Cloud edge collaborative visual slam method and system based on feature point prediction

Abstract

The invention discloses a cloud edge collaborative visual slam method and system based on feature point prediction. The method comprises the following steps: adding a plurality of initial continuous frames of images into a sliding window; feature point extraction is carried out on all pixels; establishing a quadtree as prior information of a corresponding frame image, and merging and establishing global prior information; predicting feature point distribution of a next frame of image by using global prior information and adopting a multi-step Markov probability model; adjusting the branches in the new quadtree; and adding the next frame of image into the sliding window, removing the earliest frame of image and the corresponding prior information, and predicting the feature point distribution of the subsequent frame of image by using the image in the sliding window and the corresponding prior information. The method considers that the camera has continuity in motion in the space and the feature points between the front frame and the rear frame have spatial relation, and by constructing the multi-step Markov probability model, the calculation amount is reduced on the premise of ensuring the positioning and mapping effects, and the calculation resource consumption of a CPU (Central Processing Unit) is greatly saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot vision, and in particular relates to a cloud-edge collaborative vision SLAM method and system based on feature point prediction. Background Art

[0002] With the widespread adoption of smart devices and the rapid development of technologies such as mobile robots, drones, and augmented reality, simultaneous localization and mapping (SLAM) technology faces new challenges and opportunities. Traditional SLAM systems typically rely on a single device to complete all computing tasks, placing extremely high demands on resource-constrained mobile devices. To achieve stable and efficient SLAM functionality within the constraints of limited computing power, storage space, and battery life, cloud-edge collaborative SLAM architectures have emerged and become a hot research topic.

[0003] Cloud-edge collaboration (SLAM) rationally distributes computing tasks, offloading compute-intensive workloads to the cloud while retaining the necessary real-time processing capabilities on edge devices, thereby striking a balance between system performance and resource consumption. This architecture is particularly suitable for small devices equipped with low-power processors, such as smartphones, drones, and service robots. Although these devices have limited computing power, cloud-edge collaboration can overcome the limitations of local hardware and achieve a SLAM experience close to that of high-performance computing platforms.

[0004] In practical applications, low-computing-power devices have put forward many requirements for the cloud-edge collaborative SLAM architecture, including reducing the amount of computing and CPU burden as much as possible. Summary of the Invention

[0005] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology and provide a cloud-edge collaborative visual SLAM method and system based on feature point prediction. Feature points are extracted based on quadtree and Markov properties, which reduces the amount of calculation while ensuring the positioning and mapping effects.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: One aspect of the present invention provides a cloud-edge collaborative visual SLAM method based on feature point prediction, comprising the following steps: Add the initial continuous frames of images into the sliding window; Extract feature points from all pixels of several initial consecutive frames of images; Build a quadtree for the feature points extracted from each frame; Save the quadtree as the prior information of the corresponding frame image; Merge the prior information corresponding to several initial consecutive frames of images and establish a new quadtree as the global prior information; The global prior information is used to predict the distribution of feature points of the next frame image using a multi-step Markov probability model; Adjust the branches in the new quadtree according to the actual feature point distribution of the next frame of image; The next frame image is added to the sliding window, and the earliest frame image and the corresponding prior information are removed. The image in the sliding window and the corresponding prior information are used to predict the distribution of feature points of subsequent frame images.

[0007] As a preferred technical solution, the process of establishing a quadtree for the feature points extracted from each frame includes the following steps: Create a root node representing the entire image area; Starting from the root node, check whether the current node has been divided. If it has not been divided and the spatial density of the feature points in the node ( N / Area )> ρ 1, then divide the current area into four sub-areas and assign the feature points to the corresponding sub-areas; N is the number of feature points of the current node, ρ 1 is the set spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area; Node center ( x c , y c ) is divided into two parts: ; The area has reached the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ ∈[6,8], stop dividing. W and H are the width and height of the image frame, Q 1- Q 4 is the four sub-areas obtained by division; The process is repeated recursively until the feature points are assigned to appropriate leaf nodes.

[0008] As a preferred technical solution, the new quadtree is established as global prior information, where the number of node feature points ,in w 1, w 2,…, w n is the weight of images in different time frames. The subscript represents the sequence number. The weight value is determined by the time sequence of the image frames. The new image frame has a larger weight, and vice versa. V 1, V 2,…, V n is the number of feature points in different time frame images, and the subscript indicates the sequence number; Remove the spatial density of feature points ( N / Area )< ρ 2 leaf nodes, where N is the number of feature points of the current node, ρ 2 is the preset spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area.

[0009] As a preferred technical solution, the method of using global prior information and adopting a multi-step Markov probability model to predict the distribution of feature points of the next frame image is specifically as follows: Define a multi-step Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,...,X n-k+1 ), P ( X n+1 | X n , X n-1 ,..., X 0) indicates the distribution state of feature points of the current image X n+1 ,and X n+1 All previous distribution states X n , X n-1 ,..., X 0 are all related; P ( X n+1 | X n , X n-1 ,..., X n-k+1 ) means X n+1 Restricted to only X n+1 Previously limited k Status X n , X n-1 ,..., X n-k+1 related; pass X n , X n-1 ,..., X n-k+1 predict n +1 image feature point distribution status : ; in E [∙] is the maximum likelihood.

[0010] As a preferred technical solution, X n+1 Only with X n+1 Previously limited k= 3 states X n , X n-1 ,..., X n-2Related, we get the third-order Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-2 ); Assume that the state transition of the feature point distribution between the previous and next image frames in a short time satisfies the linear relationship: X n+1 = aX n + bX n-1 + cX n-2 +...+ X 1+ ε , ε is Gaussian noise, then: ; in, a, b, c is the linear coefficient.

[0011] As a preferred technical solution, the actual feature point distribution of the following frame of image adjusts the branches in the new quadtree, specifically comprising the following steps: Extract feature points: extract feature points from the image corresponding to the leaf node, count the number, and update the leaf node data; Node splitting: spatial density of feature points within a node ( N / Area )> ρ 1, then divide the current area into four sub-areas, assign the feature points to the corresponding sub-areas, and recursively process until the feature points are assigned to the appropriate leaf nodes; N is the number of feature points of the current node, ρ 1 is the set spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 xmin 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area; when the area reaches the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ ∈[6,8], stop dividing. W and H is the width and height of the image frame; Local reconstruction: If a parent node P There are only leaf nodes, and the spatial density of feature points of each leaf node ( N’ / Area' )< ρ 3, then complete the parent node P The four leaf nodes of N’ is the number of feature points of leaf nodes, Area' is the area of ​​the leaf node, ρ 3 is the set leaf node space density threshold; Extract feature points from the completed leaf node area. If the parent node is completed P If the number of feature points of each leaf node does not reach the set threshold T2, then the parent node P Become a leaf node and reconstruct it locally upwards; if the parent node is completed P If there is at least one leaf node whose number of feature points reaches the set threshold T2, then P The parent node P 1 is the root node, executing the step of establishing a quadtree for the feature points extracted from each frame; The upward local reconstruction is specifically as follows: P The parent node P 1 Is there only a leaf node? If so, P 1. Execute the local reconstruction step.

[0012] As a preferred technical solution, the number of quadtree layers is set at the beginning of the step of adjusting the branches in the new quadtree, and the corresponding leaf nodes whose number of feature points is less than the set threshold T2 are assigned a value of -1, indicating the abandoned area. The abandoned nodes do not play a role when merging.

[0013] Another aspect of the present invention further provides a cloud-edge collaborative visual SLAM system based on feature point prediction, comprising an edge terminal, an edge server, and a cloud server that communicate with each other through a communication module; The edge end applies the above-mentioned cloud-edge collaborative visual SLAM method based on feature point prediction to obtain the distribution of image feature points and construct and store submaps; The edge server is used to restore the data uploaded by the edge end according to the SLAM visual odometry to obtain the real-time path and roadside points, and optimize the data through the SLAM backend to obtain the edge map; store the edge map; The cloud server is used to merge the edge maps uploaded by the edge servers as a global map, and optimize and store the global map.

[0014] Another aspect of the present invention also provides a cloud-edge collaborative visual SLAM device based on feature point prediction, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and at least one processor are interconnected through a line; at least one processor calls the instructions in the memory to enable the cloud-edge collaborative visual SLAM device based on feature point prediction to execute a cloud-edge collaborative visual SLAM method based on feature point prediction as described above.

[0015] Another aspect of the present invention further provides a storage medium storing a program, which, when executed by a processor, implements the above-mentioned cloud-edge collaborative visual SLAM method based on feature point prediction.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) The present invention takes into account the continuity of the camera's movement in space and the spatial connection of the feature points between the previous and next frames. By constructing a multi-step Markov probability model and then constructing an algorithm, the CPU computing resource consumption is greatly saved.

[0017] (2) The present invention adapts to the needs of low-computing power devices through cloud-edge collaboration, extracts and manages visual odometry feature points while ensuring system robustness, and overcomes the problem in the prior art that extracting feature points (for example, orb feature points) from the entire image is very resource-consuming and time-consuming. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a cloud-edge collaborative visual SLAM method based on feature point prediction according to an embodiment of the present invention; Figure 2 is a flow chart of establishing a quadtree according to an embodiment of the present invention; Figure 3 Schematic diagram of the correspondence between a quadtree and an image according to an embodiment of the present invention; Figure 4 1 is a schematic diagram of quadtree weighted merging according to an embodiment of the present invention; Figure 52 is a schematic diagram of dynamic adjustment of a priori quadtree according to an embodiment of the present invention; Figure 6 is a schematic diagram of window sliding according to an embodiment of the present invention; Figure 7 This is a structural diagram of a cloud-edge collaborative vision SLAM system based on feature point prediction in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a cloud-edge collaborative visual SLAM method based on feature point prediction, including the following steps: S1. Input image: Add several initial consecutive frames of images into the sliding window.

[0021] S2. Extraction of global feature points of the image: Extract feature points from all pixels of several initial consecutive frames of images.

[0022] S3, establish quadtree: establish quadtree to manage image feature points for feature points extracted from each frame. Figure 2 As shown, the corresponding relationship between the image and the quadtree is as follows Figure 3 shown.

[0023] A quadtree is a tree-like data structure used to efficiently manage two-dimensional spaces. It recursively partitions a region into four equal-sized subregions, with each node having a maximum of four children. Quadtrees are widely used in image processing, enabling fast spatial queries and collision detection. Their hierarchical structure optimizes storage and retrieval efficiency, making them particularly well-suited for processing unevenly distributed data.

[0024] Furthermore, step S3 includes the following steps: S301, initializing the root node: creating a root node representing the entire image area.

[0025] S302, insert feature points: starting from the root node, check whether the current node has been divided. If it has not been divided and the spatial density of feature points in the node (number of points per unit area) N / Area )> ρ1 then divide the current area into four sub-areas and assign the feature points to the corresponding sub-areas; N is the number of feature points of the current node, ρ 1 is the set spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area.

[0026] Partitioning rules (based on node center ( x c , y c ) as the boundary): ; The area has reached the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ ∈[6,8], stop dividing. W and H are the width and height of the image frame, Q 1- Q 4 represents the four sub-areas obtained by division.

[0027] S303: Recursively process until the feature points are assigned to appropriate leaf nodes.

[0028] S4. Store as prior information: The quadtree corresponding to the image frame in the sliding window is stored as prior information.

[0029] S5. Establish a quadtree with prior information: Merge the quadtrees corresponding to the images in the sliding window to establish a new quadtree as global prior information. The number of node feature points is ,in w 1, w 2,…, w n is the weight of images in different time frames. The subscript represents the sequence number. The weight value is determined by the time sequence of the image frames. The new image frame has a larger weight, and vice versa. V 1,V 2,…, V n is the number of feature points in different time frame images, and the subscript indicates the sequence number. N / Area )< ρ 2's leaf node, ρ 2 is the preset spatial density threshold, which is less than or equal to ρ 1. The specific value is adjusted according to the scenario and experience. The process of merging quadtrees A and B into C is as follows Figure 4 As shown, let the spatial density of the feature points of the leaf node a2 of the quadtree A be ( N / Area )< ρ 2. The nodes at the same position in quadtree A and quadtree B are weighted and merged to the corresponding position in quadtree C. During the merging process, the nodes with zero feature points do not play a role. For example, there is no leaf node a in quadtree B. l , a2, a3, a4, so the number of corresponding feature points is 0, and the weight w•v =0, then in the third layer only the leaf nodes a1, a3, a4 of quadtree A and the leaf nodes b1 and b2 of quadtree B are retained in quadtree C.

[0030] S6. Using global prior information, a multi-step Markov probability model is used to predict the distribution of feature points of the next frame image.

[0031] The Markov property refers to the characteristic of a random process that, given the current state, the future state depends only on the current state and is independent of the past state. Mathematical expression: For a discrete-time random process { X n}, if: P ( X n+1 = x | X n = x n , X n-1 = x n-1 ,..., X 0= x 0)= P ( X n+1 = x | X n = x n ); The process is said to have Markov property.

[0032] When the state at a certain moment depends on the previous k When the state at a certain moment (not just the current moment) is k order Markov process.

[0033] The mathematical expression is as follows: P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-k+1 ); That is, the conditional probability of the future state depends on the recent k A historical state.

[0034] Define a multi-step Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-k+1 ), P ( X n+1 | X n , X n-1 ,..., X 0) indicates the distribution state of feature points of the current image X n+1 ,and X n+1 All previous distribution states X n , X n-1 ,..., X 0 are all related;P ( X n+1 | X n , X n-1 ,..., X n-k+1 ) means X n+1 Restricted to only X n+1 Previously limited k Status X n , X n-1 ,..., X n-k+1 related; pass X n , X n-1 ,..., X n-k+1 predict n +1 image feature point distribution status : ; in E [∙] is the maximum likelihood.

[0035] Furthermore, in this embodiment, let X n+1 Only with X n+1 Previously limited k= 3 states X n , X n-1 ,..., X n-2 Related, we get the third-order Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-2 ); Assume that the state transition of the feature point distribution between the previous and next image frames in a short time satisfies the linear relationship:X n+1 = aX n + bX n-1 + cX n-2 +...+ X 1+ ε , ε is Gaussian noise, then: ; in, a, b, c is the linear coefficient.

[0036] More specifically, a, b, c It can be expressed as: ; ; .

[0037] S7, dynamic adjustment of the prior quadtree: adjust the branches in the new quadtree in step S5 according to the actual distribution of the image feature points of the next frame of image. Figure 5 shown.

[0038] Furthermore, step S7 includes the following steps: S701, extract feature points: extract feature points from the image corresponding to the leaf node, count the number, and update the leaf node data.

[0039] S702, node splitting: spatial density of feature points within the node (number of points per unit area) N / Area )> ρ 1, then divide the current area into four sub-areas, assign the feature points to the corresponding sub-areas, and recursively process until the feature points are assigned to the appropriate leaf nodes.

[0040] The area has reached the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ Stop dividing when ∈[6,8].

[0041] S703, local reconstruction: If a parent node P There are only leaf nodes, and the spatial density of feature points of each leaf node ( N’ / Area' )< ρ 3, then complete the parent node P The four leaf nodes of N’is the number of feature points of leaf nodes, Area' is the area of ​​the leaf node, ρ 3 is the set leaf node space density threshold, which is less than or equal to ρ 1. The specific value is adjusted according to the scenario and experience. Extract feature points from the completed leaf node area. If the parent node is completed P If the number of feature points of each leaf node does not reach the set threshold T2, then the parent node P Become a leaf node and reconstruct it locally upwards; if the parent node is completed P If there is at least one leaf node whose number of feature points reaches the set threshold T2, then P The parent node P 1 is the root node, execute step S3 to build a quadtree; The upward local reconstruction is specifically as follows: P The parent node P 1 Is there only a leaf node? If so, P 1Execute step S703.

[0042] Furthermore, since adding and removing branches of a quadtree is bloated and complicated in actual operation, in this embodiment, the number of quadtree layers is set at the beginning of the step of adjusting the branches in the new quadtree, and the corresponding leaf nodes whose number of feature points of the leaf nodes is less than the set threshold T2 are assigned a value of -1, indicating an abandoned area, and the abandoned nodes do not play a role when merged.

[0043] S8, window sliding: remove the oldest image frame in the sliding window and add the new image frame, such as Figure 6 shown.

[0044] This approach processes sequence data by dynamically maintaining a fixed-size sliding window of local data. Its core principle is that as the sequence progresses, the sliding window, like a sliding viewport, discards old data and incorporates new data, enabling real-time analysis and computation of continuous data. This approach cleverly strikes a balance between time and space efficiency, avoiding the overhead of processing the entire dataset while capturing local features of the data. It is particularly suitable for scenarios such as time series analysis, real-time signal processing, and streaming data mining.

[0045] Example 2: like Figure 7 As shown, in this embodiment, a cloud-edge collaborative visual SLAM system based on feature point prediction is provided, which includes an edge terminal, an edge server and a cloud server that communicate with each other through a communication module; The edge end applies a cloud-edge collaborative visual SLAM method based on feature point prediction of the above embodiment to obtain the distribution of image feature points and construct and store a submap; The edge server is used to restore the data uploaded by the edge end according to the SLAM visual odometry to obtain the real-time path and roadside points, and optimize the data through the SLAM backend to obtain the edge map; store the edge map; The cloud server is used to merge the edge maps uploaded by the edge servers as a global map, and optimize and store the global map.

[0046] It should be noted here that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure is divided into different functional modules to complete all or part of the functions described above. The system is a cloud-edge collaborative visual slam method based on feature point prediction applied to the above embodiment.

[0047] Example 3: In this embodiment, a cloud-edge collaborative visual SLAM device based on feature point prediction is provided, including: a memory and at least one processor, the memory storing instructions, the memory and at least one processor being interconnected through a line; at least one processor calls the instructions in the memory to enable the cloud-edge collaborative visual SLAM device based on feature point prediction to execute a cloud-edge collaborative visual SLAM method based on feature point prediction as described in the above embodiment.

[0048] Example 4: In this embodiment, a storage medium is provided, which stores a program. When the program is executed by a processor, a cloud-edge collaborative visual SLAM method based on feature point prediction in the above embodiment is implemented, specifically: Add the initial continuous frames of images into the sliding window; Extracting feature points from all pixels of the image of several initial consecutive frames; Build a quadtree for the feature points extracted from each frame; Save the quadtree as the prior information of the corresponding frame image; Merging the prior information corresponding to the images of the initial consecutive frames to establish a new quadtree as global prior information; The global prior information is used to predict the distribution of feature points of the next frame image using a multi-step Markov probability model; Adjust the branches in the new quadtree according to the actual feature point distribution of the next frame of image; The image of the next frame is added to the sliding window, and the earliest frame image and the corresponding prior information are removed. The image in the sliding window and the corresponding prior information are used to predict the distribution of feature points of subsequent frame images.

[0049] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0050] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A cloud-edge collaborative visual SLAM method based on feature point prediction, characterized in that: The steps include: Add the initial continuous frames of images into the sliding window; Extract feature points from all pixels of several initial consecutive frames of images; Build a quadtree for the feature points extracted from each frame; Save the quadtree as the prior information of the corresponding frame image; Merge the prior information corresponding to several initial consecutive frames of images and establish a new quadtree as the global prior information; The global prior information is used to predict the distribution of feature points of the next frame image using a multi-step Markov probability model; Adjust the branches in the new quadtree according to the actual feature point distribution of the next frame of image; The next frame image is added to the sliding window, and the earliest frame image and the corresponding prior information are removed. The image in the sliding window and the corresponding prior information are used to predict the distribution of feature points of subsequent frame images.

2. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 1 is characterized in that: The process of establishing a quadtree for the feature points extracted from each frame includes the following steps: Create a root node representing the entire image area; Starting from the root node, check whether the current node has been divided. If it has not been divided and the spatial density of the feature points in the node ( N / Area )> ρ 1, then divide the current area into four sub-areas and assign the feature points to the corresponding sub-areas; N is the number of feature points of the current node, ρ 1 is the set spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area; Node center ( x c , y c ) is divided into two parts: ； The region has reached the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ ∈[6,8], stop dividing. W and H are the width and height of the image frame, Q 1- Q 4 is the four sub-areas obtained by division; The process is repeated recursively until the feature points are assigned to appropriate leaf nodes.

3. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 1 is characterized in that: The new quadtree is established as global prior information, where the number of node feature points ,in w 1, w 2,…, w n is the weight of images in different time frames. The subscript represents the sequence number. The weight value is determined by the time sequence of the image frames. The new image frame has a larger weight, and vice versa. V 1, V 2,…, V n is the number of feature points in different time frame images, and the subscript indicates the sequence number; Remove the spatial density of feature points ( N / Area )< ρ 2 leaf nodes, where N is the number of feature points of the current node, ρ 2 is the preset spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area.

4. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 1 is characterized in that: The method of using the global prior information and adopting the multi-step Markov probability model to predict the distribution of feature points of the next frame image is specifically as follows: Define a multi-step Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-k+1 ), P ( X n+1 | X n , X n-1 ,..., X 0) indicates the distribution state of feature points of the current image X n+1 ,and X n+1 All previous distribution states X n , X n-1 ,..., X 0 are all related; P ( X n+1 | X n , X n-1 ,..., X n-k+1 ) means X n+1 Restricted to only X n+1 Previously limited k Status X n , X n-1 ,..., X n-k+1 related; pass X n , X n-1 ,..., X n-k+1 predict n +1 image feature point distribution status : ； in E [∙] is the maximum likelihood.

5. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 4 is characterized in that: make X n+1 Only with X n+1 Previously limited k= 3 states X n , X n-1 ,..., X n-2 Related, we get the third-order Markov probability model P ( X n+1 | X n , X n-1 ,..., X 0)= P ( X n+1 | X n , X n-1 ,..., X n-2 ); Assume that the state transition of the feature point distribution between the previous and next image frames in a short time satisfies the linear relationship: X n+1 = aX n + bX n-1 + cX n-2 +...+ X 1+ ε , ε is Gaussian noise, then: ； in, a, b, c is the linear coefficient.

6. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 1 is characterized in that: The actual feature point distribution of the next frame of image adjusts the branches in the new quadtree, specifically comprising the following steps: Extract feature points: extract feature points from the image corresponding to the leaf node, count the number, and update the leaf node data; Node splitting: spatial density of feature points within a node ( N / Area )> ρ 1, then divide the current area into four sub-areas, assign the feature points to the corresponding sub-areas, and recursively process until the feature points are assigned to the appropriate leaf nodes; N is the number of feature points of the current node, ρ 1 is the set spatial density threshold, Area =( x max - x min )×( y max - y min ) is the area of ​​the current node, x max 、 x min 、 y max and y min The maximum and minimum values ​​of the horizontal and vertical coordinates of the current area; when the area reaches the minimum size limit s = W / (2 n’ )× H / (2 n’ ), n’ ∈[6,8], stop dividing. W and H is the width and height of the image frame; Local reconstruction: If a parent node P There are only leaf nodes, and the spatial density of feature points of each leaf node ( N’ / Area' )< ρ 3, then complete the parent node P The four leaf nodes of N’ is the number of feature points of leaf nodes, Area' is the area of ​​the leaf node, ρ 3 is the set leaf node space density threshold; Extract feature points from the completed leaf node area. If the parent node is completed P If the number of feature points of each leaf node does not reach the set threshold T2, then the parent node P Become a leaf node and reconstruct it locally upwards; if the parent node is completed P If there is at least one leaf node whose number of feature points reaches the set threshold T2, then P The parent node P 1 is the root node, executing the step of establishing a quadtree for the feature points extracted from each frame; The upward local reconstruction is specifically as follows: P The parent node P 1 Is there only a leaf node? If so, P 1. Execute the local reconstruction step.

7. The cloud-edge collaborative visual SLAM method based on feature point prediction according to claim 6 is characterized in that: At the beginning of the step of adjusting the branches in the new quadtree, the number of quadtree layers is set, and the corresponding leaf nodes whose number of feature points is less than the set threshold T2 are assigned a value of -1, indicating an abandoned area. The abandoned nodes do not play a role when merging.

8. A cloud-edge collaborative visual SLAM system based on feature point prediction, characterized in that: including edge terminals, edge servers, and cloud servers that communicate with each other through communication modules; The edge end applies a cloud-edge collaborative visual SLAM method based on feature point prediction according to any one of claims 1 to 7 to obtain the distribution of image feature points and construct and store a submap; The edge server is used to restore the data uploaded by the edge end according to the SLAM visual odometry to obtain the real-time path and roadside points, and optimize the data through the SLAM backend to obtain the edge map; store the edge map; The cloud server is used to merge the edge maps uploaded by the edge servers into a global map, and optimize and store the global map.

9. A cloud-edge collaborative visual SLAM device based on feature point prediction, characterized in that: include: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line; At least one of the processors calls the instructions in the memory so that the cloud-edge collaborative vision SLAM device based on feature point prediction executes the cloud-edge collaborative vision SLAM method based on feature point prediction as described in any one of claims 1-7.

10. A storage medium storing a program, characterized in that: When the program is executed by the processor, the cloud-edge collaborative visual SLAM method based on feature point prediction described in any one of claims 1-7 is implemented.

Images