Neural network model-based forward stacker navigation system and method

Abstract

The invention belongs to the technical field of piling cars. The invention provides a reach stacker navigation system and method based on a neural network model. The method comprises the following steps: extracting first dynamic and static information of obstacles on the ground and second dynamic and static information of goods in a goods shelf area from a video image of a front target area shot by a camera; performing collision probability analysis on the second dynamic and static information by using a neural network model, performing matching according to the collision probability obtained by analysis to obtain a trajectory number, and performing planning according to the first dynamic and static information and the second dynamic and static information to obtain a plurality of advancing trajectories corresponding to the trajectory number in the front target area; and when the forward moving type piling car is about to arrive at the front target area, one of the advancing tracks is screened out to serve as a target advancing track, and the forward moving type piling car is controlled to run according to the target advancing track. The track of the forward-moving type piling car is planned based on the ground obstacles and the goods on the goods shelf at the same time, and higher navigation safety is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of stacker trucks, and in particular to a navigation system and method for a reach stacker truck based on a neural network model. Background Art

[0002] In the field of modern logistics and warehousing, reach stackers, as an important material handling equipment, are widely used in warehouse cargo handling, stacking and other operations.

[0003] Earlier, the navigation methods of reach stackers mainly relied on traditional track navigation, magnetic stripe navigation, etc. Track navigation requires laying special tracks on the ground, and the stacker travels along the tracks. Although this method has high stability, the cost of laying tracks is high. In addition, when the warehouse layout is adjusted later, the track changes are difficult and costly, and lack flexibility. Magnetic stripe navigation is to stick magnetic strips on the ground, and the stacker uses sensors to identify the magnetic strips for navigation. However, the magnetic strips are easily affected by the environment. For example, dust and moisture on the ground will interfere with the sensor's recognition of the magnetic strips, resulting in reduced navigation accuracy. In complex warehouse environments, the route planning of magnetic stripe navigation is relatively simple and difficult to meet diverse operational needs. With the increasing complexity of the warehousing environment and the continuous improvement of the requirements for logistics operation efficiency, these existing navigation methods have gradually become unable to meet the needs of reach stackers in terms of precise positioning, efficient path planning, and flexible adaptation to different operating scenarios.

[0004] At present, navigation methods based on image recognition technology have been applied to reach stackers. That is, a camera is placed at a suitable position directly in front of the reach stacker, and the distribution of objects in front of the reach stacker to be driven is identified based on the video images taken by the camera, so as to plan a forward path for the reach stacker. The reach stacker can move along the forward path without collision.

[0005] The above-mentioned navigation methods based on video images mainly focus on the distribution of objects on the ground, that is, the planned forward path can only ensure that the reach stacker does not collide with objects on the ground. However, in the storage area, some goods on the shelves may protrude from the shelves due to irregular placement. These goods may collide with the lift frame of the reach stacker with a certain height. The existing technology lacks consideration for the goods on the shelves, so it is necessary to improve the video navigation method of the reach stacker. Summary of the invention

[0006] In this regard, the present invention provides a reach stacker navigation method, system, computer storage medium and computer program product based on a neural network model to solve the above technical problems.

[0007] The present invention discloses a navigation method for a reach stacker based on a neural network model. The method comprises the following steps: receiving a video image of a front target area shot by a camera disposed in front of the reach stacker, extracting first dynamic and static information of obstacles located in a ground area from the video image, and extracting second dynamic and static information of goods located in a shelf area; performing collision probability analysis on the second dynamic and static information using a neural network model, obtaining a number of trajectories according to the collision probability matching obtained by the analysis, and planning a plurality of forward trajectories corresponding to the number of trajectories located in the front target area according to the first dynamic and static information and the second dynamic and static information; when the reach stacker is about to arrive at the front target area, selecting one of the forward trajectories as a target forward trajectory, and controlling the reach stacker to travel according to the target forward trajectory.

[0008] Preferably, the use of a neural network model to perform a collision probability analysis on the second dynamic and static information, and obtaining a number of trajectories according to the collision probability obtained by the analysis, includes: judging whether there are goods protruding from the edge of the shelf on the shelf in the target area ahead according to the second dynamic and static information; if not, setting the collision probability to be lower than the probability threshold, and the number of trajectories obtained by matching is the first number of trajectories; if so, setting the collision probability to be higher than the probability threshold, and then further: judging the protruding size and motion information of the target goods protruding from the edge of the shelf according to the second dynamic and static information; if the motion information indicates that the target goods are in a stationary state, calculating the second number of trajectories according to the first protruding size of the target goods and the driving width of the target area ahead; if the motion information indicates that the target goods are in a shaking state, using a neural network model to predict and analyze the motion information, obtain the second protruding size of the target goods, and calculate the third number of trajectories according to the second protruding size of the target goods and the driving width of the target area ahead; wherein, the first number of trajectories is lower than the second number of trajectories and the third number of trajectories.

[0009] Preferably, when the collision probability is higher than the probability threshold, it also includes: determining the protrusion size and distribution position of each of the target goods, calculating the median value of each of the protrusion sizes, and using it as the target protrusion size; and deriving the distribution concentration of the target goods in the front target area according to each of the distribution position evaluations; deriving the corresponding cargo inclination coefficient according to the target protrusion size and the distribution concentration, and adjusting the second trajectory number to the fourth trajectory number according to the cargo inclination coefficient, or adjusting the third trajectory number to the fifth trajectory number.

[0010] Furthermore, according to the above cargo tilt coefficient, the second track number obtained above can be adjusted to a slightly larger fourth track number, or the third track number can be adjusted to a slightly larger fifth track number. Of course, when the distribution concentration of each target cargo is low, it means that the placement of these target cargoes is not standardized and is an individual case, not the above-mentioned overall placement is not standardized, and there will be no overall tilt. Therefore, the cargo tilt coefficient can be set to 1, that is, the second track number and the third track number obtained above are not adjusted.

[0011] Preferably, the distribution concentration degree is obtained by an evaluation model based on a classification algorithm.

[0012] Preferably, when the reach stacker is about to arrive at the front target area, one is selected from each of the forward trajectories as the target forward trajectory, including: in the process of the reach stacker approaching the front target area, the third dynamic and static information of the goods located in the shelf area is obtained multiple times, and the degree of deviation between the latest third dynamic and static information and the second dynamic and static information is evaluated; when the degree of deviation is greater than a deviation threshold, the reach stacker is set to be at a first distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; when the degree of deviation is less than the deviation threshold, the reach stacker is set to be at a second distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; wherein the first distance is less than the second distance.

[0013] The present invention also discloses a navigation system for a reach stacker based on a neural network model, the system comprising a navigation controller and a camera deployed in front of the reach stacker, the navigation controller being used to implement the following steps: receiving a video image of a front target area captured by a camera, extracting first dynamic and static information of obstacles located in a ground area from the video image, and extracting second dynamic and static information of goods located in a shelf area; performing collision probability analysis on the second dynamic and static information using a neural network model, obtaining a number of trajectories according to the collision probability matching obtained by the analysis, and planning a plurality of forward trajectories corresponding to the number of trajectories located in the front target area according to the first dynamic and static information and the second dynamic and static information; when the reach stacker is about to arrive at the front target area, selecting one of the forward trajectories as a target forward trajectory, and controlling the reach stacker to travel according to the target forward trajectory.

[0014] Preferably, the use of a neural network model to perform a collision probability analysis on the second dynamic and static information, and obtaining a number of trajectories according to the collision probability obtained by the analysis, includes: judging whether there are goods protruding from the edge of the shelf on the shelf in the target area ahead according to the second dynamic and static information; if not, setting the collision probability to be lower than the probability threshold, and the number of trajectories obtained by matching is the first number of trajectories; if so, setting the collision probability to be higher than the probability threshold, and then further: judging the protruding size and motion information of the target goods protruding from the edge of the shelf according to the second dynamic and static information; if the motion information indicates that the target goods are in a stationary state, calculating the second number of trajectories according to the first protruding size of the target goods and the driving width of the target area ahead; if the motion information indicates that the target goods are in a shaking state, using a neural network model to predict and analyze the motion information, obtain the second protruding size of the target goods, and calculate the third number of trajectories according to the second protruding size of the target goods and the driving width of the target area ahead; wherein, the first number of trajectories is lower than the second number of trajectories and the third number of trajectories.

[0015] Preferably, when the collision probability is higher than the probability threshold, it also includes: determining the protrusion size and distribution position of each of the target goods, calculating the median value of each of the protrusion sizes, and using it as the target protrusion size; and deriving the distribution concentration of the target goods in the front target area according to each of the distribution position evaluations; deriving the corresponding cargo inclination coefficient according to the target protrusion size and the distribution concentration, and adjusting the second trajectory number to the fourth trajectory number according to the cargo inclination coefficient, or adjusting the third trajectory number to the fifth trajectory number.

[0016] Furthermore, according to the above cargo tilt coefficient, the second track number obtained above can be adjusted to a slightly larger fourth track number, or the third track number can be adjusted to a slightly larger fifth track number. Of course, when the distribution concentration of each target cargo is low, it means that the placement of these target cargoes is not standardized and is an individual case, not the above-mentioned overall placement is not standardized, and there will be no overall tilt. Therefore, the cargo tilt coefficient can be set to 1, that is, the second track number and the third track number obtained above are not adjusted.

[0017] Preferably, the distribution concentration degree is obtained by an evaluation model based on a classification algorithm.

[0018] Preferably, when the reach stacker is about to arrive at the front target area, one is selected from each of the forward trajectories as the target forward trajectory, including: in the process of the reach stacker approaching the front target area, the third dynamic and static information of the goods located in the shelf area is obtained multiple times, and the degree of deviation between the latest third dynamic and static information and the second dynamic and static information is evaluated; when the degree of deviation is greater than a deviation threshold, the reach stacker is set to be at a first distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; when the degree of deviation is less than the deviation threshold, the reach stacker is set to be at a second distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; wherein the first distance is less than the second distance.

[0019] The present invention also discloses an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the methods described above when executed by the processor.

[0020] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements any of the above methods when executed by a processor.

[0021] The present invention also discloses a computer program product. The computer program product contains computer program codes. When the computer program codes are executed by a processor, the method described in any of the above items is implemented.

[0022] The beneficial effect of the present invention is that the scheme of the present invention simultaneously considers the influence of ground obstacles and irregularly placed goods on the shelves on the movement of the reach stacker, so as to plan a more reasonable forward trajectory, thereby improving the driving safety of the reach stacker in the storage area. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 It is a flow chart of a navigation method for a reach stacker based on a neural network model disclosed in an embodiment of the present invention.

[0025] Figure 2It is a structural schematic diagram of a reach stacker navigation system based on a neural network model disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0027] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0028] The reach stacker in the present invention has a self-moving function, that is, it can realize self-planning and walking execution of the path in the storage area at least with the cooperation of the camera and the navigation controller. Since it belongs to the existing technology, other functional components will not be described in detail.

[0029] like Figure 1 As shown, an embodiment of the present invention discloses a reach stacker navigation method based on a neural network model, the method comprising the following steps: receiving a video image of a front target area taken by a camera deployed in front of the reach stacker, extracting first dynamic and static information of obstacles located in the ground area from the video image, and extracting second dynamic and static information of goods located in the shelf area.

[0030] In this step, the present invention arranges a camera in front of the reach stacker (for example, on top of the lifting frame), and its function is to capture in real time the video image of the target area in front that the reach stacker needs to pass. From the captured video image, the first dynamic and static information of the obstacles located in the ground area is extracted through specific image recognition algorithms and techniques. The ground obstacles here may include temporary sundries, goods, other reach stackers, etc. placed on the ground. The first dynamic and static information includes whether the ground obstacles are stationary or in motion, as well as their position, shape, size and other characteristics, which is helpful for subsequent planning of a safe driving path.

[0031] At the same time, the second dynamic and static information of the goods located in the shelf area is extracted from the video image. The second dynamic and static information includes whether there is any goods protruding from the shelf, the size of the goods protruding from the shelf, whether the goods on the shelf have dynamic signs such as shaking, etc. As described in the background technology, if the goods on the shelf protrude from the shelf, it is possible to collide with the forward stacker, so obtaining this part of information is of great significance to ensure the safety of the stacker.

[0032] A neural network model is used to perform a collision probability analysis on the second dynamic and static information, and a number of trajectories is obtained according to the collision probability matching obtained by the analysis. A plurality of forward trajectories corresponding to the number of trajectories located in the front target area are planned according to the first dynamic and static information and the second dynamic and static information.

[0033] In this step, the present invention pre-builds a neural network model, which has been trained with a large amount of data and has strong pattern recognition and data analysis capabilities. It can accurately calculate the probability of a reach stacker colliding with the goods on the shelf under different driving conditions based on the characteristic information of the goods. For example, when the neural network model detects that the goods on a certain shelf are obviously protruding, it will calculate the probability of the reach stacker colliding with it based on the size and angle of the protrusion and the size of the stacker itself.

[0034] When the reach stacker has not yet reached the target area ahead, multiple forward trajectories are planned based on the above-mentioned first dynamic and static information of the ground obstacles, the second dynamic and static information of the shelf goods, and the path planning algorithm (such as the Dijkstra algorithm, the rapidly exploring random tree (RRT) algorithm, the A* algorithm, etc.) for further screening by the reach stacker.

[0035] At the same time, the present invention also uses a pre-built neural network model to perform a collision probability analysis on the second dynamic and static information of the goods on the shelf, and determines the number of the above-mentioned forward trajectories based on the analysis results of the collision probability. When the collision probability between the forward stacker and the goods on the shelf in the front target area is higher, in order to ensure that the stacker can drive safely, it is necessary to plan more alternative trajectories, such as 7-10, to increase the possibility of finding a safe path later; conversely, if the collision probability is lower, the number of trajectories can be appropriately reduced, such as 3-5, to improve the efficiency of path planning.

[0036] It should be noted that when planning, the forward trajectory obtained by planning must avoid ground obstacles and try to meet the driving requirements of the stacker, such as driving stability and reasonable turning radius.

[0037] When the reach stacker is about to reach the front target area, one of the forward trajectories is selected as a target forward trajectory, and the reach stacker is controlled to travel according to the target forward trajectory.

[0038] In this step, when the reach stacker is about to reach the target area ahead, the reach stacker needs to select an optimal trajectory from the multiple forward trajectories obtained from the aforementioned plan according to the actual situation, and connect the selected target forward trajectory with the currently executed trajectory, so as to smoothly enter the target area ahead.

[0039] The solution of the present invention simultaneously considers the impact of ground obstacles and irregularly placed goods on the shelves on the movement of the reach stacker, so as to plan a more reasonable forward trajectory, thereby improving the driving safety of the reach stacker in the storage area.

[0040] Preferably, the use of a neural network model to perform a collision probability analysis on the second dynamic and static information, and obtaining a number of trajectories according to the collision probability obtained by the analysis, includes: judging whether there are goods protruding from the edge of the shelf on the shelf in the target area ahead according to the second dynamic and static information; if not, setting the collision probability to be lower than the probability threshold, and the number of trajectories obtained by matching is the first number of trajectories; if so, setting the collision probability to be higher than the probability threshold, and then further: judging the protruding size and motion information of the target goods protruding from the edge of the shelf according to the second dynamic and static information; if the motion information indicates that the target goods are in a stationary state, calculating the second number of trajectories according to the first protruding size of the target goods and the driving width of the target area ahead; if the motion information indicates that the target goods are in a shaking state, using a neural network model to predict and analyze the motion information, obtain the second protruding size of the target goods, and calculate the third number of trajectories according to the second protruding size of the target goods and the driving width of the target area ahead; wherein, the first number of trajectories is lower than the second number of trajectories and the third number of trajectories.

[0041] In an embodiment of the present invention, the present invention is configured to decide different strategies for determining the number of trajectories based on different types represented by the second dynamic and static information, specifically as follows: the second dynamic and static information includes whether there are target goods protruding from the shelf edge, the size of these target goods protruding from the shelf edge, and also includes the motion information of these target goods, that is, whether they are in a stationary state or in a moving state. The above information can be used to analyze the collision risk of the stacker in the target area ahead, and then determine the number of alternative forward trajectories for subsequent planning.

[0042] 1) The goods do not protrude from the shelf, that is, the target goods do not exist: When the judgment result is that there is no goods protruding from the edge of the shelf, since the possibility of a collision between the reach stacker and the goods on the shelf is extremely low in this case, the collision probability can be set lower than the probability threshold. The probability threshold here is a value set based on actual experience and safety standards, which is used to define the level of collision risk. In this low-risk situation, in order to simplify the path planning process and improve planning efficiency, the number of matched trajectories is set to the minimum number of first trajectories. In this case, the planned trajectory is only a relatively direct and efficient path planned based on factors such as ground obstacle information and the destination of the reach stacker. The number of first trajectories is, for example, 1 or 2.

[0043] 2) Goods protrude from the shelf, that is, there are target goods: If it is determined that there are target goods protruding from the edge of the shelf, the collision probability is set to be higher than the probability threshold. Specifically, the second dynamic and static information also includes the motion state of the target goods. If the motion state indicates that the target goods are in a stationary state, the size of the target goods protruding from the edge of the shelf, that is, the first protrusion size, is known and fixed. The second track number is directly calculated based on the first protrusion size of the target goods and the driving width of the target area in front. For example, , where is a constant greater than 1.

[0044] Among them, when the first protruding size of the target cargo is larger, the present invention sets and plans more alternative trajectories, so that when, for example, when arriving at the area where the target cargo is located and finding that its actual protruding size is larger than the first protruding size, other alternative trajectories can be quickly switched without temporary planning; and when the driving width of the target area ahead is larger, since the driving space is sufficient, fewer alternative trajectories are correspondingly set and planned (the spacing between each alternative trajectory is larger), so that the reach stacker can be controlled to move on a trajectory that is farther away from the target cargo.

[0045] In addition, the motion state may also indicate that the target cargo is in a shaking state. For example, the forward stacker on the back of the shelf is sorting the cargo, and the target cargo may be affected by the sorting and shake. The final protrusion size of the target cargo in a shaking state cannot be determined at the moment, so the present invention further uses the aforementioned pre-built neural network model to conduct an in-depth analysis of the motion information of the target cargo in a shaking state, that is, based on the shaking law, starting position, shaking speed and direction of the target cargo and other motion information, predict the possible second protrusion size of the target cargo in the future. Then, the aforementioned method is used to calculate the number of third trajectories corresponding to the second protrusion size: , where is a constant greater than 1.

[0046] Preferably, when the collision probability is higher than the probability threshold, it also includes: determining the protrusion size and distribution position of each of the target goods, calculating the median value of each of the protrusion sizes, and using it as the target protrusion size; and deriving the distribution concentration of the target goods in the front target area according to each of the distribution position evaluations; deriving the corresponding cargo inclination coefficient according to the target protrusion size and the distribution concentration, and adjusting the second trajectory number to the fourth trajectory number according to the cargo inclination coefficient, or adjusting the third trajectory number to the fifth trajectory number.

[0047] In the embodiment of the present invention, the aforementioned embodiment only considers the protruding size of each target cargo itself to determine the planned number of alternative trajectories. However, in fact, when the protruding size of the target cargo protruding from the edge of the shelf is generally large and these target cargoes are concentratedly distributed, it means that these target cargoes are not placed in a standardized manner as a whole. In this case, these target cargoes are more likely to tilt as a whole, that is, gradually tilt outward under the action of gravity. When the reach stacker has not reached the target driving area in front, the overall tilt degree of these target cargoes is not high. At this time, the number of the second trajectory and the number of the third trajectory mentioned above can be determined based on the median value of the protruding size of these target cargoes; when the reach stacker reaches the target driving area in front, the vibration of the reach stacker will cause these target cargoes to tilt faster, resulting in the latest median value of the protruding size becoming much larger than before, and thus there is a greater probability of collision with the reach stacker.

[0048] Therefore, in view of the above situation, the present invention is configured to evaluate the distribution concentration of each target cargo according to its distribution position, and then obtain the corresponding cargo tilt coefficient according to the above target protruding size and the size of the distribution concentration obtained by the evaluation. The cargo tilt coefficient represents the difference between the actual protruding size of these target cargoes when the reach stacker reaches the front target area (specifically the distribution area of ​​these target cargoes) and the initial protruding size of these target cargoes when the reach stacker has not reached the front target area. In other words, the larger the cargo tilt coefficient (for example, 1.2), the larger the above difference; the smaller the cargo tilt coefficient (for example, 1.0, 1.1), the smaller the above difference.

[0049] It should be noted that the corresponding relationship between the cargo tilt coefficient, the distribution concentration and the target protruding size can also be determined by using a comparison table, and the corresponding relationship is obviously positively correlated, so the details will not be repeated. The comparison table is as follows:

[0050] Furthermore, according to the cargo tilt coefficient, the second track number obtained above can be adjusted to a slightly larger fourth track number, or the third track number can be adjusted to a slightly larger fifth track number. Of course, when the distribution concentration of each target cargo is low, it means that the irregular placement of these target cargoes is an individual case and does not belong to the above-mentioned irregular overall placement, and there will be no overall tilt. Therefore, the cargo tilt coefficient can be set to 1, that is, the second track number and the third track number obtained above are not adjusted. The adjustment here is, for example, a multiplication operation.

[0051] Preferably, the distribution concentration degree is obtained by an evaluation model based on a classification algorithm.

[0052] In the embodiment of the present invention, a model dedicated to evaluating the above distribution concentration is pre-built, and such a model can be built and trained based on algorithms such as SVM (support vector machine), decision tree, random forest, etc. The specific process will not be repeated.

[0053] Preferably, when the reach stacker is about to arrive at the front target area, one is selected from each of the forward trajectories as the target forward trajectory, including: in the process of the reach stacker approaching the front target area, the third dynamic and static information of the goods located in the shelf area is obtained multiple times, and the degree of deviation between the latest third dynamic and static information and the second dynamic and static information is evaluated; when the degree of deviation is greater than a deviation threshold, the reach stacker is set to be at a first distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; when the degree of deviation is less than the deviation threshold, the reach stacker is set to be at a second distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; wherein the first distance is less than the second distance.

[0054] In the embodiment of the present invention, since the multiple forward trajectories obtained by the aforementioned planning are planned when the reach stacker is a certain distance (for example, 5m) away from the target area in front, the dynamic and static information of the goods located in the shelf area may have undergone significant changes during its driving within the certain distance, it is necessary to optimize the screening timing of the target forward trajectory.

[0055] Specifically, the present invention is arranged to obtain the third dynamic and static information of the goods located in the shelf area multiple times during the process of the forward-moving stacker approaching the edge of the front target area, and evaluate the degree of deviation from the initial second dynamic and static information. The evaluation of the degree of deviation is at least based on the above-mentioned protruding size and motion information. For example, when the protruding size of the target goods at the previous and next moments becomes larger and / or the motion information becomes stronger (speed increases, shaking amplitude increases, etc.), the corresponding degree of deviation is greater; conversely, when the protruding size of the target goods at the previous and next moments becomes smaller and / or the motion information becomes weaker (speed slows down, shaking amplitude decreases, etc.), the corresponding degree of deviation is also greater. In the reverse case of the above situation, the corresponding degree of deviation is smaller.

[0056] When the above-mentioned deviation degree is greater than the deviation threshold, the present invention is arranged to select one from each forward trajectory as the target forward trajectory when the reach stacker is closer to the target area ahead, so that the passing safety of the selected target forward trajectory can be improved; and when the above-mentioned deviation degree is less than the deviation threshold, the present invention is arranged to select one from each forward trajectory as the target forward trajectory when the reach stacker is farther from the target area ahead, so that the target forward trajectory can be determined as early as possible, thereby reserving more time for the connection between the target forward trajectory and the current driving trajectory, and reducing the probability of the reach stacker stopping and waiting (calculating the connection trajectory).

[0057] like Figure 2 As shown, an embodiment of the present invention further discloses a reach stacker navigation system based on a neural network model, the system comprising a navigation controller and a camera deployed in front of the reach stacker, the navigation controller being used to implement the following steps: receiving a video image of a front target area captured by a camera, extracting first dynamic and static information of obstacles located in a ground area from the video image, and extracting second dynamic and static information of goods located in a shelf area; performing collision probability analysis on the second dynamic and static information using a neural network model, obtaining a number of trajectories according to the collision probability matching obtained by the analysis, and planning a plurality of forward trajectories corresponding to the number of trajectories located in the front target area according to the first dynamic and static information and the second dynamic and static information; when the reach stacker is about to arrive at the front target area, selecting one of the forward trajectories as a target forward trajectory, and controlling the reach stacker to travel according to the target forward trajectory.

[0058] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the methods described above when executed by the processor.

[0059] An embodiment of the present invention further discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any of the preceding items is implemented.

[0060] The embodiment of the present invention further discloses a computer program product. The computer program product contains computer program code. When the computer program code is executed by a processor, the method described in any of the above items is implemented.

[0061] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A navigation method for a reach stacker based on a neural network model, characterized in that: The method comprises the following steps: receiving a video image of a front target area captured by a camera disposed in front of a reach stacker, extracting first dynamic and static information of obstacles located in a ground area from the video image, and extracting second dynamic and static information of goods located in a shelf area; Using a neural network model to perform a collision probability analysis on the second dynamic and static information, obtaining a number of trajectories according to the collision probability matching obtained by the analysis, and planning and obtaining a plurality of forward trajectories corresponding to the number of trajectories located in the front target area according to the first dynamic and static information and the second dynamic and static information; When the reach stacker is about to reach the front target area, one of the forward trajectories is selected as a target forward trajectory, and the reach stacker is controlled to travel according to the target forward trajectory.

2. The method for navigating a reach stacker based on a neural network model according to claim 1, characterized in that: A neural network model is used to perform collision probability analysis on the second dynamic and static information, and the number of trajectories is obtained according to the collision probability obtained by the analysis, including: judging whether there are goods protruding from the shelf edge on the shelf in the target area ahead according to the second dynamic and static information; if not, setting the collision probability to be lower than the probability threshold, and the number of trajectories obtained by matching is the first number of trajectories; if yes, setting the collision probability to be higher than the probability threshold, and then further: judging the protruding size and motion information of the target goods protruding from the shelf edge according to the second dynamic and static information; if the motion information indicates that the target goods are in a stationary state, the second number of trajectories is calculated according to the first protruding size of the target goods and the driving width of the target area ahead; if the motion information indicates that the target goods are in a shaking state, the neural network model is used to predict and analyze the motion information to obtain the second protruding size of the target goods, and the third number of trajectories is calculated according to the second protruding size of the target goods and the driving width of the target area ahead; wherein the first number of trajectories is lower than the second number of trajectories and the third number of trajectories.

3. The method for navigating a reach stacker based on a neural network model according to claim 2, characterized in that: When the collision probability is higher than the probability threshold, it also includes: determining the protruding size and distribution position of each of the target goods, calculating the median value of each of the protruding sizes, and using it as the target protruding size; and obtaining the distribution concentration of the target goods in the front target area according to each of the distribution positions; obtaining the corresponding cargo inclination coefficient according to the target protruding size and the distribution concentration, and adjusting the second trajectory number to the fourth trajectory number according to the cargo inclination coefficient, or adjusting the third trajectory number to the fifth trajectory number.

4. The method for navigating a reach stacker based on a neural network model according to claim 3, characterized in that: The distribution concentration degree is obtained by an evaluation model based on a classification algorithm.

5. The method for navigating a reach stacker based on a neural network model according to claim 1, characterized in that: When the reach stacker is about to arrive at the front target area, one is selected from each of the forward trajectories as the target forward trajectory, including: in the process of the reach stacker approaching the front target area, the third dynamic and static information of the goods located in the shelf area is obtained multiple times, and the degree of deviation between the latest third dynamic and static information and the second dynamic and static information is evaluated; when the degree of deviation is greater than a deviation threshold, the reach stacker is set to be at a first distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; when the degree of deviation is less than the deviation threshold, the reach stacker is set to be at a second distance from the front target area, and one is selected from each of the forward trajectories as the target forward trajectory; wherein the first distance is less than the second distance.

6. A forward-reaching stacker navigation system based on a neural network model, characterized in that: The system includes a navigation controller and a camera deployed in front of a reach stacker, wherein the navigation controller is used to implement the following steps: receiving a video image of a target area in front captured by the camera, extracting first dynamic and static information of obstacles located in a ground area from the video image, and extracting second dynamic and static information of goods located in a shelf area; Using a neural network model to perform a collision probability analysis on the second dynamic and static information, obtaining a number of trajectories according to the collision probability matching obtained by the analysis, and planning and obtaining a plurality of forward trajectories corresponding to the number of trajectories located in the front target area according to the first dynamic and static information and the second dynamic and static information; When the reach stacker is about to reach the front target area, one of the forward trajectories is selected as a target forward trajectory, and the reach stacker is controlled to travel according to the target forward trajectory.

7. The reach stacker navigation system based on a neural network model according to claim 6, characterized in that: The method of using a neural network model to perform a collision probability analysis on the second dynamic and static information, and obtaining a number of trajectories according to the collision probability obtained by the analysis, includes: judging whether there are goods protruding from the shelf edge on the shelf in the target area ahead according to the second dynamic and static information; if not, setting the collision probability to be lower than the probability threshold, and the number of trajectories obtained by matching is the first number of trajectories; if yes, setting the collision probability to be higher than the probability threshold, and then further: judging the protruding size and motion information of the target goods protruding from the shelf edge according to the second dynamic and static information; if the motion information indicates that the target goods are in a stationary state, calculating the second number of trajectories according to the first protruding size of the target goods and the driving width of the target area ahead; if the motion information indicates that the target goods are in a shaking state, using a neural network model to predict and analyze the motion information, obtaining the second protruding size of the target goods, and calculating the third number of trajectories according to the second protruding size of the target goods and the driving width of the target area ahead; wherein the first number of trajectories is lower than the second number of trajectories and the third number of trajectories.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method according to any one of claims 1 to 5 when executed by the processor.

9. A computer storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

10. A computer program product, characterized in that: The computer program product is prepackaged with computer program codes, and the computer program codes implement the method according to any one of claims 1 to 5 when executed by a processor.

Images