An unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling

Abstract

The present application discloses an unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling, which includes pre-scanning a target to construct an unstructured terrain point cloud; constructing a digital elevation model and a digital gradient model based on the unstructured terrain point cloud model to divide the terrain obstacle area in the unstructured terrain; using a static kinematic model to calculate the critical obstacle height of the passable terrain and realizing the semantic segmentation of risk obstacles in the terrain obstacle area; combining the dynamic roll motion process of an articulated engineering vehicle to design a steady-state margin angle index to realize the judgment of the dynamic stable state of the articulated engineering vehicle; constructing a vehicle-terrain coupling solution method to semantically segment the unstructured terrain into a safe area, a primary danger area, and a secondary danger area. The beneficial effects of the present application are as follows: It expands the semantic information dimension of the unstructured terrain, effectively provides multi-level danger warning information for the driving operation of work vehicles, and guides the vehicles to operate safely and efficiently.

Description

Technical Field

[0001] The present application relates to the technical field of unstructured terrain point cloud mapping, and particularly to an unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling. Background Art

[0002] Unmanned engineering vehicles exhibit great application potential in unstructured terrain environments such as large-scale construction, tunnel environment operations, and disaster relief. High-precision maps are a basic function of unmanned driving systems. They not only rely on high-resolution terrain data but also require real-time and accurate semantic information to ensure that unmanned vehicles can safely and effectively carry out operation planning.

[0003] In unstructured terrain environments, the driving speed of engineering vehicles during operation tasks is relatively low, mainly because the irregularity of the terrain may cause the vehicle to roll over due to instability. Most engineering vehicles have a relatively high center of mass, resulting in limited driver vision. When the assessment of terrain passability is inaccurate, the risk of vehicle rollover increases. By establishing a vehicle-terrain coupling model, it is possible to more accurately predict the driving attitude information of the vehicle and obtain the risk warning state of the engineering vehicle. However, there is a large time delay in the real-time calculation of this coupling model, and it cannot provide sufficient vehicle control decision-making and feedback response time. Therefore, by dividing the point cloud data of unstructured terrain into safety warning states and integrating these warning states into the high-precision map as semantic information, it is possible to directly provide regional driving safety information and control response feedback for the vehicle in advance, and optimize the operation performance of unmanned engineering vehicles on the premise of ensuring safety. Summary of the Invention

[0004] One of the purposes of the present application is to provide an unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling that can solve at least one of the defects in the above background art.

[0005] To achieve at least one of the above purposes, the technical solution adopted in the present application is: an unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling, which pre-scans the target to construct an unstructured terrain point cloud, and the safety semantic segmentation method implemented by the unstructured terrain point cloud model includes the following steps.

[0006] Step 1: Construct a digital model.

[0007] Based on the unstructured terrain point cloud model, a digital elevation model and a digital gradient model are constructed, and the terrain obstacle area in the unstructured terrain is divided through the digital gradient model.

[0008] Step 2: Semantic segmentation of risk obstacles.

[0009] Using the static kinematic model of an articulated engineering vehicle, calculate the critical obstacle height of the passable terrain. Based on this, semantic segmentation of risk obstacles is achieved in the terrain obstacle area.

[0010] Step 3: Judge the dynamic stability state of the articulated engineering vehicle.

[0011] Combined with the dynamic roll motion process of the articulated engineering vehicle, design a simple and easy-to-use steady-state margin angle index. Based on this, judge the dynamic stability state of the articulated engineering vehicle.

[0012] Step 4: Unstructured terrain safety semantic segmentation.

[0013] Construct a vehicle-terrain coupling solution method, and according to the calculated steady-state margin angle index, segment the unstructured terrain safety semantics into three levels: safe area, primary danger area, and secondary danger area.

[0014] Preferably, when the indexing object in the unstructured terrain point cloud model in Step 1 is the elevation value, the digital elevation model can be used, and the specific calculation is as follows:

[0015]

[0016] Among them, z ( x , y ) is the indexed elevation value; N is the total number of neighborhood point clouds in the unstructured terrain point cloud model; d is the rasterized side length, that is, the model resolution; point cloud ( x i , y i , z i ) is the neighborhood point near the point cloud ( x , y , z ).

[0017] The discrete gradient of the digital elevation model is expressed as:

[0018] \forall f=\left [ {{^{{g}_{x}}_{{g}_{y}}}} \right ]=\left [ {\frac {\partial f} {\partial x}\, \frac {\partial f} {\partial y}} \right ]=\left [ {f(x+1,y)-f(x,y)\, f(x,y+1)-f(x,y)} \right ]{\,}^{T}

[0019] Among them, g x and g y are respectively the gradient components of the point ([[]]END]] x , y ) in the x and y directions.

[0020] The gradient components of the digital gradient model are calculated as:

[0021]

[0022]

[0023] Among them, l is the side length of the operator grid.

[0024] By calculating the gradient of the digital elevation model, the terrain obstacle area in the unstructured terrain can be divided.

[0025] Preferably, in step 2, select the steering hinge point as the vehicle coordinate origin to construct the kinematic model of the articulated engineering vehicle; the total vehicle mass mainly includes the mass of the front body G 1 , the mass of the front wheels G 2 , the mass of the front swing arms G 3 , the mass of the rear body G 4 and the mass of the rear wheels G5; the center of gravity G(x0, y0, z0) of the vehicle is:

[0026]

[0027] Among them, M n is the mass of each vehicle body part; G n is the coordinate of each vehicle body part.

[0028] The tire contact points of each tire are obtained through geometric motion relationships. The left front wheel P 1l (x 1l , y 1l , z 1l ) , the right front wheel P 1r (x 1r , y 1r , z 1r) and the left rear wheel P 2l (x 2l , y 2l , z 2l ) and the right rear wheel P 2r (x 2r , y 2r , z 2r ) have the following ground contact coordinates:

[0029]

[0030]

[0031] {R}_{1}=\left [ {{^{^{\cos {\theta}}_{\sin {\theta}}}_{{\,}^{0}_{\,}}{^{{\,}^{-\sin {\theta}}_{\cos {\theta}}}_{{\,}^{0}_{\,}}{^{^{0}_{0}\,}_{^{1}_{\,}\,}\,}}}} \right ] and {R}_{2}=\left [ {{^{^{\cos {\alpha}}_{\, \, \, 0}}_{{\,}^{-\sin {\alpha}}_{\,}}{{}^{^{0}_{1}\,}_{{\,}^{0}_{\,}}{^{^{\sin {\alpha}}_{0}\,}_{^{\cos {\alpha}}_{\,}\,}\,}}}} \right ]

[0032] where W is half of the wheelbase width; L 1 is the distance from the front axle to the center of gravity; L 2 is the distance from the rear axle to the center of gravity; H is the height of the steering articulation point; θ is the articulated steering angle; α is the swing angle of the rear axle.

[0033] During the static rollover motion of an articulated engineering vehicle, the roll angle generated at low speed φ is defined as:

[0034]

[0035] where, h is the height of the obstacle; D is the projected distance from the tire contact point P 1l to the roll axis.

[0036] After that, taking a safety factor δ , the critical obstacle height h 0 can be derived as:

[0037]

[0038] where, φ max is the limit roll angle.

[0039] Preferably, by calculating the limit roll angle of the articulated engineering vehicle, the critical obstacle height of the passable terrain is obtained; the critical obstacle height of the passable terrain is compared with the actual obstacle elevation value obtained from the digital elevation model. When the actual obstacle elevation value is less than the critical obstacle height, the invalid obstacle is filtered out, and finally, the risk obstacles are further segmented in the unstructured terrain.

[0040] Preferably, in step 3, by constructing the angle between the resultant force vector of the vehicle mass and the normal vector of the state surface, the steady-state margin angle of the articulated engineering vehicle is expressed as:

[0041] ,

[0042] where, ψ 1 is the first-level steady-state margin angle, which is the minimum angle between the resultant force vector and the first-level instability plane; ψ 2 is the second-level steady-state margin angle, which is the minimum angle between the resultant force vector and the second-level instability plane; is the normal vector of the plane corresponding to the first-level instability state; is the normal vector of the plane corresponding to the second-level instability state; is the resultant force vector of the vehicle mass.

[0043] Based on the steady-state margin angle index, the driving state of the articulated engineering vehicle can be divided into three categories: when ψ 1 <0, ψ 2 <0, it is in a stable state; whenψ 1 > 0, ψ 2 < When it is less than 0, it is in the first - level instability state; when ψ 1 > 0, ψ 2 > 0, it is in the second - level instability state.

[0044] Preferably, in step 4, the process of solving the steady - state margin angle is as follows: First, calculate the X / Y coordinates of the four vehicle tire contact points, the coordinates of the vehicle center of gravity and the rear - axle hinge point; then, index the digital elevation model to determine their elevation values in the Z direction; finally, dynamically solve the steady - state margin angle in real - time through the normal vector and resultant force vector of each state surface to realize the stability state judgment of the articulated engineering vehicle.

[0045] Preferably, set the vehicle steering angle to three cases: straight - ahead, maximum left - turn, and maximum right - turn, and set the vehicle heading angle to eight directions: east, south, west, north, southeast, southwest, northwest, and northeast; a total of 24 state combinations are constructed to solve the stability state of the vehicle in different motion directions, and after solving, select the steady - state margin angle with the worst driving state among the 24 combinations to judge the current vehicle stability state.

[0046] Preferably, according to the stable state, first - level instability state, and second - level instability state of the steady - state margin angle, non - structured terrain safety semantic segmentation is respectively divided into three levels: safe area, first - level dangerous area, and second - level dangerous area.

[0047] Compared with the prior art, the beneficial effects of the present application are as follows:

[0048] (1) The present invention constructs a kinematic model considering the static roll - over characteristics of the articulated engineering vehicle, obtains the critical obstacle height that can pass through the terrain, and effectively marks and distinguishes the risk obstacles in terrain mapping based on this.

[0049] (2) The present invention designs a simple and easy - to - use steady - state margin angle index. Through the vehicle - terrain coupling solution method, this index can effectively characterize the dynamic roll - over stability of the articulated engineering vehicle and improve the accuracy of non - structured terrain safety semantic segmentation.

[0050] (3) The method of the present invention divides non - structured terrain safety semantic segmentation into three levels: safe area, first - level dangerous area, and second - level dangerous area, expands the semantic information dimension of non - structured terrain, effectively provides multi - level danger warning information for the driving operation of work vehicles, and guides the vehicle to operate safely and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flowchart of the method of the present application.

[0052] Figure 2 This is a schematic diagram of the digital gradient model after grayscale conversion in this application.

[0053] Figure 3 This is a schematic diagram of the coordinate system of an articulated engineering vehicle in this application.

[0054] Figure 4 This is a schematic diagram of the static rollover process of an articulated engineering vehicle in this application.

[0055] Figure 5 This is a schematic diagram of the steady-state margin angle of an articulated engineering vehicle in this application.

[0056] Figure 6 This is a schematic diagram of the calculation process of the steady-state margin angle in this application.

[0057] Figure 7 This is a schematic diagram of the non-structured terrain safety semantic segmentation in this application. Detailed implementation manners

[0058] Next, in combination with the detailed implementation manners, the present application will be further described. It should be noted that, on the premise of no conflict, the following-described embodiments or technical features can be arbitrarily combined with each other to form new embodiments.

[0059] In the description of the present application, it should be noted that for orientation terms, such as the terms "center", "horizontal", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., the orientation and position relationships indicated are based on the orientation or position relationship shown in the drawings. This is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of the present application.

[0060] It should be noted that the terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence.

[0061] The terms "including" and "having" and any variations thereof in the description and claims of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0062] One aspect of the present application provides an unstructured terrain safety semantic segmentation method considering vehicle-terrain coupling. First, in order to generate an unstructured terrain point cloud model for safety semantic segmentation, methods such as on-vehicle lidar or drone oblique photography are used to obtain the scanned point cloud data of the unstructured terrain. After obtaining the scanned point cloud data, filtering methods are used to filter out the abnormal noise points and invalid weed points in the point cloud. Finally, an optimized and reconstructed unstructured terrain point cloud model is obtained.

[0063] In this embodiment, as Figure 1 shown, the steps of implementing the safety semantic segmentation process in the unstructured terrain point cloud model are as follows.

[0064] Step 1: Construct a digital elevation model and a digital gradient model.

[0065] Specifically, a digital elevation model and a digital gradient model are constructed based on the unstructured terrain point cloud model, and the terrain obstacle area in the unstructured terrain is divided through the digital gradient model.

[0066] First, when the index object in the unstructured terrain point cloud model is the elevation value, it is the digital elevation model. It is obtained through the two-dimensional grid data of the unstructured terrain, which contains the geometric space coordinates of the X / Y / Z axes. The elevation value can be obtained through the grid voxelization mean processing of the neighboring point cloud. The specific process is as follows:

[0067]

[0068] Where, z ( x , y ) is the indexed elevation value; N is the total number of neighboring point clouds in the unstructured terrain point cloud model; d is the grid side length, that is, the model resolution; the point cloud ( x i , y i , z i ) is the neighboring point near the point cloud ( x , y , z ).

[0069] Then, based on the digital elevation model, the gradient value of each point in the unstructured terrain point cloud model is calculated to obtain the digital gradient model. By solving the x and y direction gradient values of each point in the digital elevation model, the discrete gradient of the digital elevation model is expressed as:

[0070] \forall f=\left [ {{^{{g}_{x}}_{{g}_{y}}}} \right ]=\left [ {\frac {\partial f} {\partial x}\, \frac {\partial f} {\partial y}} \right ]=\left [ {f(x + 1,y)-f(x,y)\, f(x,y + 1)-f(x,y)} \right ]{\,}^{T}

[0071] where \(g\) x and \(g\) y are the gradient components of the point \((x, y)\) in the \(x\) and \(y\) directions, respectively.

[0072] Furthermore, the gradient components can be obtained by calculating the Sobel operator, which is commonly used for image edge detection and segmentation. To approximately solve the absolute gradient, an improved Sobel operator is designed in this application to establish a digital gradient model, and its operation matrix is described as:

[0073]

[0074]

[0075] where l is the side length of the operator grid.

[0076] Specifically, by calculating the gradient of the digital elevation model, the gray-scaled digital gradient model is as Figure 2 shown, Figure 2 in which the contour area of the terrain obstacle can be clearly observed, thus effectively dividing the terrain obstacle area in the unstructured terrain.

[0077] Step 2: Semantic segmentation of risk obstacles.

[0078] To distinguish risk obstacles among terrain obstacles, the static rollover stability of the engineering vehicle needs to be further considered. Considering the static kinematic model of the articulated engineering vehicle, the critical obstacle height that can pass through the terrain is calculated, and based on this, the semantic segmentation of risk obstacles is further realized in the terrain obstacle area.

[0079] First, taking the articulated engineering vehicle as the operation object in the unstructured terrain environment, the steering articulation point is selected as the vehicle coordinate origin to construct the kinematic model of the articulated engineering vehicle. As Figure 3 shown, the total vehicle mass mainly includes the mass of the front body G 1 and the mass of the front wheels G 2, Front swing arm mass G 3 , Rear body mass G 4 and Rear wheel mass G 5 . The center of gravity of a specific vehicle G ( x 0 , y 0 , z 0 ) is calculated by the formula:

[0080]

[0081] where, M n is the mass of each vehicle body part; G n is the coordinate of each vehicle body part.

[0082] Further, in order to obtain the tire contact points, the left front wheel P 1l ( x 1l , y 1l , z 1l ), right front wheel P 1r ( x 1r , y 1r , z 1r ), left rear wheel P 2l ( x 2l , y 2l , z 2l ) and right rear wheel P 2r ( x 2r , y 2r , z 2r ) are obtained through geometric motion relationships and their grounding coordinates are:

[0083]

[0084]

[0085] {R}_{1}=\left [ {{^{^{\cos {\theta}}_{\sin {\theta}}}_{{\,}^{0}_{\,}}{^{{\,}^{-\sin {\theta}}_{\cos {\theta}}}_{{\,}^{0}_{\,}}{^{^{0}_{0}\,}_{^{1}_{\,}\,}\,}}}} \right ] ,{R}_{2}=\left [ {{^{^{\cos {\alpha}}_{\, \, \, 0}}_{{\,}^{-\sin {\alpha}}_{\,}}{{}^{^{0}_{1}\,}_{{\,}^{0}_{\,}}{^{^{\sin {\alpha}}_{0}\,}_{^{\cos {\alpha}}_{\,}\,}\,}}}} \right ]

[0086] where W is half of the track width; L 1 is the distance from the front axle to the center of gravity; L 2 is the distance from the rear axle to the center of gravity; H is the height of the steering articulation point; θ is the articulation steering angle; α is the swing angle of the rear axle.

[0087] It should be noted that in order to consider the influence of the obstacle height on the vehicle's passing ability, in this application, the roll angle threshold when the vehicle crosses the obstacle is calculated to eliminate the ineffective obstacles that do not affect the driving stability of the engineering vehicle. For this purpose, a critical height threshold is defined to determine whether the terrain obstacle height obtained from the digital elevation model will cause the vehicle to roll over.

[0088] Specifically, the static rollover process of the articulated engineering vehicle is as shown in Figure 4 . When the left front wheel of the vehicle is driving on the obstacle, and the roll angle caused by the wheel being lifted is less than the maximum restraint angle of the rear swing bridge, specifically as shown in (a) of Figure 4 and (b) of Figure 4 , the vehicle body will roll around P 1r P 2 axis. The roll angle generated at low speed φ is defined as:

[0089]

[0090] Among them, h is the height of the obstacle; D is from point P 1l to P 1r P 2 the projection distance to the

[0091] Furthermore, when the roll angle exceeds the maximum constraint angle of the rear swing axle, the vehicle body will rotate around P 1r P 2r the Figure 4 axis to reach the rollover critical state, as specifically shown in (c) of D = 2 W . After the vehicle undergoes these two roll motions in sequence, there will be a risk of rollover, and the limit roll angle can be obtained by calculating the vehicle geometric parameters. By selecting a safety factor δ , the critical obstacle height h 0 , h 0 can be derived as follows.

[0092]

[0093] Among them, φ max is the limit roll angle.

[0094] It can be understood that by calculating the limit roll angle of the articulated engineering vehicle, the critical obstacle height of the passable terrain can be obtained. By comparing with the actual obstacle elevation value obtained from the digital elevation model, when the actual obstacle elevation value is less than the critical obstacle height, the invalid obstacle is filtered out, and finally the risk obstacles are further segmented in the unstructured terrain.

[0095] Step 3: Judge the dynamic stability state of the articulated engineering vehicle.

[0096] The driving safety in a complex unstructured terrain environment should further consider the dynamic motion relationship between the vehicle and the terrain. For this purpose, in combination with the dynamic roll motion process of the articulated engineering vehicle, the present invention designs a simple and easy-to-use steady-state margin angle index, and based on this, the dynamic stability state of the articulated engineering vehicle is judged.

[0097] Specifically, as shown in (a) of Figure 5 , when the vehicle is in a stable state, the resultant force F is located in the plane Δ GP 1lP 1r , Δ GP 1l P 2 and Δ GP 1r P 2 within the stable range of the envelope. As shown in (b) of Figure 5 when the resultant force F is located within the roll range formed by the planes Δ GP 1l P 2 , Δ GP 1r P 2 , Δ GP 1l P 2l , Δ GP 1r P 2r and Δ GP 2l P 2r , the vehicle will roll around the lateral roll axis, and the vehicle motion state is defined as the first - level instability state. As shown in (c) of Figure 5 if the resultant force F exceeds the maximum roll range formed by the planes Δ GP 1l P 1r , Δ GP 1l P 2l , Δ GP 2l P 2r and Δ GP 1r P 2r , the vehicle will roll over, and the vehicle motion state is defined as the second - level instability state.

[0098] Further, the steady - state margin angle can be obtained by calculating the included angle between the normal vectors of each plane. The normal vector corresponding to the first - level instability state can be calculated by the following formula.

[0099] ,

[0100] Even further, the normal vector It can be calculated by the same method, and the resultant force vector Then it can be calculated from the vehicle gravity and lateral acceleration, and can be expressed as:

[0101]

[0102] Finally, the steady-state margin angle of the articulated construction vehicle is expressed as:

[0103] ,

[0104] Wherein, ψ 1 is the first-level steady-state margin angle, which is the minimum angle between the resultant force vector and the first-level instability plane; ψ 2 is the second-level steady-state margin angle, which is the minimum angle between the resultant force vector and the second-level instability plane.

[0105] Through the above calculations, based on the steady-state margin angle index, the driving states of articulated construction vehicles can be divided into three categories: when ψ 1 <0, ψ 2 <0, it is in a stable state; when ψ 1 >0, ψ 2 <0, it is in a first-level instability state; when ψ 1 >0, ψ 2 >0, it is in a second-level instability state.

[0106] Step 4: Unstructured terrain safety semantic segmentation.

[0107] By constructing a vehicle-terrain coupling solution method, the steady-state margin angle index is further calculated on the basis of static risk obstacles, and the unstructured terrain safety semantic is segmented into three levels according to the dynamic stability state: safe area, first-level danger area and second-level danger area.

[0108] Specifically, as Figure 6 shown, the X / Y coordinates of the four vehicle tire contact points, the vehicle center of gravity and the coordinates of the rear axle hinge point can be calculated through Step 2, and then the digital elevation model is indexed to determine their elevation values in the Z direction. Finally, the normal vector and resultant force vector of each state surface can be obtained through Step 3, and the steady-state margin angle can be solved in real time dynamically, so as to realize the judgment of the stable state of the articulated construction vehicle.

[0109] Furthermore, since different tire contact points will be generated when the vehicle moves in different directions, in order to solve the steady state within all driving areas at each vehicle coordinate, the vehicle steering angle is set to three cases: straight ahead, maximum left turn, and maximum right turn, and the vehicle heading angle is set to eight directions: east, south, west, north, southeast, southwest, northwest, and northeast. A total of 24 state combinations are constructed to calculate the steady state of the vehicle in different moving directions, and the steady state margin angle with the worst driving state among the 24 combinations is selected to judge the current vehicle steady state.

[0110] Specifically, as Figure 7 shown, select an actual terrain and calculate it according to the method in this application. According to the steady state, primary instability state, and secondary instability state of the steady state margin angle, the unstructured terrain safety semantics can be segmented into three levels: safe area, primary danger area, and secondary danger area. From Figure 7 it can be seen that as the slope of the unstructured terrain increases, the primary danger area and the secondary danger area increase significantly, and the secondary danger area appears in the slope terrain with a larger slope, making the vehicle rollover risk increase. At this time, the driving safety will be greatly reduced, which effectively provides danger warning information for the operation of the work vehicle and guides the vehicle to operate safely.

[0111] The above describes the basic principle, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of this application. Without departing from the spirit and scope of this application, this application will have various changes and improvements, and these changes and improvements all fall within the scope of this application claimed. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A safety semantic segmentation method for unstructured terrain considering vehicle-terrain coupling, characterized in that: The target is scanned in advance to construct an unstructured terrain point cloud. The secure semantic segmentation method implemented by the unstructured terrain point cloud model includes the following steps: Step 1: Build a digital model; Building a digital elevation model and a digital gradient model based on the unstructured terrain point cloud model, and dividing terrain obstacle areas in the unstructured terrain by using the digital gradient model; Step 2: Semantic segmentation of risk obstacles; The static kinematic model of the articulated engineering vehicle is used to calculate the critical obstacle height that can pass through the terrain. Based on this, the semantic segmentation of risk obstacles is achieved in the terrain obstacle area. Step 3: Determine the dynamic stability state of the articulated engineering vehicle; Combined with the dynamic roll motion process of articulated engineering vehicles, a simple and easy-to-use steady-state margin angle index is designed, based on which the dynamic stability state judgment of articulated engineering vehicles is realized; By constructing the angle between the vehicle mass force vector and the state surface normal vector, the steady-state margin angle of the articulated engineering vehicle is expressed as: ， ； in, ψ 1 is the primary steady-state margin angle, which is the minimum angle between the resultant force vector and the primary instability plane; ψ 2 is the secondary stable margin angle, which is the minimum angle between the resultant force vector and the secondary instability plane; is the plane normal vector corresponding to the first-level instability state; is the plane normal vector corresponding to the secondary instability state; is the vehicle mass force vector; Based on the steady-state margin angle index, the driving state of articulated engineering vehicles can be divided into three categories: ψ 1<0, ψ 2<0 is a stable state; ψ 1>0, ψ When 2<0, it is the first-level instability state; when ψ 1>0, ψ When 2>0, it is the secondary instability state; Step 4: Unstructured terrain safety semantic segmentation; A vehicle-terrain coupling solution method is constructed, and the safety semantics of the non-structural terrain are divided into three levels: safe area, primary dangerous area and secondary dangerous area according to the stable state, primary instability state and secondary instability state of the steady-state margin angle.

2. The method for safe semantic segmentation of unstructured terrain considering vehicle-terrain coupling according to claim 1, characterized in that: When the index object in the unstructured terrain point cloud model in step 1 is an elevation value, the digital elevation model is used, and the specific calculation is as follows: ; in, z ( x , y ) is the elevation value of the index; N is the total number of points in the neighborhood point cloud model of unstructured terrain; d is the rasterized edge length, i.e. the model resolution; point cloud ( x i , y i , z i ) is the point cloud ( x , y , z ) nearby neighboring points; The discrete gradient of the digital elevation model is expressed as: ； in, g x , g y They are points ( x , y )exist x , y Directional gradient component; The gradient component of the digital gradient model is calculated as: ； ； in, l is the side length of the operator grid; By calculating the gradient of the digital elevation model, the terrain obstacle area in the unstructured terrain can be divided.

3. The method for safe semantic segmentation of unstructured terrain considering vehicle-terrain coupling according to claim 1, characterized in that: In step 2, the steering hinge point is selected as the vehicle coordinate origin to construct the kinematic model of the articulated engineering vehicle; the vehicle mass mainly includes the front body mass G1, the front wheel mass G2, the front swing arm mass G3, the rear body mass G4 and the rear wheel mass G5; the center of gravity G(x0, y0, z0) of the vehicle is: ； in, M n is the mass of each body part; G n is the coordinates of each body part; The contact points of each tire are obtained through geometric motion relationships. P 1l ( x 1l , y 1l , z 1l ), right front wheel P 1r ( x 1r , y 1r , z 1r ), left rear wheel P 2l ( x 2l , y 2l , z 2l ) and right rear wheel P 2r ( x 2r , y 2r , z 2r )’s ground coordinates are: ； ； , ； in, W 1 / 2 wheelbase width; L 1 is the distance from the front axle to the center of gravity; L 2 is the distance from the rear axle to the center of gravity; H is the height of the steering hinge point; θ is the articulated steering angle; α is the swing angle of the rear axle; During the static rollover motion of an articulated construction vehicle, the roll angle generated at low speed φ Defined as: ； in, h is the obstacle height; D From the tire contact point P 1l Projected distance to the roll axis; Then, take a safety factor δ , the critical obstacle height can be derived h 0 is: ； in, φ max is the limit rollover angle.

4. The method for safe semantic segmentation of non-structural terrain considering vehicle-terrain coupling as claimed in claim 3, characterized in that: By calculating the maximum rollover angle of the articulated engineering vehicle, the critical obstacle height of the passable terrain is obtained; the critical obstacle height of the passable terrain is compared with the actual obstacle elevation value obtained in the digital elevation model. When the actual obstacle elevation value is less than the critical obstacle height, the invalid obstacle is filtered out, and finally the risk obstacle is further segmented in the unstructured terrain.

5. The method for safe semantic segmentation of unstructured terrain considering vehicle-terrain coupling according to claim 1, characterized in that: The process of solving the steady-state margin angle is as follows: first, calculate the X / Y coordinates of the four vehicle tire contact points, the coordinates of the vehicle center of gravity and the rear axle articulation point; then, index the digital elevation model to determine their elevation values ​​in the Z direction; finally, the steady-state margin angle is obtained through real-time dynamic solution of the normal vector and resultant force vector of each state surface to realize the stable state judgment of the articulated engineering vehicle.

6. The method for safe semantic segmentation of non-structural terrain considering vehicle-terrain coupling as claimed in claim 5, characterized in that: The vehicle steering angle is set to three situations: straight driving, maximum left turn and maximum right turn, and the vehicle heading angle is set to eight directions: east, south, west, north, southeast, southwest, northwest and northeast. A total of 24 state combinations are constructed to solve the stability of the vehicle in different movement directions. After solving, the steady-state margin angle with the worst driving state among the 24 combinations is selected to determine the current vehicle stability state.

Images