CN113110482B - Indoor environment robot exploration method and system based on priori information heuristic method - Google Patents

Abstract

The invention relates to a heuristic indoor environment robot exploration method and system based on prior information, which comprises the following steps: the robot acquires data of surrounding environment information through a sensor carried by the robot; updating a part of map into a known area based on data of the surrounding environment information to obtain an updated map; performing boundary extraction on the updated map by using two fast search random trees to obtain RRT boundary points; identifying a heuristic object and carrying out position estimation on the heuristic object, constructing a prior area by taking the position of the heuristic object as a reference, and extracting boundary points in the prior area to obtain room boundary points; based on the RRT boundary points and the room boundary points, the robot performs indoor environment exploration. The robot preferentially explores the environment in the prior area, can preferentially explore one room area and then turn to other areas, effectively reduces backtracking in the exploration process, and improves exploration efficiency.

Description

Indoor environment robot exploration method and system based on prior information heuristic

technical field

The invention relates to the technical field of robot exploration, in particular to an indoor environment robot exploration method and system based on a priori information heuristic.

Background technique

The breakthrough of artificial intelligence technology has brought huge opportunities to the research of mobile service robots. The application scenarios and service models of intelligent public service robots continue to expand, driving the rapid growth of my country's service robot market. Autonomous exploration of mobile robots refers to the process that robots build a complete environment map by moving in a new environment without any prior knowledge. At present, the main autonomous exploration methods can be roughly divided into two categories: grid map-based exploration methods and feature map-based exploration methods. Among them, the method based on the grid map has rich representation information and high resolution, which is conducive to the construction of the information gain model of the boundary point. Therefore, we also use the method based on the grid map to carry out research on the autonomous exploration of robots. The mainstream research method based on grid map is the exploration method based on boundary points, which divides the detection space into known areas and unknown areas, and guides the robot to collect information to update the map. In order to detect the environment more effectively, this detection method mainly focuses on how to detect and select boundaries. The choice of boundaries directly affects the exploration efficiency. At present, most of the existing robot exploration strategies focus on how to design a boundary point information gain model to select a boundary point with a larger profit value. However, this information gain model only considers the cost of the exploration path at the current moment and the information gain brought by updating a small part of the map around the boundary point, while ignoring the geometric continuity of obstacles in the environment. For example, in an indoor environment, after a robot enters a room to explore, it generally has two actions: ① After completing the exploration of the room area, it turns to the outside of the room to explore; ② The room area has not been explored and turns to the outside of the room, and then enters again. Room exploration. Obviously, the second action is more time-consuming and less efficient; the expected benefit of the first action that continues to explore downwards is greater. However, current robot exploration strategies evaluate candidate boundary points independently, ignoring the potential benefits of environmental structure, so the backtracking phenomenon of exploration strategies like the second action often occurs.

SUMMARY OF THE INVENTION

Therefore, the technical problem to be solved by the present invention is to overcome the technical defect of the backtracking phenomenon in the process of robot exploration in the prior art, ignoring the potential benefits of the environmental structure.

In order to solve the above technical problems, the present invention provides an indoor environment robot exploration method based on a priori information heuristic, including the following steps:

S1. The robot collects the data of the surrounding environment information through the sensor carried by itself;

S2. Update a part of the map to a known area based on the data of the surrounding environment information, and obtain the updated map;

S3. Use two fast search random trees to extract the boundary of the updated map to obtain the RRT boundary point;

S4. Identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract boundary points in the priori region, and obtain room boundary points; wherein, the heuristic object is a door;

S5. Based on the RRT boundary point and the room boundary point, the robot explores the indoor environment.

Preferably, the S5 includes:

S51. When the room boundary point exists, select the room boundary point with the largest income value as the target point, and when there is no room boundary point, select the RRT boundary point with the largest income value as the target point;

S52, guide the robot to navigate to the target point.

Preferably, the profit value of the boundary point R1 _f =w ₁ *I _f -w ₂ *N _f ,

Among them, If is the information gain, the information gain refers to the number of unknown grids within the information gain radius _r =1 of the centroid point,

N _f is the path cost, and the path cost refers to the Euclidean distance between the robot's current position and the position of the center of mass;

w ₁ and w ₂ are custom weights, which are constants.

Preferably, after the S5, it also includes:

S6. When no boundary point is detected in the a priori area, it means that the area has been explored, and the model of the prior area is destroyed to form the next heuristic area;

S7, loop S1-S6 until the robot explores the entire environment and obtains a grid map.

Preferably, the S3 includes:

S31, in the initialization phase, add the starting point to the tree structure as the root node, wherein the starting points of the two trees are artificially set in the free area of the map;

S32, randomly sprinkle points in the map area as candidate points;

S33. If the candidate point is in the known area, traverse all existing nodes on the tree structure, select the node closest to the candidate point as the nearest neighbor point, and use the connection line from the nearest neighbor point to the candidate node as the growth direction. If the distance between the point and the candidate node exceeds the preset step size, the nearest point grows one step size along the growth direction, and the reached point is used as the growth point. If the distance does not exceed the step size, the candidate point is used as the growth point;

If the candidate point is in the unknown area, first find the nearest tree node of the candidate point, and the connection line from the nearest point to the candidate point is used as the growth direction, and the nearest point grows forward along the growth direction, and the place where it reaches the boundary is used as the growth direction. boundary point.

Preferably, the S33 further includes:

S34. Perform collision detection on the map for the connection between the growth point and the candidate node, specifically including:

Traverse all grid points on the connection between the growth point and the candidate node, and judge the grid state of the grid point;

If the state of the grid point is occupied, the collision detection fails, and returns to S32 to collect points again;

If the connection between the growing point and the candidate node does not encounter obstacles, add the connection between the candidate point, the growing point and the candidate node to the tree structure.

Preferably, in said S4, identifying heuristic objects and performing position estimation on them, including:

Construct a lightweight network, complete the recognition of heuristic objects based on the deep learning method, and obtain the coordinate information of the heuristic objects; wherein, the lightweight network includes a convolution layer, an inverse residual block, a pooling layer and an SSP layer .

Preferably, in the S4, a priori region is constructed based on the position of the heuristic object, including:

When the robot recognizes the heuristic object, if the robot's position is below the heuristic object, the estimated room area is above the heuristic object. If the robot's position is above the heuristic object, the estimated room area below the position of the heuristic object;

Wherein, the length of the estimated room area is the position of the heuristic object extending to both sides by a length a, and the width of the estimated area is the position of the heuristic object extending backward by a length 2b. Among them, the parameters a and b are set by experience.

Preferably, in said S4, extracting boundary points in the priori region to obtain room boundary points, including the following steps:

Perform binarization processing on the image of the prior area to obtain a binarized image, in which the obstacles of the binarized image are white, and the rest of the area is black;

Perform color flipping on the binarized image to obtain a color flipped image, wherein the obstacles in the color flipped image are black, and the rest of the area is white;

The Canny operator is used to detect the edge of the binarized image. In the detection result, the edge of the image is set to white, and the rest of the area is black;

Perform a bitwise AND operation on the binarized image and the color-flipped image to remove excess white edges to obtain a final image, where the boundary between the known area and the unknown area in the final image consists of straight lines;

Extract the center of gravity of the line, which is the room boundary point.

The invention also discloses an indoor environment robot exploration system based on a priori information heuristic, which is characterized by comprising:

a data acquisition module, the data acquisition module is used for the robot to collect the data of the surrounding environment information through the sensor carried by the robot;

A positioning and mapping module, the positioning and mapping module updates a part of the map as a known area based on the data of the surrounding environment information, and obtains the updated map;

RRT boundary point extraction module, the RRT boundary point extraction module uses two fast search random trees to perform boundary extraction on the updated map to obtain RRT boundary points;

The room boundary point extraction module is used to identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract the boundary point in the priori region, and obtain the room border point;

An environment exploration module, the environment exploration module is based on the RRT boundary point and the room boundary point, and the robot performs indoor environment exploration.

The above-mentioned technical scheme of the present invention has the following advantages compared with the prior art:

By introducing a heuristic prior information exploration module, when the robot recognizes the heuristic object, it will first explore the environment in the prior area, so that the robot can first explore a room area and then turn to other areas, effectively reducing the amount of time in the exploration process. Retrospect the phenomenon and improve the efficiency of exploration.

Description of drawings

Fig. 1 is the schematic diagram of the indoor environment robot exploration method in the present invention;

Fig. 2 is the flow chart of robot exploration in the present invention;

Fig. 3 is the process schematic diagram of fast search random tree extraction boundary point;

Fig. 4 is the structural schematic diagram of the structure of the prior region;

Fig. 5 is a scene-simulation environment structure diagram;

Fig. 6 is the structure diagram of the simulation environment of scene two;

FIG. 7 is a structural diagram of the simulation environment of scene three.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

Referring to FIG. 3 , the present invention discloses a method for exploring an indoor environment robot based on a priori information heuristic, including the following steps:

Step 1: The robot collects data of surrounding environment information through sensors carried by itself. At the beginning, the entire grid map is unknown, the robot is in a certain position in the environment, and the robot obtains the data of the surrounding environment information through the sensors it carries.

Step 2: Update a part of the map to a known area based on the data of the surrounding environment information, and obtain the updated map.

The localization and map construction (SLAM) module receives sensor data, updates a part of the map to a known area, and the map constructed by SLAM can in turn correct the robot's pose.

Step 3. Use two fast search random trees to extract the boundaries of the updated map to obtain RRT boundary points, including the following steps:

S31, in the initialization phase, add the starting point to the tree structure as the root node, wherein the starting points of the two trees are artificially set in the free area of the map;

S32, randomly sprinkle points in the map area as candidate points;

S33. As shown in Figure 3, we take three sampling points 1, 2, and 3 as examples to illustrate the boundary point extraction process. If the candidate point is in the known area, traverse all existing nodes on the tree structure, select the node closest to the candidate point as the nearest neighbor point, and the connection line from the nearest neighbor point to the candidate node as the growth direction. As shown in candidate point 2 in FIG. 3 , candidate point 2 is in a known region, and the dotted line with an arrow in the figure is the growth direction. If the distance between the nearest neighbor point and the candidate node exceeds the preset step size, the nearest neighbor point will grow one step size along the growth direction, and the reached point will be used as the growth point. If the distance does not exceed the step size, the candidate point will be used as the growth point. growing point;

If the candidate point is in the unknown area, first find the nearest tree node of the candidate point, and the connection line from the nearest point to the candidate point is used as the growth direction, and the nearest point grows forward along the growth direction, and the place where it reaches the boundary is used as the growth direction. boundary point.

The above method embodies the traversal from the nearest point along the growth direction, and the grid state is unknown as the boundary point. As shown in candidate point 1 in Figure 3, candidate point 2 is in an unknown area, and the dotted line with an arrow in the figure is the growth direction, then along the growth direction, the place where it reaches the boundary is used as the boundary point.

S34. Perform collision detection on the map for the connection between the growth point and the candidate node, specifically including:

Traverse all grid points on the connection between the growth point and the candidate node, and judge the grid state of the grid point;

If the state of the grid point is occupied (that is, an obstacle), the collision detection fails, and returns to S32 to collect points again;

If the connection between the growing point and the candidate node does not encounter obstacles, add the connection between the candidate point, the growing point and the candidate node to the tree structure. As shown in candidate point 3 in Figure 3, the connection line between the candidate point and the nearest point crosses the obstacle, so the collision detection fails, and the point must be re-collected.

The algorithm in the present invention uses two fast search random numbers, divided into a global tree and a local tree. The global tree extracts the boundary points through the above steps. The extraction principle and growth method of the local tree for boundary points are the same as the global tree. grow back.

In step 3, for the RRT boundary points, the RRT boundary points can be filtered and eliminated. Specifically, the detected boundary points are clustered by the mean-shift clustering algorithm to obtain the centroid points, so that some boundary points can be filtered out. Reduce computational cost. At the same time, at each moment, the grid state of the boundary point and the value in the costmap (costmap) are detected (the costmap divides each grid value between 0 and 255, and the white value is 255, which represents the idle state; The black value is 0, representing obstacles; the value between is gray, representing unknown), if the grid state is idle (indicating that the grid point has been explored) and the value in the costmap exceeds a certain threshold, it indicates The area where the grid point is located has been explored, and the grid point is also culled as an invalid point.

Step 4: Identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract boundary points in the priori region, and obtain room boundary points; wherein, the heuristic object is a door.

In the fourth step, identifying the heuristic object and estimating its position includes: constructing a lightweight network, completing the identification of the heuristic object based on a deep learning method, and obtaining the coordinate information of the heuristic object; wherein, the light The magnitude network includes convolutional layers, inverse residual blocks, pooling layers, and SSP layers. In the present invention, the improved lightweight network can quickly complete object recognition and publish the position coordinates of the door.

1. The present invention uses a deep learning-based target detection method for heuristic object recognition, which is an improved lightweight network based on YOLOv4_tiny. This lightweight network is mainly composed of convolutional layers, inverse residual blocks, pooling layers and SPP (Spatial Pyramid Pooling) layers, with a total of 42 layers, and the output is simplified to two layers. When the input size is 416×416×3, the corresponding output layers are 13×13×255 and 26×26×255, respectively. The inverse residual block in the backbone network can effectively increase the dimension of feature extraction, while the SPP layer located in the deep layer spatially fuses local and global features. Compared with YOLOv4_tiny, the detection accuracy of this lightweight network is improved by 19.2%, which is close to YOLOv4, but almost 4 times faster than YOLOv4. Overall, this lightweight network is very efficient in terms of speed and accuracy, and it is suitable for heuristic object recognition.

2. After the heuristic object is detected using the above network, the two-dimensional image coordinate information of the heuristic object (position in the depth camera coordinate system) can be obtained. Next, the mapping from two-dimensional points to three-dimensional points is realized. We assume that the coordinates of the center point P of the output heuristic object on the two-dimensional plane are (x', y'), which are in the three-dimensional coordinate system (that is, in the world coordinate system). The coordinates under the position below) are (x, y, z), and there is the following mapping relationship between them:

In order to facilitate the actual conversion calculation, we use homogeneous coordinates to represent the camera parameters, so the conversion from 2D points to 3D points can be expressed as:

The camera internal parameters (f _x , f _y , c _x , c _y ) can be obtained by subscribing to the topic published by the depth camera under ROS. The z-coordinate of the heuristic object has no effect on the construction of the prior area in the following steps, so it is not here. add a description.

In the fourth step, a priori area is constructed based on the position of the heuristic object, including: as shown in Figure 4, when the robot recognizes the heuristic object, if the robot position is below the heuristic object, the estimated The room area is above the position of the heuristic object, and if the robot is positioned above the heuristic object, the estimated room area is below the position of the heuristic object; wherein, the estimated length of the room area is the heuristic The position of the object extends to both sides by a length a, and the width of the estimated area is the position of the heuristic object that extends backward by a length of 2b. Among them, the parameters a and b are set by experience.

The present invention constructs a priori area based on the position of the heuristic object. This area follows human perception habits. The area behind the door is a room. larger than the actual area size.

In the fourth step, the boundary points are extracted in the prior area to obtain the room boundary points, including the following steps:

1. Since the original image is a grayscale image (the obstacle is black, the unknown area is gray, and the free area is white), the present invention can directly perform binarization processing, in which the threshold adopts an adaptive method, and the obstacle is set as White, the rest of the area is black;

2. Perform color inversion on the binarized image to obtain a color-reversed image, wherein the obstacles of the color-reversed image are black, and the remaining areas are white;

3. The Canny operator is used to detect the edge of the binarized image. In the detection result, the edge of the image is set to white, and the rest of the area is black;

4. Perform a bitwise AND operation on the binarized image and the color-flipped image to remove excess white edges to obtain a final image, where the boundary between the known area and the unknown area in the final image consists of a straight line ;

5. Extract the center of gravity of the line, which is the room boundary point.

For the room boundary points in step 4, the invalid room boundary points are filtered and eliminated. Specifically, the detected boundary points are clustered by the mean-shift clustering algorithm to obtain the centroid points, which can filter out a part of the boundary points and reduce the computational cost. consume. At the same time, at each moment, the grid state of the boundary point and the value in the costmap (costmap) are detected (the costmap divides each grid value between 0 and 255, and the white value is 255, which represents the idle state; The black value is 0, representing obstacles; the value between is gray, representing unknown), if the grid state is idle (indicating that the grid point has been explored) and the value in the costmap exceeds a certain threshold, it indicates The area where the grid point is located has been explored, and the grid point should also be culled as an invalid point.

Step 5. Based on the RRT boundary points and room boundary points, the robot conducts indoor environment exploration, including:

S51. When the room boundary point exists, the robot preferentially selects the room boundary point to explore, so that the robot can enter the priori area to explore preferentially after recognizing the heuristic object. When the room boundary point exists, it means that the prior area has not been explored, and then the RRT boundary point will be selected for exploration after the room boundary point has been explored. In this way, the robot can explore one area and then turn to other areas to explore according to our artificial assumptions.

When the room boundary point exists, the room boundary point with the largest revenue value is selected as the target point; when there is no room boundary point, the RRT boundary point with the largest revenue value is selected as the target point;

The income value of the boundary point R1 _f = w ₁ *I _f -w ₂ *N _f ,

Among them, If is the information gain, the information gain refers to the number of unknown grids within the information gain radius _r =1 of the centroid point,

N _f is the path cost, and the path cost refers to the Euclidean distance between the robot's current position and the position of the center of mass;

w ₁ and w ₂ are custom weights, which are constants.

S52, guide the robot to navigate to the target point. Using the A* global path planning algorithm to quickly plan a path from the robot's current position to the target point in a known environment, and combined with the DWA local path planning algorithm, the robot can make good use of the local environment information to complete obstacle avoidance. The combination guides the robot to navigate to the target point and update the map.

Step 6. When no boundary point is detected in the priori region, it means that the region has been explored, and the model of the priori region is destroyed to form the next heuristic region.

Step 7: Repeat steps 1 to 6 until the robot explores the entire environment and obtains a grid map.

The invention discloses an indoor environment robot exploration system based on a priori information heuristic, including a data acquisition module, a positioning and mapping module, an RRT boundary point extraction module, a room boundary point extraction module and an environment exploration module.

The data collection module is used for the robot to collect the data of the surrounding environment information through the sensor carried by the robot.

The positioning and mapping module updates a part of the map as a known area based on the data of the surrounding environment information, and obtains the updated map.

The RRT boundary point extraction module uses two fast search random trees to perform boundary extraction on the updated map to obtain RRT boundary points.

The room boundary point extraction module is used to identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract the boundary point in the priori region, and obtain the room boundary point.

An environment exploration module, the environment exploration module is based on the RRT boundary point and the room boundary point, and the robot performs indoor environment exploration.

In order to fully prove the effectiveness of the present invention, the present invention and the autonomous exploration algorithm based on fast search random tree (referred to as "RRTs algorithm" below) are compared in two simulation scenarios. For each experimental setting, a total of 20 experiments were performed. Including 10 times using the RRTs algorithm and 10 times our improved algorithm. The metrics for experimental comparison are the time spent exploring the entire environment and the length of the path traveled.

Table 1 is the experimental data of the first scenario, Table 2 is the experimental data of the second scenario, and Table 3 is the experimental data of the third scenario. As shown in Figure 5, in the first scenario, compared with the RRTs algorithm, the method of the present invention reduces the exploration time by 34.9% and the length of the exploration path by 24.5%. As shown in Fig. 6, in the second scenario, the method of the present invention reduces the exploration time by 34.04% and the length of the exploration path by 35.9% compared with the RRTs algorithm. As shown in Figure 7, in the third scenario, the method of the present invention reduces the exploration time by 12.8% and the length of the exploration path by 16.9% compared with the RRTs algorithm.

Table 1

Table 2

table 3

By introducing a heuristic prior information exploration module, when the robot recognizes the heuristic object, it will first explore the environment in the prior area, so that the robot can first explore a room area and then turn to other areas, effectively reducing the amount of time in the exploration process. Retrospect the phenomenon and improve the efficiency of exploration.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

Obviously, the above-mentioned embodiments are only examples for clear description, and are not intended to limit the implementation manner. For those of ordinary skill in the art, other different forms of changes or modifications can also be made on the basis of the above description. There is no need and cannot be exhaustive of all implementations here. And the obvious changes or changes derived from this are still within the protection scope of the present invention.

Claims (3)

1. an indoor environment robot exploration method based on a priori information heuristic, is characterized in that, comprises the following steps: S1. The robot collects the data of the surrounding environment information through the sensor carried by itself; S2. Update a part of the map to a known area based on the data of the surrounding environment information, and obtain the updated map; S3. Use two fast search random trees to perform boundary extraction on the updated map to obtain RRT boundary points, including: S31, in the initialization phase, add the starting point to the tree structure as the root node, wherein the starting points of the two trees are artificially set in the free area of the map; S32, randomly sprinkle points in the map area as candidate points; S33. If the candidate point is in the known area, traverse all existing nodes on the tree structure, select the node closest to the candidate point as the nearest neighbor point, and use the connection line from the nearest neighbor point to the candidate node as the growth direction. If the distance between the point and the candidate node exceeds the preset step size, the nearest point grows one step size along the growth direction, and the reached point is used as the growth point. If the distance does not exceed the step size, the candidate point is used as the growth point; If the candidate point is in the unknown area, first find the nearest tree node of the candidate point, and the connection line from the nearest point to the candidate point is used as the growth direction, and the nearest point grows forward along the growth direction, and the place where it reaches the boundary is used as the growth direction. border point; S34. Perform collision detection on the map for the connection between the growth point and the candidate node, specifically including: Traverse all grid points on the connection between the growth point and the candidate node, and judge the grid state of the grid point; If the state of the grid point is occupied, the collision detection fails, and returns to S32 to collect points again; If the connection between the growing point and the candidate node does not encounter obstacles, add the connection between the candidate point, the growing point and the candidate node to the tree structure; S4. Identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract boundary points in the priori region, and obtain room boundary points; wherein, the heuristic object is a door; Among them, heuristic objects are identified and their position estimated, including: Construct a lightweight network, complete the recognition of heuristic objects based on the deep learning method, and obtain the coordinate information of the heuristic objects; wherein, the lightweight network includes a convolution layer, an inverse residual block, a pooling layer and an SSP layer ; Among them, a priori region is constructed based on the position of the heuristic object, including: When the robot recognizes the heuristic object, if the robot's position is below the heuristic object, the estimated room area is above the heuristic object. If the robot's position is above the heuristic object, the estimated room area below the position of the heuristic object; Wherein, the length of the estimated room area is the position of the heuristic object extending to both sides by a length a, and the width of the estimated area is the position of the heuristic object extending back the length 2b; wherein, the parameters a and b are set by experience Certainly; Among them, the boundary points are extracted in the prior area, and the room boundary points are obtained, including the following steps: Perform binarization processing on the image of the prior area to obtain a binarized image, in which the obstacles of the binarized image are white, and the rest of the area is black; Perform color flipping on the binarized image to obtain a color flipped image, wherein the obstacles in the color flipped image are black, and the rest of the area is white; The Canny operator is used to detect the edge of the binarized image. In the detection result, the edge of the image is set to white, and the rest of the area is black; Perform a bitwise AND operation on the binarized image and the color-flipped image to remove excess white edges to obtain a final image, where the boundary between the known area and the unknown area in the final image consists of straight lines; Extract the center of gravity of the straight line, which is the room boundary point; S5. Based on RRT boundary points and room boundary points, the robot performs indoor environment exploration, including: S51. When the room boundary point exists, select the room boundary point with the largest income value as the target point, and when there is no room boundary point, select the RRT boundary point with the largest income value as the target point, wherein the income value of the boundary point R1 _f = w ₁ *I _f -w ₂ *N _f , Among them, If is the information gain, the information gain refers to the number of unknown grids within the information gain radius _r =1 of the centroid point, N _f is the path cost, and the path cost refers to the Euclidean distance between the robot's current position and the position of the center of mass; w ₁ and w ₂ are custom weights, which are constants; S52, guide the robot to navigate to the target point. 2. The indoor environment robot exploration method based on a priori information heuristic according to claim 1, is characterized in that, after described S5 also comprises: S6. When no boundary point is detected in the a priori area, it means that the area has been explored, and the model of the prior area is destroyed to form the next heuristic area; S7, loop S1-S6 until the robot explores the entire environment and obtains a grid map. 3. An indoor environment robot exploration system based on a priori information heuristic is characterized in that, comprising: a data acquisition module, the data acquisition module is used for the robot to collect the data of the surrounding environment information through the sensor carried by itself; A positioning and mapping module, the positioning and mapping module updates a part of the map as a known area based on the data of the surrounding environment information, and obtains the updated map; The RRT boundary point extraction module, the RRT boundary point extraction module uses two fast search random trees to perform boundary extraction on the updated map to obtain RRT boundary points, including: S31. In the initialization phase, the starting point is added to the tree structure as the root node, wherein the starting points of the two trees are artificially set in the free area of the map; S32, randomly sprinkle points in the map area as candidate points; S33. If the candidate point is in the known area, traverse all existing nodes on the tree structure, select the node closest to the candidate point as the nearest neighbor point, and use the connection line from the nearest neighbor point to the candidate node as the growth direction. If the distance between the point and the candidate node exceeds the preset step size, the nearest point grows one step size along the growth direction, and the reached point is used as the growth point. If the distance does not exceed the step size, the candidate point is used as the growth point; If the candidate point is in the unknown area, first find the nearest tree node of the candidate point, and the connection line from the nearest point to the candidate point is used as the growth direction, and the nearest point grows forward along the growth direction, and the place where it reaches the boundary is used as the growth direction. border point; S34. Perform collision detection on the map for the connection between the growth point and the candidate node, specifically including: Traverse all grid points on the connection line between the growth point and the candidate node, and judge the grid state of the grid point; If the state of the grid point is occupied, the collision detection fails, and returns to S32 to collect points again; If the connection between the growing point and the candidate node does not encounter obstacles, add the connection between the candidate point, the growing point and the candidate node to the tree structure; The room boundary point extraction module is used to identify the heuristic object and perform position estimation on it, construct a priori region based on the position of the heuristic object, extract the boundary point in the priori region, and obtain the room Boundary point; wherein, identifying the heuristic object and estimating its position includes: constructing a lightweight network, completing the identification of the heuristic object based on a deep learning method, and obtaining the coordinate information of the heuristic object; wherein, the lightweight The level network includes convolution layer, inverse residual block, pooling layer and SSP layer; among them, the prior region is constructed based on the position of the heuristic object, including: when the robot recognizes the heuristic object, if the robot position is in the heuristic object below the heuristic object, the estimated room area is above the position of the heuristic object, and if the robot is positioned above the heuristic object, the estimated room area is below the position of the heuristic object; The length of the estimated room area is the position of the heuristic object extending to both sides by a length a, and the width of the estimated area is the position of the heuristic object extending back the length 2b, where the parameters a and b are set by experience; Extracting boundary points from the prior area to obtain room boundary points includes the following steps: performing binarization processing on the image in the prior area to obtain a binary image, wherein the obstacles in the binary image are white, and the rest of the area is black; color-flip the binarized image to obtain a color-flipped image, wherein the obstacles in the color-flipped image are black, and the rest of the area is white; the Canny operator is used to detect the edge of the binarized image , in the detection result, the edge of the image is set to white, and the rest of the area is black; the bitwise AND operation is performed on the binarized image and the color-flipped image to remove the excess white edge, and the final image is obtained. The boundary between the known area and the unknown area consists of a straight line; the center of gravity of the extracted line is the room boundary point; Environment exploration module, the environment exploration module is based on the RRT boundary point and the room boundary point, and the robot performs indoor environment exploration: when the room boundary point exists, the room boundary point with the largest profit value is selected as the target point, when there is no room boundary point. , then select the RRT boundary point with the largest profit value as the target point; guide the robot to navigate to the target point, where the profit value of the boundary point R1 _f = w ₁ *I _f -w ₂ *N _f , where _{If is} the information gain , the information gain refers to the number of unknown grids within the information gain radius r=1 of the centroid point, N _f is the path cost, and the path cost refers to the Euclidean distance between the robot's current position and the centroid point; w ₁ and w ₂ are the self- Define the weight as a constant.

Images