Intelligent station driving monitoring method, equipment and system

Abstract

The invention discloses a smart station driving monitoring method, device and system, and the method comprises the steps: a main station recognition step: calling a visual pickup device to scan environment information, forming a station topographic map, and forming a parking space three-dimensional coordinate on the station topographic map; a vehicle identification step: identifying information of all vehicles in the range of the visual pickup device, forming three-dimensional coordinates of the vehicles, marking positions and contours of static vehicles to form a static information set, and marking moving vehicles to form a moving information set; in the motion tracking step, the motion information set is responded, and visual pickup equipment is called to track vehicle movement based on a Hungary algorithm; a collision risk analysis step: identifying at least three feature points on the outer contour of the vehicle in the motion information set, calculating the minimum Euclidean distance between the feature points and surrounding entities based on a local attention mechanism, and outputting collision early warning information when the minimum Euclidean distance reaches a threshold value; and a collision early warning step: responding to the collision early warning information, and calling a directional acoustic array and / or an underground lamp to give out collision early warning. Through the scheme, more efficient and orderly driving monitoring and management can be realized.

Description

Technical Field

[0001] This invention relates to the field of smart depot technology, and in particular to a smart depot driving monitoring method, equipment and system. Background Technology

[0002] With the development of the times, vehicles have become the most numerous means of transportation, and due to the large number of vehicles, more and more large parking lots have appeared in various public places. Even with the existence of a large number of parking lots, problems such as blind spots of vehicles, parking congestion, difficulty in parking, and easy collisions caused by the large number of vehicles still plague parking lot operators.

[0003] With the gradual maturation of basic detection equipment such as various sensors and cameras, and the increasingly diversified role of AI algorithms, multiple detection devices are being added to the monitoring process of parking lots. Based on the integration of AI algorithms, more intelligent functions that were previously impossible with basic monitoring systems can be realized. This can effectively help operators solve the aforementioned problems. Therefore, more and more parking lots, especially those with high accident rates, are beginning to introduce new intelligent monitoring systems and solutions. Summary of the Invention

[0004] The main objective of this invention is to provide a smart depot driving monitoring method, device, and system, which aims to achieve richer depot monitoring capabilities by introducing an AI system, especially in scenarios where vehicles are driving within the depot, to achieve better dynamic monitoring results.

[0005] To achieve the above objectives, the present invention proposes a smart depot operation monitoring method, comprising: The site identification process involves calling a visual pickup device to scan environmental information, generating a site topographic map, and then creating three-dimensional coordinates of parking spaces on the site topographic map. The vehicle recognition process involves identifying all vehicle information within the range of the visual acquisition device, forming three-dimensional coordinates of the vehicles, marking the positions and outlines of stationary vehicles to form a static information set, and marking moving vehicles to form a motion information set. The motion tracking steps involve responding to a set of motion information and calling upon a visual pickup device to track vehicle movement based on a Hungarian algorithm. The collision risk analysis steps involve identifying at least three feature points on the outer contour of the vehicle with concentrated motion information, calculating the minimum Euclidean distance between the feature points and surrounding entities based on a local attention mechanism, responding to external input collision warning thresholds, and outputting collision warning information when the threshold is reached. The collision warning procedure involves responding to the collision warning information by calling a directional acoustic array and / or in-ground lights to issue a collision warning.

[0006] The collision risk step involves invoking a spatiotemporal token mechanism to predict the vehicle's path for the next 5 seconds through a linear transformation of historical trajectories.

[0007] Furthermore, the motion tracking steps also include: responding to the motion information set, marking the dynamic vehicle, and calling on at least two visual pickup devices closest to it to output focusing instructions to track the dynamic vehicle; Each vision pickup device matches the maximum number of vehicles it tracks based on the number of its cameras, and the focusing command output to each vision pickup device does not exceed the number of its cameras.

[0008] Furthermore, when the focusing commands output to a single vision pickup device exceed the number of its cameras, the following judgment is made: Are there still available visual acquisition devices that can observe the dynamic vehicle and are in an idle state? If so, invoke the vision pickup device. If not, generate an early warning message and output it to the mobile terminal.

[0009] Furthermore, it also includes a blind spot screening step, which calls all visual pickup devices to move within their entire range of movement to form blind spot identification data, calls the site topographic map to compare with the blind spot identification data, identifies areas that cannot be covered by the visual range of the visual pickup devices, and outputs a set of blind spots.

[0010] Furthermore, the blind spot screening step also includes: continuously retrieving static information sets and motion information sets, simultaneously incorporating comparison actions, identifying changes in areas that cannot be covered by the visual range of the visual pickup device, and outputting a dynamic blind spot set.

[0011] Furthermore, the blind spot screening step also includes: responding to the dynamic blind spot set, calling up the lidar, and assisting the visual pickup device in filling in the blind spot location information.

[0012] Furthermore, a two-stream network model is embedded in all of the aforementioned visual pickup devices.

[0013] Furthermore, it also includes a path guidance step, which calls the station topographic map and static information set and motion information set, analyzes the motion intention of moving vehicles and available parking spaces, calls the ant colony optimization algorithm, calculates multi-vehicle paths, and guides vehicles to available parking spaces.

[0014] This invention also proposes a smart depot driving monitoring device, which is connected to the driving monitoring method described above, including: Base; At least two cameras are mounted on a base, all of which are arranged in a direction perpendicular to the base, and the cameras are rotatable within a horizontal plane at least 120 degrees. A directional sound array is mounted on a base and rotates 360 degrees in the horizontal plane. The lidar is mounted on a base and rotates 360 degrees in the horizontal plane. The central storage unit stores a dual-stream network model, and all cameras and / or directional sound arrays and / or lidars are signal-connected to the central storage unit.

[0015] This invention also proposes a smart depot operation monitoring system, comprising: The hardware layer includes various types of hardware used to perform information acquisition activities, including... The visual acquisition device is installed at least one location, such as on the station's support column, roof, or wall. A directional sound array is installed at least one location, such as on the support column, roof, or wall of the site. The lidar is installed at least one location, such as on the support column, roof, or wall of the site. The perception layer calls data from the hardware layer to perform operations such as site identification, vehicle identification, and blind spot screening. The decision-making layer calls upon the data output from the perception layer to execute various operations in the motion tracking, path guidance, and collision risk analysis steps. The execution layer calls the data output from the decision layer and executes the collision warning steps.

[0016] By adopting the above-described solution, the technical solution of this invention can achieve the following main functions: In complex parking lot environments, it can effectively identify and predict vehicle movement trajectories, reduce the probability of collisions, and improve the safety of parking lots. The system anticipates the driving intentions of vehicles in the depot and plans and adjusts driving routes accordingly to improve the smoothness of traffic flow in the depot. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram illustrating the steps of the driving monitoring method of the present invention; Figure 2 This is a schematic diagram of the hierarchical modules of the driving monitoring system of the present invention; Figure 3 This is a schematic diagram of the first embodiment of the driving monitoring device of the present invention; Figure 4 This is a schematic diagram of a second embodiment of the driving monitoring device of the present invention; Figure 5 This is a cross-sectional view of a second embodiment of the driving monitoring device of the present invention; Figure 6 A schematic diagram of a smart depot using the driving monitoring method, equipment and system of the present invention; Figure 7 A schematic diagram of a smart depot using the driving monitoring method and system of the present invention; Figure 8 This is a schematic diagram of a support column in a smart depot using the driving monitoring method and system of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0022] like Figure 1 As shown, this invention discloses a smart depot driving monitoring method, including the following steps and operations: In the site identification step 101, the visual pickup device is invoked to scan the environmental information within the recognition range of all visual pickup devices, and the BEV semantic segmentation network of the ParkPredict+ model is introduced. The AI ​​analysis is performed by inputting a real-time bird's-eye view stitched from multiple cameras, and the three-dimensional coordinates and attribute labels of each parking space in the space are output. At the same time, the site topographic map is constructed.

[0023] During the construction of the site topographic map, SLAM technology is used to fuse LiDAR point cloud and visual data to generate a site topographic map with centimeter-level accuracy, and all parking space information is marked on the topographic map.

[0024] Among them, the attribute tags can be status information such as occupied, idle, or overdue.

[0025] In vehicle recognition step 102, information collected from visual pickup devices, geomagnetic sensing devices, and lidar is simultaneously retrieved. Based on a dual-stream network model, vehicle information is identified, and the static vehicle information is integrated into a static information set, while the moving vehicle information is formed into a motion information set.

[0026] In the vehicle information recognition process, the RGB stream of the dual-stream network model calls the LPRNet model to recognize vehicle outlines and license plates, while the optical flow of the dual-stream network model is used to perform motion h-vector calculations for moving vehicles based on the Farneback algorithm.

[0027] The generated static information set includes vehicle contour data and vehicle 3D coordinate data, with the 3D coordinate data stored as a point cloud matrix. The motion information set includes motion data such as the vehicle's position, trajectory sequence, velocity vector, and acceleration vector.

[0028] In motion tracking step 103, in response to the motion information set, the visual pickup device is invoked to track vehicle movement.

[0029] The motion tracking step uses the Hungarian algorithm to track vehicle movement, calling the vehicle coordinates and visual pickup device positions from the motion information set, and binding the qualified visual pickup device with the vehicle information so that the qualified visual pickup device always tracks the vehicle's movement.

[0030] In the process of selecting suitable vision pickup devices, after the above calculations, the two most suitable vision pickup devices are retained to track the same vehicle, thereby providing a more comprehensive tracking of vehicle movement.

[0031] In any embodiment of this application, when a visual pickup device has multiple cameras, the number of vehicles it can track simultaneously is constrained based on the number of cameras it has. For example, when a visual pickup device has 3 cameras, its maximum tracking number is set to 3, with each camera independently tracking one vehicle.

[0032] When the focus command output to a single vision pickup device exceeds the number of its cameras, the following judgment is made: Are there still any idle vision pickup devices that can observe the dynamic vehicle and are in an idle state? If so, invoke the vision pickup device. If not, generate an early warning message and output it to the mobile terminal.

[0033] When multiple qualified vision pickup devices are competing for a vehicle target, each qualified vision pickup device is evaluated based on the following formula, and only the two with the highest scores are bound together.

[0034] The formula is: score = 0.4*(1 / distance) + 0.3*(1 / occlusion rate) + 0.3*(1 / deflection angle).

[0035] In the collision risk analysis step 104, at least three feature points on the outer contour of the vehicle are identified in the motion information set. The minimum Euclidean distance between the feature points and surrounding entities is calculated based on the local attention mechanism. In response to the collision warning threshold input from the outside, collision warning information is output when the threshold is reached.

[0036] The feature points are the most prominent or representative points of the vehicle's outline, such as the front corner, rear corner, and midpoint of the side door.

[0037] In any embodiment of this application, when calculating the Euclidean distance between the feature point and the surrounding entities, the weights are also adjusted by dynamic weighting. For example, in one embodiment, the weight of the front corner is assigned as 0.6, and the weights of the others are all 0.2 to better adapt to the actual situation that driving collisions mostly occur at the front of the vehicle.

[0038] In any embodiment of this application, the collision risk analysis step further introduces an AI spatiotemporal token mechanism, which calls the historical trajectory of the vehicle's movement in the motion information set, calculates and outputs linear trajectory prediction data for the next five seconds, and predicts the possibility of the vehicle colliding with surrounding entities based on the linear trajectory prediction data.

[0039] Specifically, when the distance between a vehicle feature point and surrounding entities in the trajectory linear prediction data is less than two meters and the relative speed is greater than 5 km / h, it is identified as a potential collision and a collision warning is output.

[0040] Collision warning step 105: In response to the collision warning information, call the directional acoustic array and / or in-ground lights to issue a collision warning.

[0041] The directional acoustic array can send out warning information to vehicles that are about to collide. At the same time, the collision warning step filters the in-ground lights located in the visible range in front of the vehicle that is about to collide and simultaneously reminds the vehicle of the collision.

[0042] In one embodiment of this application, the collision warning step also performs an evidence consolidation action. When a collision is predicted, a total of 10 minutes of video clips are automatically captured to facilitate the operator's review of the evidence. When an actual collision is detected, a total of 15 minutes of clips before and after the collision are automatically captured to facilitate the operator's review of the evidence.

[0043] Through the above technical solution, when a vehicle is driving in a parking lot, its trajectory can be effectively calculated, and the driver can be promptly reminded to avoid unnecessary collisions. Especially when multiple vehicles are moving at the same time, because each vehicle has a focused camera, the collision warning analysis can be performed with that vehicle as the main focus, thereby achieving multiple protections. Combined with the high speed and dynamic blur characteristics of AI algorithms, it can quickly predict and analyze complex situations, thereby effectively predicting and reminding drivers of situations that are prone to collisions, such as passing other vehicles, traveling in opposite directions, and passing other vehicles at corners.

[0044] In any embodiment of this application, a blind spot screening step 106 is also included, which calls all visual pickup devices to move within all their movement ranges to form blind spot identification data, calls the site topographic map to compare with the blind spot identification data, identifies areas that cannot be covered by the visual range of the visual pickup devices, and outputs a blind spot set.

[0045] The blind spot screening step automatically executes a dynamic scanning protocol, performing a scan and updating the blind spot identification data and blind spot set every 30 minutes.

[0046] The blind spot screening step also includes: continuously retrieving static information sets and motion information sets, simultaneously incorporating the above blind spot comparison actions, identifying changes in areas that cannot be covered by the visual range of the visual pickup device, and outputting a dynamic blind spot set.

[0047] In any embodiment of this application, the blind spot screening step 106 also periodically calls the blind spot set and the dynamic blind spot set, and calls the lidar near the corresponding blind spot to scan the blind spot location, assisting the visual pickup device in filling in the blind spot location information. Since the image generated by the lidar is point cloud array data, the ICP point cloud registration algorithm is used to fuse it with the visual data, thereby achieving the updating of the complete site topographic map and the elimination of blind spots.

[0048] In any embodiment of this application, during the process of collecting dynamic blind spot sets, Kalman filtering is invoked to smooth the changes in the position of dynamic blind spots, thereby forming a dynamic blind spot change presentation that is more in line with visual habits.

[0049] In any embodiment of this application, a path guidance step 107 is also included, which calls the station topographic map and static information set and motion information set, analyzes the motion intention of the moving vehicles and the available parking spaces, calls the ant colony optimization algorithm, calculates the multi-vehicle path, and guides the vehicles to the available parking spaces.

[0050] The path guidance step includes building an intent recognition model. During the building process, the LSTM+ spatiotemporal attention mechanism is called to input information such as the vehicle's speed, frequency of method changes, and relative distance changes from the exit into the model. This distinguishes and determines the vehicle's intent to find a parking space and its intent to leave, and plans its route based on the determination results.

[0051] In the process of route planning, this application adopts ant colony optimization to achieve path planning, and the core formula is: Δτ = (1-ρ)·τ + Σ(Q / L_k) in: Δτ is the updated value of the pheromone concentration, i.e., the concentration of the new generation of pheromones. It is used to cover the pheromones on the original planned path, so as to realize the dynamic adjustment of the path quality.

[0052] In (1-ρ)·τ, ρ is the evaporation value, used to simulate the natural evaporation of pheromones and avoid the algorithm becoming too noisy and familiar with local optima. In any embodiment of this application, the evaporation value represents the state in which existing paths are diluted in the algorithm. The typical value of the evaporation value is 0.3-0.7, and the larger the value, the stronger the ability to explore new paths.

[0053] τ represents the pheromone concentration on the current path.

[0054] (1-ρ)·τ represents the (1-ρ) proportion of the original pheromones retained, enabling the operation of the forgetting mechanism.

[0055] In Σ(Q / L_k), Q represents the pheromone intensity function, which is used to control the scaling factor of the pheromone increment. Its adjustment trend is positively correlated with the problem size. For example, in any embodiment of this application, the larger the parking lot and the more complex the path, the larger Q should be.

[0056] L_k represents the path length of ant k (i.e., the moving car in any embodiment of this application). The shorter the path, the greater the pheromone increment.

[0057] Σ is a contribution summation formula used to optimize the path, acquire more information enhancements, and form a positive feedback loop.

[0058] In summary, the ant colony model can achieve path enhancement, obtaining higher pheromone increments for short paths and the opposite for long paths, thus adapting to parking and departure analyses of different distances.

[0059] At the same time, the existence of volatile values ​​can prevent the algorithm from getting stuck in local optima, making vehicle management tend towards the global optimal path.

[0060] Furthermore, in any embodiment of this application, when the path obtained by the ant colony model is output, a smooth path that avoids intersections is output to further reduce the occurrence of vehicles passing each other, thereby improving the smoothness and safety of vehicle driving monitoring and control.

[0061] The present invention also proposes a smart depot driving monitoring device, which is used to realize the function of the visual recognition device in the above-mentioned smart depot driving monitoring method.

[0062] like Figure 2-4 As shown, the intelligent station driving monitoring device disclosed in any embodiment of this application includes a base 320, a base column 310 fixedly connected to the base, and three cameras 301 rotatably connected to the base column. All cameras are arranged along the length of the base column. The number of cameras can be adjusted to one or more as needed.

[0063] like Figure 2 As shown in one embodiment of this application, each camera 301 can rotate 120 degrees in the horizontal plane.

[0064] like Figure 3-5 As shown, in another embodiment of this application, each camera 301 can rotate 360 ​​degrees around the base column. A gear ring is provided inside the base column at the position corresponding to each camera, and each camera meshes with the corresponding gear ring and can rotate around the gear ring under the drive of a drive motor.

[0065] By integrating multiple cameras into one device through the aforementioned driving monitoring device 300, the related functions of the visual pickup device mentioned in the driving monitoring method can be realized. This can save control costs and enable multiple cameras to freely track moving vehicles.

[0066] like Figure 3-5 As shown, a directional acoustic array 303 is also rotatably connected to the base column. The directional acoustic array is located at the end of the base column corresponding to the camera 301 away from the base. The directional acoustic array 303 can achieve the function of rotating around the base column 360 degrees by meshing with the toothed ring at the corresponding position in the base column.

[0067] When it is necessary to issue an audible alert to a designated vehicle, the directional acoustic matrix 302 can be rotated to the same position as the camera 301 tracking the vehicle and a corresponding alert can be issued.

[0068] like Figure 3-5 As shown, a lidar 302 is also rotatably connected to the base column. The lidar is set at the end of the base column away from the camera. By setting it at a different height from the camera, it can supplement the camera's field of view, thereby realizing the corresponding function of the above-mentioned driving monitoring method.

[0069] The base is equipped with a central storage unit that stores a dual-stream network model. All cameras, directional sound arrays, and lidars are connected to the central storage unit. The central storage unit writes the dual-stream network model into the cameras and lidars.

[0070] like Figure 6 The image shows a parking lot station 401 using the aforementioned driving monitoring device 300. By installing the driving monitoring device 300 between parking spaces, the driving of vehicles in the parking lot station can be monitored by calling multiple cameras and lidar.

[0071] like Figure 6 As shown, when using the driving monitoring method disclosed in this application, the driver can also be prompted or warned by the ground-mounted in-ground light 305.

[0072] This invention also proposes a smart depot operation monitoring system, such as... Figure 2 As shown, the system includes the aforementioned intelligent depot driving monitoring equipment and method. The specific structure of the intelligent depot driving monitoring method and equipment is as described in the above embodiments. Since this system adopts all the technical solutions of all the above embodiments, it possesses at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be elaborated upon here. It includes a perception layer, a decision layer, an execution layer, and a hardware layer, wherein: Hardware layer 202 includes various hardware for performing information acquisition activities, including visual pickup devices, directional acoustic arrays, in-ground lights, lidar, mobile terminals, etc. The visual pickup devices are installed on the walls, ceilings, and tops of support columns of the site. The directional acoustic arrays are installed in the same or similar positions as the visual pickup devices. The in-ground lights are installed under the driving roads of the parking lot. The lidar is installed in the same or similar positions as the visual pickup devices and on the side walls of the support columns.

[0073] The perception layer 201 calls the data from the hardware layer to perform the station identification step, vehicle identification step, and blind spot screening step.

[0074] The decision layer 203 calls the data output of the perception layer to execute various operations in the motion tracking step, path guidance step, and collision risk analysis step.

[0075] Execution layer 204 calls the data output of the decision layer to perform the collision warning step.

[0076] The above system enables system-level method implementation, allowing for effective monitoring and management of station operations through the electronic control system.

[0077] like Figure 7-8As shown, in one embodiment of this application, a smart parking station 402 using a smart parking station driving monitoring method and system is disclosed. In this system, support columns 410 for supporting various monitoring devices are provided between parking spaces. A lidar 302 is installed on the side wall of the support column 410. The camera 301 used by the visual pickup device is a common high-definition camera, located on the top of the support column. An in-ground light 304 is installed on the driving road of the parking station.

[0078] The above description is merely an optional embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for intelligent station operation monitoring, characterized in that, include: The site identification process involves calling a visual pickup device to scan environmental information, generating a site topographic map, and then creating three-dimensional coordinates of parking spaces on the site topographic map. The vehicle recognition process involves identifying all vehicle information within the range of the visual acquisition device, forming three-dimensional coordinates of the vehicles, marking the positions and outlines of stationary vehicles to form a static information set, and marking moving vehicles to form a motion information set. The motion tracking steps involve responding to a set of motion information and calling upon a visual pickup device to track vehicle movement based on a Hungarian algorithm. The collision risk analysis steps involve identifying at least three feature points on the outer contour of the vehicle with concentrated motion information, calculating the minimum Euclidean distance between the feature points and surrounding entities based on a local attention mechanism, responding to external input collision warning thresholds, and outputting collision warning information when the threshold is reached. The collision warning procedure involves responding to the collision warning information by calling a directional acoustic array and / or in-ground lights to issue a collision warning. The collision risk step involves invoking a spatiotemporal token mechanism to predict the vehicle's path for the next 5 seconds through a linear transformation of historical trajectories.

2. The driving monitoring method as described in claim 1, characterized in that, The motion tracking steps also include: responding to the motion information set, marking the dynamic vehicle, and calling on at least two visual pickup devices that are closest to it to output focusing instructions to track the dynamic vehicle; Each vision pickup device matches the maximum number of vehicles it tracks based on the number of its cameras, and the focusing command output to each vision pickup device does not exceed the number of its cameras.

3. The driving monitoring method as described in claim 2, characterized in that, When the focus command output to a single vision pickup device exceeds the number of its cameras, the following judgment is made: Are there still available visual acquisition devices that can observe the dynamic vehicle and are in an idle state? If so, invoke the vision pickup device. If not, generate an early warning message and output it to the mobile terminal.

4. The driving monitoring method as described in claim 1, characterized in that, It also includes a blind spot screening step, which calls all visual pickup devices to move within their entire range of movement to form blind spot identification data, calls the site topographic map to compare with the blind spot identification data, identifies areas that cannot be covered by the visual range of the visual pickup devices, and outputs a set of blind spots.

5. The driving monitoring method as described in claim 4, characterized in that, The blind spot screening step also includes: continuously retrieving static information sets and motion information sets, simultaneously incorporating comparison actions, identifying changes in areas that cannot be covered by the visual range of the visual pickup device, and outputting a dynamic blind spot set.

6. The driving monitoring method as described in claim 5, characterized in that, The blind spot screening step also includes: responding to the dynamic blind spot set, calling up the lidar, and assisting the visual pickup device in filling in the blind spot location information.

7. The driving monitoring method as described in any one of claims 1-6, characterized in that, A two-stream network model is embedded in all the aforementioned visual pickup devices.

8. The driving monitoring method as described in any one of claims 1-6, characterized in that, It also includes a path guidance step, which calls the station topographic map and static information set and motion information set, analyzes the motion intention of moving vehicles and available parking spaces, calls the ant colony optimization algorithm, calculates multi-vehicle paths, and guides vehicles to available parking spaces.

9. A smart depot driving monitoring device, integrated with the driving monitoring method as described in any one of claims 1-8, characterized in that, include: Base; At least two cameras are mounted on a base, all of which are arranged in a direction perpendicular to the base, and the cameras are rotatable within a horizontal plane at least 120 degrees. A directional sound array is mounted on a base and rotates 360 degrees in the horizontal plane. The lidar is mounted on a base and rotates 360 degrees in the horizontal plane. The central storage unit stores a dual-stream network model, and all cameras and / or directional sound arrays and / or lidars are signal-connected to the central storage unit.

10. A smart depot driving monitoring system, comprising the driving monitoring method as described in any one of claims 1-8 and / or the driving monitoring equipment as described in claim 9, characterized in that, include: The hardware layer includes various types of hardware used to perform information acquisition activities, including visual pickup devices and / or directional acoustic arrays and / or lidar; The perception layer calls data from the hardware layer to perform operations such as site identification, vehicle identification, and blind spot screening. The decision-making layer calls upon the data output from the perception layer to execute various operations in the motion tracking, path guidance, and collision risk analysis steps. The execution layer calls the data output from the decision layer and executes the collision warning steps.

Images