the same-position mapping refers to representing the channel feature enhancement result of each static point at the position of the corresponding static point.

[0236] In this embodiment, the boundary nested feature refers to the association relationship or overlapping relationship between the channel feature enhancement results of adjacent static points.

[0237] In this embodiment, the stitching seam is a stitching position used to stitch the channel feature enhancement results of adjacent static points.

[0238] In this embodiment, the depth of field information is used to characterize the distance between different positions of the static point and the robot's onboard camera.

[0239] In this embodiment, the distance parameter refers to the specific distance length between different static points.

[0240] In this embodiment, the feature representation refers to the representative features presented by each static point.

[0241] The working principle and beneficial effects of the above technical solution are: by processing the channel enhancement results of static points, the channel feature enhancement results of adjacent static points are effectively spliced, and the features of different static points are comprehensively and effectively determined. Secondly, the spliced ​​results are analyzed to effectively determine the directions and distances between different static points and the robot's on-board camera and between different static points. Finally, the environment map is constructed based on the relative directions and distance parameters between different static points and the feature representation of each static point, and the determined robot position is marked in the environment map, so as to realize comprehensive and reliable construction of the environment map, providing a reliable reference basis for the robot's terrain driving.

[0242] Embodiment 9:

[0243] Based on Example 1, this embodiment provides a method for constructing an environment map based on dynamic vision. In step 3, the robot's own positioning and environment map are dynamically updated based on the robot's operating state, including:

[0244] Obtain the running status of the robot, and when the position of the robot is determined to have changed based on the running status, synchronously trigger the robot's onboard camera to reacquire the environment image sequence;

[0245] Recheck the reacquired environment image sequence, and determine the change parameters of the robot's own positioning and the channel characteristics of the static point based on the recheck results;

[0246] Determine the update nodes for the robot's own positioning and the environment map and the update parameters corresponding to the update nodes based on the change parameters of the channel characteristics of the robot's own positioning and the static points;

[0247] The robot's own positioning and environment map are dynamically updated based on the update nodes and the update parameters corresponding to the update nodes.

[0248] In this embodiment, the change parameter refers to the change in the robot's own positioning and the channel characteristics of the static point, including the amplitude of the change in the viewing angle and the distance at which the static object changes.

[0249] In this embodiment, the update node refers to the specific data and location points for updating the robot's own positioning and environment map.

[0250] In this embodiment, the update parameter refers to the specific update degree corresponding to the update of the update node, including the direction and distance of the change in the robot's own positioning and the specific object that needs to be replaced in the environment map.

[0251] The working principle and beneficial effects of the above technical solution are: by real-time monitoring of the robot's operating status, it is possible to timely and effectively collect new environmental image sequences when the position changes according to the operating status, and analyze and process the collected new environmental image sequences, thereby realizing the robot's own positioning and dynamic updating of the environmental map, ensuring the real-time effectiveness of the environmental map.

[0252] Embodiment 10:

[0253] Based on Example 1, this embodiment provides a method for constructing an environment map based on dynamic vision. In step 3, the robot's own positioning and environment map are dynamically updated based on the robot's operating state, including:

[0254] Obtain a dynamically updated real-time environment map, and determine the robot's direction of travel based on the real-time environment map and the environment map at the previous moment;

[0255] At the same time, the distribution of obstacles in the direction of travel of the robot is determined based on the real-time environment map, and a guidance route for the robot in the direction of travel is generated based on the obstacle distribution;

[0256] A travel control instruction is generated based on the guidance route, and the robot is controlled to travel according to the guidance route based on the travel control instruction.

[0257] The working principle and beneficial effects of the above technical solution are: by analyzing the dynamically updated real-time environmental map, the robot's travel direction can be accurately and effectively determined. At the same time, the distribution of obstacles in the robot's travel direction can be determined through the real-time environmental map, so that it is convenient to determine the guidance route of the robot in the travel direction according to the obstacle distribution, which is convenient for guiding the robot. Finally, a travel control instruction is generated according to the obtained guidance route, so that the robot can be controlled to move safely and reliably through the travel control instruction, which provides great convenience and guarantee for the control of the robot.

[0258] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing an environment map based on dynamic vision, characterized in that: include: Step 1: Collect the environment image sequence based on the robot's onboard camera, and extract the feature points of each environment image in the environment image sequence based on the attention mechanism of YOLOv8n; Step 2: Based on the sparse optical flow method, the feature points are distinguished into dynamic points and static points, and the dynamic points are eliminated based on the distinction results. At the same time, the channel features of the static points are enhanced; Step 3: Based on the channel feature enhancement results of static points, the robot is positioned and the environment map is constructed, and the self-positioning and environment map are dynamically updated based on the robot's operating status.

2. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 1, the environment image sequence is collected based on the robot's onboard camera, including: Obtain the viewing angle parameters of the robot and use the viewing angle parameters as the first camera configuration parameters; Obtaining image acquisition requirements based on the management terminal, parsing the image acquisition requirements, obtaining image acquisition frequency and basic working parameters of the camera, and using the image acquisition frequency and basic working parameters as second camera configuration parameters; The robot's onboard camera is configured based on the first camera configuration parameter and the second camera configuration parameter, and an interactive link between the management terminal and the robot's onboard camera is established according to the robot's main control center based on the configuration result; Based on the management terminal, an image acquisition instruction is sent to the robot's onboard camera according to the interactive link, and based on the sending result, the robot's onboard camera is started to acquire the environment image sequence.

3. The method for constructing an environment map based on dynamic vision according to claim 2, characterized in that: Based on the sent results, the robot's onboard camera is started to collect environmental image sequences, including: Obtaining the collected environment image sequence, and randomly selecting a preset sampling amount of environment image samples from the environment image sequence; Perform pixel traversal on each environment image sample respectively to obtain corresponding pixel features, and analyze the pixel features based on a preset quality evaluation system to obtain the image quality of each environment image sample; Determine the image acquisition pass rate of the robot's onboard camera based on the image quality of each environmental image sample, and optimize the robot's onboard camera when the pass rate does not meet a preset threshold; Based on the optimization results, the environmental image sequence captured by the robot's onboard camera is transmitted back to the management terminal for storage.

4. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 1, the feature points of each environment image in the environment image sequence are extracted based on the attention mechanism of YOLOv8n, including: Acquire the obtained environment image sequence, and perform pixel traversal on each environment image in the environment image sequence based on a preset window size to obtain each pixel and a corresponding set of surrounding pixels; Performing sum and average processing on each pixel and the corresponding surrounding pixel set to obtain a new pixel value corresponding to each pixel, and performing a first replacement on the original pixel value of each pixel based on the new pixel value to obtain a denoised environment image of the environment image; At the same time, a brightness requirement interval of the environment image is determined based on the image processing requirement, and a brightness adjustment coefficient for the environment image is determined based on the brightness requirement interval; Traversing each pixel in the denoised environment image in a row-by-row order, and obtaining a pixel value of each pixel based on the pixel traversal result; Performing a linear operation on the pixel value of each pixel and the brightness adjustment coefficient, and obtaining an adjusted pixel value of each pixel based on the linear operation result; Performing a second replacement on the pixel value of each pixel in the denoised environment image based on the adjusted pixel value, and performing a smoothing process on the adjusted pixel value of each pixel point based on the second replacement result to obtain an environment image with uniform brightness; Based on the attention mechanism of YOLOv8n, the environment image with uniform brightness is processed to extract the corresponding feature points.

5. The method for constructing an environment map based on dynamic vision according to claim 4, characterized in that: The contrast-enhanced environment image is processed based on the attention mechanism of YOLOv8n to extract the corresponding feature points, including: Obtaining the obtained uniform brightness environment image, and inputting the uniform brightness environment image into a convolutional neural network for processing; Based on the convolutional neural network, the uniform brightness environment image is analyzed, the sensitive image area in the uniform brightness environment image is determined, and the environment detection is performed on the sensitive image area to obtain the preliminary image features of the uniform brightness environment image; Determine key information based on feature extraction requirements, and adapt parameters of YOLOv8n's attention mechanism based on key information; Performing attention analysis on the preliminary image features based on the parameter adaptation results, obtaining the attention weights corresponding to different feature positions in each preliminary image feature, and multiplying the attention weights with different feature positions in the preliminary image features to obtain weighted image features; Key feature position points are selected based on weighted image features, and the key feature position points are determined as the final selected feature points.

6. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 2, the feature points are distinguished into dynamic points and static points based on the sparse optical flow method, and the dynamic points are removed based on the distinction results, including: Acquire the obtained environment image sequence, and split the environment image sequence to obtain each frame of the environment image; Select two adjacent frames of environmental images from the environmental image sequence, and locate the same feature point in the two adjacent frames of environmental images. At the same time, based on the positioning result, perform constant processing on the pixel value brightness of the same feature point in the two adjacent frames of environmental images; Based on the locking result, the time t of the previous frame environment image and the time t+Δt of the next frame environment image are extracted. At the same time, the pixel point of the previous frame environment image is set to I, and the pixel coordinates of I are (x, y, t), and the pixel point position of the next frame is (x+Δx, y+Δy, t+Δt); Based on the constant brightness of pixel values, we get: I(x,y,t)=I(x+Δx,y+Δy,t+Δt); Perform Taylor series expansion on the right side of the above formula: Among them, α is the higher-order infinitesimal of Taylor series expansion; Combining the above formulas, we can get: Assume that at time , the speed of the pixels on the x-axis and y-axis is: At the same time, the pixel gradient I of the two adjacent frames of environmental images x and I y Determine, at the same time, the brightness value I at time t t to determine; Based on I x ,I y and I t Transform the above formula: I x V x +I y V y =-I t ; According to the principle of spatial consistency, take the extracted feature point as the center, take a square window with a side length of h = 3, use a 3*3 window, take 9 pixels, and establish 9 equations in the following form; The least square method is used to solve the above formula, and the average streamer vector is calculated using the solution obtained. The average streamer vector is: Among them, v x and v y is the average streamer value in the horizontal and vertical directions, then the equation for filtering the dynamic points to be removed is as follows: Wherein, k is the limit value of the dynamic feature point. If it exceeds the limit value, it is a dynamic feature point, otherwise it is a static feature point. Distinguishing the feature points into dynamic points and static points based on the above formula, and locking the first dynamic point set and the static point set based on the differentiation result; At the same time, continuous image frames are retrieved from a preset image library, and tracking annotations of different moving objects in the continuous image frames are extracted; Based on the tracking and annotation results, the appearance features of different moving objects in different image frames are extracted respectively, and the appearance features of the moving objects in different image frames are subjected to correlation analysis from the starting point to the ending point based on the time series to determine the change features of the appearance features of the moving objects, wherein the change features include the displacement of the moving objects relative to the surrounding environment and the change amount of their own morphology; Based on the deep learning model, the changing features of the appearance features of the moving object are learned, and a dynamic point detection model is obtained based on the learning results; Extracting structural parameters of the dynamic point detection model, dividing the structural parameters into branches based on the business process of the dynamic point detection model, and determining model weights of different parameters in each branch based on the branch division; Determine the prunable parameters based on the model weights, and prune the prunable parameters. At the same time, compress the required number of bits of the parameters based on the pruning results to obtain a lightweight dynamic point detection model. Analyzing the static point set based on the lightweight dynamic point detection model, and obtaining a second dynamic point set based on the analysis result; The first dynamic point set and the second dynamic point set are aggregated, and a elimination operation is performed on the aggregated dynamic point set.

7. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 2, channel feature enhancement is performed on static points, including: Obtain the static point, and perform feature extraction on the static point and surrounding pixels in the environment image to obtain multi-channel features corresponding to the static point; Analyze each channel feature to obtain the distribution and weight of multi-channel features, and determine the difference enhancement strategy for multi-channel features based on the distribution and weight; Based on the difference enhancement strategy, personalized channel enhancement processing is performed on multi-channel features, and the personalized channel enhancement processing results of different channel features are associated and summarized to complete the channel feature enhancement processing of static points.

8. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 3, the robot is positioned and the environment map is constructed based on the channel feature enhancement results of the static points, including: Obtain the channel feature enhancement results of the static points and extract the original position information of the static points in the environment image sequence; Determine the distribution characteristics of each static point based on the original position information, and perform co-location mapping on the channel feature enhancement results of each static point based on the distribution characteristics; Based on the same-position mapping results, boundary nested features of channel feature enhancement results of adjacent static points are obtained, and based on the boundary nested feature pairs, the joint seams of channel feature enhancement results of adjacent static points are determined; The channel feature enhancement results of adjacent static points are spliced ​​based on the splicing seams, and the depth of field information corresponding to the channel enhancement results of different static points is extracted based on the splicing results; Obtain the configuration parameters of the robot's onboard camera, and parse the depth of field information based on the configuration parameters to obtain the position and orientation of the robot's onboard camera relative to different static points; The robot is positioned based on the position and orientation of the robot's onboard camera relative to different static points to obtain the robot's position coordinates in the current environment; The splicing results of the channel feature enhancement results of adjacent static points are visualized in a visualization coordinate system, and the relative directions and distance parameters between different static points are determined based on the visualization results; At the same time, based on the splicing result of the channel feature enhancement results of adjacent static points, feature selection is performed on the channel feature enhancement result of each static point to obtain the feature representation of each static point; The relative directions and distance parameters between different static points and the characteristic representation of each static point are used to construct an environmental map to obtain a map frame. The position of the robot is marked in the map frame according to the position coordinates of the robot in the current environment to obtain the final environmental map.

9. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 3, the robot's own positioning and environment map are dynamically updated based on the robot's operating status, including: Obtain the running status of the robot, and when the position of the robot is determined to have changed based on the running status, synchronously trigger the robot's onboard camera to reacquire the environment image sequence; Recheck the reacquired environment image sequence, and determine the change parameters of the robot's own positioning and the channel characteristics of the static point based on the recheck results; Determine the update nodes for the robot's own positioning and the environment map and the update parameters corresponding to the update nodes based on the change parameters of the channel characteristics of the robot's own positioning and the static points; The robot's own positioning and environment map are dynamically updated based on the update nodes and the update parameters corresponding to the update nodes.

10. The method for constructing an environment map based on dynamic vision according to claim 1, characterized in that: In step 3, the robot's own positioning and environment map are dynamically updated based on the robot's operating status, including: Obtain a dynamically updated real-time environment map, and determine the robot's direction of travel based on the real-time environment map and the environment map at the previous moment; At the same time, the distribution of obstacles in the direction of travel of the robot is determined based on the real-time environment map, and a guidance route for the robot in the direction of travel is generated based on the obstacle distribution; A travel control instruction is generated based on the guidance route, and the robot is controlled to travel according to the guidance route based on the travel control instruction.

Images