BIM-based automatic scene recognition and positioning navigation system for construction machinery

Abstract

The application discloses a BIM-based automatic scene recognition and positioning navigation system for construction machinery, and relates to the technical field of building construction.The system comprises a data acquisition unit, an automatic scene recognition unit, a path planning unit and an execution unit, generates an environment perception model by acquiring real-time three-dimensional space data sets and real-time obstacle information sets around the construction machinery, calculates a feature vector related to a preset scene category based on a preset BIM model, outputs a scene category label and a confidence level according to a deep learning algorithm, generates a comprehensive environment model in combination with preset geometric information of a building structure and construction progress data in the BIM model, calculates an optimal path from a current position to a target position, and finally generates an optimal navigation path and sends the optimal navigation path to the execution unit to convert the optimal navigation path into a motion control instruction for the construction machinery, so that the construction machinery can travel along the optimal navigation path and effectively avoid entering a dangerous area, thereby ensuring construction safety.

Description

Technical Field

[0001] This invention relates to the field of building construction technology, specifically to a BIM-based automatic scene recognition and positioning navigation system for building machinery. Background Technology

[0002] Building construction refers to the activities of building and installing a building in accordance with the requirements of the design documents and construction drawings after the planning and design phase of a building project. It includes the entire production process from construction preparation to project completion and acceptance, covering multiple stages such as foundation engineering, main structure, roofing engineering, and decoration engineering.

[0003] In existing navigation and positioning methods, construction machinery often struggles to effectively avoid entering dangerous areas, such as deep pits during construction, areas near high-voltage lines, and areas where hazardous materials are stored. The existence of these dangerous areas poses significant safety risks to the construction process. Furthermore, due to the lack of real-time and accurate environmental perception and intelligent recognition capabilities in existing technologies, construction machinery often fails to identify and avoid these risks in a timely manner, leading to construction accidents.

[0004] Furthermore, as the complexity and scale of construction projects continue to increase, progress control and resource management during construction are becoming increasingly important. However, existing navigation and positioning methods often fail to fully integrate the geometric information of the building structure and construction progress data, resulting in unreasonable path planning and low construction efficiency. This not only affects the overall construction progress but also increases construction costs and safety risks.

[0005] Therefore, in view of this, the present invention proposes a BIM-based automatic scene recognition and positioning navigation system for building machinery to make up for and improve the shortcomings of the existing technology. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a BIM-based automatic scene recognition and positioning navigation system for construction machinery, thereby resolving the corresponding technical issues raised in the background section.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is: a BIM-based automatic scene recognition and positioning navigation system for construction machinery, comprising: a data acquisition unit, an automatic scene recognition unit, a path planning unit, and an execution unit;

[0008] The data acquisition unit is used to collect a three-dimensional spatial dataset D and a real-time obstacle information set R around the construction machinery in real time through a sensor group installed on the construction machinery. The data is then processed by a spatiotemporal alignment and information fusion algorithm to generate an environmental perception model, which is then sent to the automatic scene recognition unit.

[0009] The automatic scene recognition unit is used to acquire an environmental perception model, and combine it with a preset BIM model to perform spatial coordinate alignment and feature association, thereby obtaining a feature vector F related to a preset scene category. SC It identifies the current scene category of the construction machinery based on deep learning algorithms, and outputs the scene category label L and confidence level A to the path planning unit.

[0010] The path planning unit is used to obtain scene category labels L and confidence levels A, and to obtain real-time obstacle information set R from the data acquisition unit. Combined with the geometric information of the pre-set building structure and construction progress data in the BIM model, it generates a comprehensive environment model E, and calculates the current position p based on the comprehensive environment model E. star to target position p goal Find the optimal path and generate the optimal navigation path p. opt And send it to the execution unit;

[0011] The execution unit is used to obtain the optimal navigation path p. opt This information is then converted into motion control commands for the construction machinery, driving it to follow the optimal navigation path.

[0012] The sensor group includes lidar, millimeter-wave radar, camera, and ultrasonic sensor.

[0013] The scene category labels include construction area, danger zone, densely populated area and obstacle zone. The confidence level A represents the reliability of the classification result and has a value range of 0-1.

[0014] The specific process of generating the environmental perception model is as follows:

[0015] S101. Obtain the 3D spatial dataset D and the real-time obstacle information set R, and perform data preprocessing to obtain the processed 3D spatial dataset D' and the real-time obstacle information set R'. Perform spatiotemporal alignment on the processed 3D spatial dataset D' and the real-time obstacle information set R' using a spatiotemporal alignment algorithm. Then, use a weighted fusion algorithm to fuse the aligned data to obtain the fused data F. DR Its formula is: F DR =(D',R');

[0016] S102, From the fused data F DR The spatial location features of construction machinery and obstacles, the shape features of obstacles, and the size features of obstacles are extracted and combined to form a feature vector F. A convolutional neural network is used to train the feature vector F to generate an environmental perception model.

[0017] The specific process of outputting scene labels and confidence scores is as follows:

[0018] S201. Obtain the environmental perception model and extract features from it to obtain the three-dimensional spatial structural features, obstacle location features, and dynamic change information features around the construction machinery, and combine them to form a comprehensive feature vector F. Z ;

[0019] S202, Combine the feature vector F Z Align the spatial coordinates with the preset BIM model to obtain the aligned comprehensive feature vector F. Z Its spatial coordinate alignment formula is: F Z '=T(F Z M BIM ), where T is the spatial coordinate alignment function, M BIM It is a pre-defined BIM model;

[0020] S203, regarding the aligned comprehensive feature vector F Z Perform feature association, and align the comprehensive feature vector F. Z Each feature in the model is matched and associated with a preset scene feature in the preset BIM model to obtain a feature vector F related to the preset scene category. SC ;

[0021] S204. Use a convolutional neural network to process the feature vector F related to the preset scene category. SC Perform scene classification to obtain scene category labels L, and output the confidence score A for each scene category. The scene classification formula is: L,A=k(F SC ), where k is the scene classifier function.

[0022] The specific process for generating the comprehensive environment model E is as follows:

[0023] S301. Obtain a real-time obstacle information set R from the data acquisition unit, wherein the real-time obstacle information set R includes obstacle positions (x, y, y). i ,y i ,z i ), velocity v i Direction θ i and size (l) i ,w i ,h i ), construct a dynamic obstacle model O i Its formula is: O i ={(x i ,y i ,z i ),v i ,θ i ,(l i ,w i ,h i )};

[0024] S302. Extract the geometric information of the building structure and construction progress data from the preset BIM model. The geometric information of the building structure includes the wall location (x... n ,y n ,z n ) and pillar position (x m ,y m ,z m The construction progress data includes the location of temporary facilities (x...). j ,y j ,z j Construct a static environment model J, whose formula is: J={(x n ,y n ,z n ),(x m ,y m ,z m ),(x j ,y j ,z j )};

[0025] S303, Dynamic obstacle model O i By integrating with the static environment model J, a comprehensive environment model E is generated, and its calculation formula is: E=J∪{O i |i=1,2,......,N}, where N is the number of obstacles.

[0026] Generate the optimal navigation path p opt The specific process is as follows:

[0027] S401. Use the A* algorithm to search for a position p in the comprehensive environment model E from the current position p. star to target position p goal The candidate path set P = {p1, p2, ..., p K}, where K is the number of candidate paths;

[0028] S402, For each candidate path p K The cost C(p) K The calculation is performed using the following formula: C(p) K )=α×C d (p K )+β×C o (p K )+γ×C s (p K And satisfy α+β+γ=1;

[0029] Among them, C d (p K() represents the distance cost, indicating the total path length;

[0030] C o (p K ) is the obstacle cost, which represents the reciprocal of the minimum distance between the path and the obstacle;

[0031] C s (p K () is the scene cost, which represents the sum of the weights of the scenes traversed by the path;

[0032] S403. Use B-spline curves to process each candidate path p. K Fitting to generate a smooth path p smooth Further optimization using gradient descent reduces the path cost C(p). smooth ), thus obtaining the optimized path p opt Its formula is: p opt =p smooth -η×▽C(p smooth ), where η is the learning rate, ▽C(p smooth ) is the gradient of the cost function;

[0033] S404, Compare all optimized paths p opt The path with the lowest cost is selected as the optimal navigation path p. opt ', and the optimal navigation path p opt 'Perform verification to obtain the distance d between all points on the path and the obstacles.' j Greater than the safe distance d safe And the cost C(p) of optimizing the path opt The cost threshold C(p) is less than the preset cost threshold. max ), and the optimal navigation path p opt 'It is converted into a navigation instruction sequence and sent to the execution unit.'

[0034] Compared with existing technologies, the beneficial effects of this invention are as follows: By acquiring real-time three-dimensional spatial datasets and real-time obstacle information sets around construction machinery, an environmental perception model is generated. Based on a preset BIM model, feature vectors related to preset scene categories are calculated. Scene category labels and confidence scores are output according to deep learning algorithms. Combined with the geometric information of the preset building structure and construction progress data in the BIM model, a comprehensive environmental model is generated. The optimal path from the current position to the target position is calculated, and finally, the optimal navigation path is generated and sent to the execution unit to be converted into motion control commands for the construction machinery. This drives the construction machinery to travel along the optimal navigation path, effectively preventing the construction machinery from entering dangerous areas, ensuring construction safety, and reducing safety risks during construction. Attached Figure Description

[0035] Figure 1This is a schematic diagram of the overall structure of a preferred embodiment of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Embodiments of the present invention:

[0038] Please refer to Figure 1 As shown, the BIM-based automatic scene recognition and positioning navigation system for construction machinery includes: a data acquisition unit, an automatic scene recognition unit, a path planning unit, and an execution unit.

[0039] The data acquisition unit is used to collect a three-dimensional spatial dataset D and a real-time obstacle information set R around the construction machinery in real time through a sensor group installed on the construction machinery. The data is then processed by a spatiotemporal alignment and information fusion algorithm to generate an environmental perception model, which is then sent to the automatic scene recognition unit.

[0040] The automatic scene recognition unit is used to acquire an environmental perception model, and combine it with a preset BIM model to perform spatial coordinate alignment and feature association, thereby obtaining a feature vector F related to a preset scene category. SC It identifies the current scene category of the construction machinery based on deep learning algorithms, and outputs the scene category label L and confidence level A to the path planning unit.

[0041] The path planning unit is used to obtain scene category labels L and confidence levels A, and to obtain real-time obstacle information set R from the data acquisition unit. Combined with the geometric information of the pre-set building structure and construction progress data in the BIM model, it generates a comprehensive environment model E, and calculates the current position p based on the comprehensive environment model E. star to target position p goal Find the optimal path and generate the optimal navigation path p. opt And send it to the execution unit;

[0042] The execution unit is used to obtain the optimal navigation path p. opt It converts these into motion control commands for the construction machinery, driving the machinery to travel along the optimal navigation path;

[0043] The sensor group includes lidar, millimeter-wave radar, camera, and ultrasonic sensor;

[0044] The scene category labels include construction area, danger area, densely populated area and obstacle area. The confidence level A represents the reliability of the classification result and has a value range of 0-1.

[0045] The specific process of generating the environmental perception model is as follows:

[0046] S101. Obtain the 3D spatial dataset D and the real-time obstacle information set R, and perform data preprocessing to obtain the processed 3D spatial dataset D' and the real-time obstacle information set R'. Perform spatiotemporal alignment on the processed 3D spatial dataset D' and the real-time obstacle information set R' using a spatiotemporal alignment algorithm. Then, use a weighted fusion algorithm to fuse the aligned data to obtain the fused data F. DR Its formula is: F DR =(D',R');

[0047] S102, From the fused data F DR The spatial location features of construction machinery and obstacles, the shape features of obstacles, and the size features of obstacles are extracted and combined to form a feature vector F. A convolutional neural network is used to train the feature vector F to generate an environmental perception model.

[0048] The specific process of outputting scene labels and confidence scores is as follows:

[0049] S201. Obtain the environmental perception model and extract features from it to obtain the three-dimensional spatial structural features, obstacle location features, and dynamic change information features around the construction machinery, and combine them to form a comprehensive feature vector F. Z ;

[0050] S202, Combine the feature vector F Z Align the spatial coordinates with the preset BIM model to obtain the aligned comprehensive feature vector F. Z Its spatial coordinate alignment formula is: F Z '=T(F Z M BIM ), where T is the spatial coordinate alignment function, M BIM It is a pre-defined BIM model;

[0051] S203, regarding the aligned comprehensive feature vector F Z Perform feature association, and align the comprehensive feature vector F. Z Each feature in the model is matched and associated with a preset scene feature in the preset BIM model to obtain a feature vector F related to the preset scene category. SC ;

[0052] S204. Use a convolutional neural network to process the feature vector F related to the preset scene category.SC Perform scene classification to obtain scene category labels L, and output the confidence score A for each scene category. The scene classification formula is: L,A=k(F SC ), where k is the scene classifier function;

[0053] The specific process for generating the comprehensive environment model E is as follows:

[0054] S301. Obtain a real-time obstacle information set R from the data acquisition unit, wherein the real-time obstacle information set R includes obstacle positions (x, y, y). i ,y i ,z i ), velocity v i Direction θ i and size (l) i ,w i ,h i ), construct a dynamic obstacle model O i Its formula is: O i ={(x i ,y i ,z i ),v i ,θ i ,(l i ,w i ,h i )};

[0055] S302. Extract the geometric information of the building structure and construction progress data from the preset BIM model. The geometric information of the building structure includes the wall location (x... n ,y n ,z n ) and pillar position (x m ,y m ,z m The construction progress data includes the location of temporary facilities (x...). j ,y j ,z j Construct a static environment model J, whose formula is: J={(x n ,y n ,z n ),(x m ,y m ,z m ),(x j ,y j ,z j )};

[0056] S303, Dynamic obstacle model O i By integrating with the static environment model J, a comprehensive environment model E is generated, and its calculation formula is: E=J∪{O i|i=1,2,......,N}, where N is the number of obstacles;

[0057] Generate the optimal navigation path p opt The specific process is as follows:

[0058] S401. Use the A* algorithm to search for a position p in the comprehensive environment model E from the current position p. star to target position p goal The candidate path set P = {p1, p2, ..., p K}, where K is the number of candidate paths;

[0059] S402, For each candidate path p K The cost C(p) K The calculation is performed using the following formula: C(p) K )=α×C d (p K )+β×C o (p K )+γ×C s (p K And satisfy α+β+γ=1;

[0060] Among them, C d (p K () represents the distance cost, indicating the total path length;

[0061] C o (p K ) is the obstacle cost, which represents the reciprocal of the minimum distance between the path and the obstacle;

[0062] C s (p K () is the scene cost, which represents the sum of the weights of the scenes traversed by the path;

[0063] S403. Use B-spline curves to process each candidate path p. K Fitting to generate a smooth path p smooth Further optimization using gradient descent reduces the path cost C(p). smooth ), thus obtaining the optimized path p opt Its formula is: p opt =p smooth -η×▽C(p smooth ), where η is the learning rate, ▽C(p smooth ) is the gradient of the cost function;

[0064] S404, Compare all optimized paths p opt The path with the lowest cost is selected as the optimal navigation path p. opt ', and the optimal navigation path popt 'Perform verification to obtain the distance d between all points on the path and the obstacles.' j Greater than the safe distance d safe And the cost C(p) of optimizing the path opt The cost threshold C(p) is less than the preset cost threshold. max ), and the optimal navigation path p opt 'It is converted into a navigation instruction sequence and sent to the execution unit.'

[0065] By acquiring real-time 3D spatial datasets and obstacle information sets around construction machinery, an environmental perception model is generated. Based on a pre-set BIM model, feature vectors related to pre-set scene categories are calculated. Scene category labels and confidence scores are output using deep learning algorithms. Combined with the geometric information of the pre-set building structure and construction progress data in the BIM model, a comprehensive environmental model is generated. The optimal path from the current location to the target location is calculated, and the optimal navigation path is finally generated and sent to the execution unit to be converted into motion control commands for the construction machinery. This drives the construction machinery to travel along the optimal navigation path, effectively preventing the construction machinery from entering dangerous areas, ensuring construction safety, and reducing safety risks during construction.

[0066] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0067] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0068] In the two embodiments provided in this application, it should be understood that the disclosed apparatus and system can be implemented in other ways; for example, the apparatus embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; furthermore, the coupling or direct coupling or communication connection between the shown or discussed mutuals can be through some interfaces, and the indirect coupling or communication connection between the apparatus or modules can be electrical, mechanical or other forms.

[0069] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A BIM-based automatic construction machine scene recognition and positioning navigation system, characterized by, Comprise: Data acquisition unit, automatic scene recognition unit, path planning unit and execution unit; The data acquisition unit is used to collect the three-dimensional space data set D and the real-time obstacle information set R around the construction machinery in real time through the sensor group arranged on the construction machinery, process them through space-time alignment and information fusion algorithm, and generate an environment perception model sent to the automatic scene recognition unit; The automatic scene recognition unit is configured to obtain an environmental perception model, perform spatial coordinate alignment and feature association in combination with a preset BIM model, and obtain a feature vector F related to a preset scene category SC And according to the deep learning algorithm, the current scene category of the construction machinery is recognized, and a scene category label L and a confidence A are output and sent to the path planning unit. The path planning unit is configured to acquire a scene category label L and a confidence A, acquire a real-time obstacle information set R from the data acquisition unit, generate a comprehensive environment model E in combination with preset geometric information of a building structure and construction progress data in the BIM model, and calculate an optimal path from a current position p star to a target position p goal based on the comprehensive environment model E, and generate an optimal navigation path p opt and send the optimal navigation path p to the execution unit. The execution unit is configured to acquire the optimal navigation path p opt and convert it into a motion control instruction of the construction machine, and drive the construction machine to travel along the optimal navigation path.

2. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 1, characterized by, The sensor group includes a laser radar, a millimeter wave radar, a camera and an ultrasonic sensor.

3. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 2, characterized in that, The scene category label includes a construction area, a dangerous area, a personnel-intensive area and an obstacle area, and the confidence A represents the reliability of the classification result, with a value range of 0-1.

4. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 3, characterized by, The specific process of generating the environment perception model is as follows: S101, acquire three-dimensional space data set D and real-time obstacle information set R and carry out data preprocessing to obtain processed three-dimensional space data set D' and real-time obstacle information set R', perform space-time alignment on the processed three-dimensional space data set D' and real-time obstacle information set R' through a space-time alignment algorithm, fuse the aligned data by using a weighted fusion algorithm, and obtain fused data F DR The formula is: F DR =(D', R'); S102、extract the spatial position features of the construction machinery and the obstacles, the shape features of the obstacles, and the size features of the obstacles from the fused data F DR The spatial position features of the construction machinery and the obstacles, the shape features of the obstacles, and the size features of the obstacles are extracted from the fused data F, combined to form a feature vector F, the feature vector F is trained by using a convolutional neural network, and an environment perception model is generated.

5. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 4, characterized in that, The specific process of outputting the scene label and the confidence is as follows: S201, acquire an environment perception model, and perform feature extraction on the environment perception model to obtain three-dimensional space structure features, obstacle position features, and dynamic change information features around the construction machinery, and combine them to form a comprehensive feature vector F Z ; S202, Combine the feature vector F Z Align the spatial coordinates with the preset BIM model to obtain the aligned comprehensive feature vector F. Z Its spatial coordinate alignment formula is: F Z '=T(F Z M BIM ), where T is the spatial coordinate alignment function, M BIM It is a pre-defined BIM model; S203, match and correlate each feature in the aligned integrated feature vector F Z with the preset scene features in the preset BIM model to obtain a feature vector F Z related to the preset scene category. SC ; S204, adopting a convolutional neural network to the feature vector F related to the preset scene category SC scene classification, obtaining a scene category label L, and outputting a confidence A of each scene category, and the scene classification formula is: L, A=k(F SC ), wherein k is a scene classifier function.

6. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 5, characterized in that, The specific process of generating the comprehensive environment model E is as follows: S301, acquiring a real-time obstacle information set R from a data acquisition unit, the real-time obstacle information set R including obstacle positions (x i ,y i ,z i ), speeds v i , directions θ i , and sizes (l i ,w i ,h i ), constructing a dynamic obstacle model O i , a formula of which is: O i ={(x i ,y i ,z i ),v i ,θ i ,(l i ,w i ,h i )} ; S302, extract the geometric information of the building structure and the construction progress data from the preset BIM model, the geometric information of the building structure includes wall position (x n ,y n ,z n ) and column position (x m ,y m ,z m ), the construction progress data includes temporary facility position (x j ,y j ,z j ), and a static environment model J is constructed, and the formula is: J={(x n ,y n ,z n ),(x m ,y m ,z m ),(x j ,y j ,z j )}. S303、fuse the dynamic obstacle model O i with the static environment model J to generate a comprehensive environment model E, whose calculation formula is: E = J∪{O i |i = 1, 2, …, N}, wherein, N is the number of obstacles.

7. The BIM-based construction machine automatic scene recognition and positioning navigation system according to claim 6, characterized in that, Generating an optimal navigation path p opt The detailed procedure is as follows: S401. Use the A* algorithm to search for a position p in the comprehensive environment model E from the current position p. star to target position p goal The candidate path set P = {p1, p2, ..., p K }, where K is the number of candidate paths; S402, calculate the cost C(p K ) of each candidate path p K , the calculation formula of which is: C(p K ) = α × C d (p K ) + β × C o (p K ) + γ × C s (p K ), and α + β + γ = 1 is satisfied; where C d (p K ) is the distance cost, representing the total length of the path. C o (p K ) is an obstacle cost representing the reciprocal of the minimum distance of the path from the obstacle; C s (p K ) is the scene cost, representing the sum of the weights of the paths through the scene; S403、adopt B spline curve to each candidate path p K fitting to generate smooth path p smooth , further optimization by gradient descent method, reduce path cost C(p smooth ), get optimized path p opt , the formula is: Wherein, η is learning rate, is the gradient of the cost function; S404, compare the cost of all optimization paths p opt , select the path with the minimum cost as the optimal navigation path p opt opt , and verify the optimal navigation path p j ' to obtain the distance d safe between all points on the path and the obstacle, and the cost C(p opt ) of the optimization path is less than the preset cost threshold C(p max ), and convert the optimal navigation path p opt ' into a navigation instruction sequence and send it to the execution unit.​

Images