Automatic parking method and system based on CSI, terminal and storage medium

Abstract

The invention provides an automatic parking method and system based on CSI, a terminal and a storage medium, and the method comprises the steps: carrying out the matching of a relative position relation between a vehicle and a parking space based on a first preset algorithm, and obtaining vehicle attitude data; sensing the antenna matrix based on the wifi-CSI, and obtaining CSI sensing data; based on the CSI sensing data and the vehicle attitude data, carrying out transfer learning on the parking space pre-built model through a second preset algorithm to obtain a reasoning sensing model; on the basis of the inference perception model, performing pose evaluation and time sequence prediction on the vehicle to obtain a space-time pose and a prediction track of the vehicle; and controlling the vehicle to be parked in the parking space based on the space-time pose and the predicted trajectory of the vehicle. The problem that in the prior art, due to interference of the environment or a closed scene, the accuracy of whole vehicle direction control is not high in the parking process is solved. Interference of the environment and a closed scene is weakened, and the accuracy of whole vehicle direction control in the parking process is improved.

Description

Technical Field

[0001] The present invention relates to the field of automobile software and implementation, and in particular to a CSI-based automatic parking method, system, terminal and storage medium. Background Art

[0002] Current Internet of Vehicles technologies mainly include several core solutions: one is the technology that combines 5G LTE, surround-view camera mapping, ultrasonic radar and odometer positioning; the second is the technology that integrates 5G LTE, lidar mapping, millimeter-wave radar and odometer positioning; the third is the technology that uses 5G LTE and pure lidar for spatiotemporal perception and positioning; this can be achieved through the parking lot space perception collaborative technology and the use of the Internet of Things to collaborate with the on-board perception unit to assist the vehicle in completing automatic parking in the Internet of Vehicles APA parking environment.

[0003] At present, the existing technology mainly uses computer vision recognition technology, such as surround-view high-definition cameras and multi-angle high-definition fisheye cameras to collect video images of parking spaces, and then uses algorithms to identify ground markings and obtain other obstacles and vehicle information in the parking lot with the help of 5G LTE, and then uses vision and the on-board inertial navigation system to correct the direction control of the entire vehicle in real time; however, the existing technology is affected by the environment or closed scenes. For example, the recognition rate is unstable in special scenes such as low light, dark light, and bright and dark light. Moreover, in closed environments or environments with many obstacles, the existing technology is prone to signal instability and signal interference. Therefore, it is easy to cause low accuracy in the direction control of the entire vehicle during parking. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes an automatic parking method, system, terminal and storage medium based on CSI, which can reduce the interference of the environment and closed scenes and improve the accuracy of the vehicle direction control during the parking process.

[0005] To achieve the above object, an embodiment of the present invention provides an automatic parking method based on CSI, comprising:

[0006] Based on a first preset algorithm, the relative position relationship between the vehicle and the berth is matched to obtain vehicle posture data;

[0007] Based on the wifi-CSI sensing antenna matrix, obtain CSI sensing data;

[0008] Based on the CSI perception data and the vehicle posture data, transfer learning is performed on the pre-built parking space model through a second preset algorithm to obtain an inference perception model;

[0009] Based on the inference perception model, the vehicle's position and posture are evaluated and time series predicted to obtain the vehicle's spatiotemporal position and predicted trajectory;

[0010] Based on the spatiotemporal position and predicted trajectory of the vehicle, the vehicle is controlled to park in a parking space.

[0011] Furthermore, the relative position relationship between the vehicle and the berth is matched based on the first preset algorithm to obtain vehicle posture data, including:

[0012] Acquire a vehicle image and a parking space image;

[0013] Preprocessing the vehicle image and the parking space image to extract vehicle feature points and parking space feature points;

[0014] Based on the first preset algorithm, the relative position relationship between the vehicle feature points and the feature points of the parking space to be parked is matched to obtain vehicle posture data.

[0015] Further, the obtaining of CSI sensing data based on the wifi-CSI sensing antenna matrix includes:

[0016] Constructing a flat panel antenna group based on a first preset group of transmitting antennas and a second preset group of receiving antennas;

[0017] Based on the flat antenna group, a wifi-CSI sensing antenna matrix is ​​constructed by orthogonally deploying a third preset group of receiving antennas;

[0018] Based on the wifi-CSI sensing antenna matrix, obtaining CSI data of each receiving antenna;

[0019] Based on the second preset algorithm, the CSI data of each receiving antenna is quotiented and signal transformed to obtain CSI perception data.

[0020] Furthermore, based on the CSI perception data and the vehicle posture data, transfer learning is performed on the pre-built parking space model to obtain an inference perception model, including:

[0021] Perform modal fusion encoding on CSI perception data and vehicle posture data to construct a comprehensive feature vector;

[0022] Performing feature extraction on the comprehensive feature vector to obtain CSI tensor parameters;

[0023] Based on the CSI tensor parameters, transfer learning is performed on the pre-built parking space model to obtain an inference perception model.

[0024] Furthermore, based on the inference perception model, the vehicle posture evaluation and time series prediction are performed to obtain the vehicle's spatiotemporal posture and predicted trajectory, including:

[0025] Based on the preset CSI sampling frequency, the vehicle CSI perception data is sorted to obtain CSI information of several time segments;

[0026] Combine CSI information of several time segments into continuous CSI time series information;

[0027] The continuous CSI time series information is input into the reasoning perception model, and the vehicle posture evaluation and time series prediction are performed to obtain the vehicle's spatiotemporal posture and predicted trajectory.

[0028] Furthermore, based on the spatiotemporal position and predicted trajectory of the vehicle, controlling the vehicle to park in a parking space includes:

[0029] Based on the vehicle's spatiotemporal position and predicted trajectory, the vehicle's driving boundary is obtained;

[0030] Based on the vehicle driving boundary, a vehicle control signal is obtained;

[0031] The vehicle is parked in the parking space by means of vehicle control signals.

[0032] The embodiment of the present invention further provides a CSI-based automatic parking system, comprising: a position acquisition module, a CSI perception data acquisition module, a reasoning perception model construction module, a reasoning prediction module and a vehicle control module;

[0033] The position acquisition module is used to match the relative position relationship between the vehicle and the berth based on a first preset algorithm to obtain vehicle posture data;

[0034] The CSI sensing data acquisition module is used to acquire CSI sensing data based on the wifi-CSI sensing antenna matrix;

[0035] The inference perception model building module is used to perform transfer learning on the pre-built parking space model through a second preset algorithm based on the CSI perception data and the vehicle posture data to obtain an inference perception model;

[0036] The inference prediction module is used to perform posture evaluation and time series prediction on the vehicle based on the inference perception model to obtain the spatiotemporal posture and predicted trajectory of the vehicle;

[0037] The vehicle control module is used to control the vehicle to park in a parking space based on the spatiotemporal position and predicted trajectory of the vehicle.

[0038] Further, the CSI sensing data acquisition module is used to acquire CSI sensing data based on the wifi-CSI sensing antenna matrix, including:

[0039] A first antenna group construction module, a second antenna group construction module, a CSI data receiving module and a CSI data processing module;

[0040] The first antenna group building module is used to build a flat antenna group based on a first preset group of transmitting antennas and a second preset group of receiving antennas;

[0041] The second antenna group construction module is used to construct a wifi-CSI sensing antenna matrix based on the flat antenna group by orthogonally deploying a third preset group of receiving antennas;

[0042] The CSI data receiving module is used to obtain CSI data of each receiving antenna based on the wifi-CSI sensing antenna matrix;

[0043] The CSI data processing module is used to perform quotient and signal transformation on the CSI data of each receiving antenna based on a second preset algorithm to obtain CSI perception data.

[0044] A CSI-based automatic parking terminal includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the CSI-based automatic parking method is implemented.

[0045] A CSI-based automatic parking storage medium, the storage medium comprising: a computer-readable storage medium and a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the CSI-based automatic parking method.

[0046] Beneficial effects:

[0047] (1) By matching the relative position of the vehicle and the parking space, the vehicle's posture data is obtained to provide an accurate data basis for subsequent perception and reasoning. By flexibly deploying the Wi-Fi-CSI sensing antenna matrix, it does not occupy resources inside the vehicle, enhances the penetration and diffraction capabilities of obstacles, and reduces the interference of complex occlusion scenes and environments on vehicle control. It can output richer tensor parameter information to improve the accuracy of vehicle direction control during parking.

[0048] (2) Through transfer learning of the pre-built parking space model, the model can quickly adapt to complex and changing environments and improve the response speed of the model. Then, an inference perception model is constructed to infer and predict the spatiotemporal position and predicted trajectory of the vehicle. Through position evaluation and time series prediction, the model's generalization ability for vehicle recognition is improved, thereby reducing the impact of multipath interference images and improving the accuracy of vehicle direction control during parking. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic flow chart of a CSI-based automatic parking method according to an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of the logical structure of an automatic parking method based on CSI provided in an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of an antenna arrangement structure of an automatic parking method based on CSI provided in an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of a sequential cloud end-to-end perception model of a CSI-based automatic parking method provided in an embodiment of the present invention;

[0053] Figure 5 A schematic diagram of vehicle-road-cloud three-terminal perception decision signal transmission of a CSI-based automatic parking method provided in an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of the module structure of a CSI-based automatic parking system provided in an embodiment of the present invention;

[0055] Figure 7 A schematic diagram of the structure of a CSI perception data acquisition module of a CSI-based automatic parking system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 , Figure 1 A schematic diagram of a CSI-based automatic parking method according to an embodiment of the present invention. Figure 2 , Figure 2 A logical structure diagram of an automatic parking method based on CSI provided in an embodiment of the present invention; Figure 2 As shown, in this embodiment, the logic of interaction between the vehicle side, the cloud side and the parking lot side will be explained, and will not be repeated below.

[0059] like Figure 1 As shown, the CSI-based automatic parking method includes steps 101 to 105, each of which is specifically as follows:

[0060] Step 101, based on a first preset algorithm, matching the relative position relationship between the vehicle and the berth to obtain vehicle posture data;

[0061] As an example of this embodiment, a vehicle image and an image of a parking space to be parked are acquired; the vehicle image and the image of the parking space to be parked are preprocessed to extract vehicle feature points and feature points of the parking space to be parked; based on a first preset algorithm, relative position relationship matching is performed on the vehicle feature points and feature points of the parking space to be parked to obtain vehicle posture data.

[0062] A specific implementation method is to obtain the relative position relationship between the vehicle and the parking space through a cloud camera CV (computer vision), install multiple cameras at different locations in the parking lot to obtain multiple perspective images of the vehicle and the parking space, and fuse the images of multiple perspectives through image registration and fusion algorithms to form a complete three-dimensional model of the vehicle and the parking space. Using the three-dimensional model, feature points (such as corner points and edge points, etc.) are extracted on the vehicle and the parking space; the feature points on the vehicle are matched with the feature points on the parking space using a feature point matching algorithm (equivalent to the first preset algorithm, such as SIFT and SURF, etc.); the relative position relationship between the vehicle and the parking space, as well as the posture information of the vehicle can be obtained through the matching results, and the corresponding information data after processing is sent to the computing terminal.

[0063] Step 102, acquiring CSI sensing data based on the wifi-CSI sensing antenna matrix;

[0064] As an example of this embodiment, based on the first preset group of transmitting antennas and the second preset group of receiving antennas, a flat panel antenna group is constructed; based on the flat panel antenna group, a wifi-CSI sensing antenna matrix is ​​constructed by orthogonally deploying a third preset group of receiving antennas; based on the wifi-CSI sensing antenna matrix, the CSI data of each receiving antenna is obtained; based on the second preset algorithm, the CSI data of each receiving antenna is divided and signal transformed to obtain CSI sensing data.

[0065] A specific implementation method is to set up an antenna group at the parking lot end, see Figure 3 , Figure 3 A schematic diagram of an antenna arrangement structure of an automatic parking method based on CSI is provided in an embodiment of the present invention, such as Figure 3As shown, the embodiment of the present invention uses a flat panel antenna (equivalent to a flat panel antenna group) arranged in a place and placed side by side with one transmitter (TX, equivalent to the first preset group of transmitting antennas) and two receivers (RX1 and RX2, equivalent to the second preset group of receiving antennas), plus an independent receiving antenna (RX3, equivalent to the third preset group of receiving antennas) deployed orthogonally to the one transmitter and two receiver antenna groups to form a wifi-CSI sensing antenna matrix; wherein the one transmitter and two receiver antenna group design combined with the ResNet deep learning algorithm can effectively reduce multipath interference and improve detection accuracy, and then combined with RX3, it can obtain wifi-CSI return signals within a larger range for correction, reducing the cumulative error of the wifi radar sensing solution in the process of identifying spatial posture. At the same time, the linear matrix design proposed in this embodiment makes full use of the obstacle diffraction and wall penetration capabilities of the wifi signal, which can reduce the interference of the echo path obstruction and small and medium-sized radio frequency shielding objects on the wifi signal, so that the signal processing module can output richer tensor parameter information.

[0066] As another example of this embodiment, after the wifi-CSI sensing antenna matrix is ​​deployed, a data acquisition model based on multi-channel CSI quotient is proposed, the signal-to-noise ratio is improved by symmetrically arranged receiving antennas (RX1 and RX2), and the quotient is obtained by using the same transmission path, the same wavelength ratio, the same antenna, the receiving end hardware design and the symmetrical layout, so as to eliminate the interference such as amplitude noise caused by the communication path and hardware. At the same time, in order to deal with the poor performance of symmetrical antennas in the simultaneous identification of multiple vehicles and the cumulative displacement error of vehicles during movement, the embodiment of the present invention uses orthogonally arranged antennas (RX3) The received CSI information is multi-pathed with the received information of RX1 and RX2 to eliminate the random offset and perceived distance and accumulated spatial position errors generated by the phase information under multiple vehicle targets, and then the orthogonality of the CSI signal amplitude and phase is restored through the signal conditioning unit, and then sent to the cloud by the router for calculation to eliminate multipath interference, etc. Specifically, the orthogonality of the signal amplitude and phase can be effectively restored through multi-path quotient. Taking a reflection path as an example, the reflection path difference between each two antenna receiving modules △d=d2(t)-d1(t). In the embodiment of the present invention, the CSI quotient is required to calculate the spatiotemporal positioning of the parked car, and the calculation formula is as follows:

[0067] H ratio =CSI RX1 / CSI RX2 =H d,1 (f,t) / H d,2 (f,t)

[0068] The position perception is then completed through the CSI quotient after Mobius transformation. As mentioned above, this method combined with the numerical processing of the jump points of the phase signal and the amplitude signal can restore the orthogonality of the CSI signal amplitude and phase to eliminate interference and make the CSI quotient signal clear and usable.

[0069] It is worth mentioning that the embodiment of the present invention also proposes a parking lot mapping perception and positioning solution that uses commercial wifi devices instead of laser radar combined with cameras, and realizes a method of receiving and capturing various CSI information sent by the transmitter through multiple wifi terminals, and designs a set of data processing programs that are pre-written into wifi terminal devices (including but not limited to Qualcomm wifi chip QCA 9X, Broadcom wifi chip and ESP32wifi chip, etc.), which can realize the collection of CSI information in duplex and half-duplex wifi terminals, thereby obtaining computer equipment for real-time posture information of vehicles. The wifi device can complete the perception task with high precision, assist the cloud computing module to complete the vehicle APA wizard, and solve the problem of dense deployment of cameras and laser radars caused by the occlusion of the complex environment of the parking lot. At the same time, the introduction of commercial wifi nodes greatly saves the design and deployment costs and terminal computing power requirements.

[0070] Step 103, based on the CSI perception data and the vehicle posture data, transfer learning is performed on the pre-built parking space model through a second preset algorithm to obtain an inference perception model;

[0071] As an example of this embodiment, modal fusion encoding is performed on CSI perception data and vehicle posture data to construct a comprehensive feature vector; feature extraction is performed on the comprehensive feature vector to obtain CSI tensor parameters; based on the CSI tensor parameters, transfer learning is performed on the pre-built parking space model to obtain an inference perception model.

[0072] For a specific implementation method, see Figure 4 , Figure 4 A schematic diagram of a sequential cloud end-to-end perception model of a CSI-based automatic parking method provided in an embodiment of the present invention; Figure 4As shown, the embodiment of the present invention proposes a one-hot encoding method for a driving CSI multimodal fusion network, that is, through a Latin hypercube DOE test combination design, the CSI information including amplitude, phase, multipath characteristics and channel characteristics is subjected to an end-to-end model output result correlation analysis, and key CSI information parameters corresponding to general parking scenes and special scenes are found (for example, in general scenes, only amplitude phase and multipath characteristics can meet the requirements of contour and obstacle recognition) for modal fusion network one-hot encoding; this method can improve the effectiveness of feature extraction after encoding and save memory resources for encoding, decoding and feature extraction operations; then, after being processed by the above-mentioned one-hot encoding method for a driving CSI multimodal fusion network, a comprehensive feature vector containing vehicle position, posture, speed and surrounding environment information is formed, and the extractor features are used to obtain CSI tensor parameters, and multiple groups of vision, radar, GPS and IMU are combined with w in a unified time slice (dt). ifi-CSI migrates the multimodal recognition perception model (equivalent to the pre-built parking space model) pre-trained according to the spatial layout of the parking lot to the end-to-end model of wifi-CSI perception (equivalent to the inference perception model), and obtains the end-to-end model of wifi-CSI perception. When training the model, the sensor signal tensor parameters are input, and the migration loss function of the model and the contour information of the identified object are output. The model adds the same time (dt) dimension and then encodes it and then puts it into the pre-trained end-to-end perception model. The model consists of one-hot encoding, feature extraction, decoding, acquisition of vehicle posture hotspot matrix, convolutional neural network ResNet, ROI pooling, complementarity of posture and contour information source information (for example: when the parking speed is fast, the posture is generally more accurate than the specific contour information, and the contour information is corrected by the posture) and correction of the output control end language. The perception prediction performed in this way can greatly enhance the generalization performance of the model in identifying vehicle size types and effectively reduce multipath interference images.

[0073] Step 104, based on the inference perception model, perform posture evaluation and time series prediction on the vehicle to obtain the spatiotemporal posture and predicted trajectory of the vehicle;

[0074] As an example of this embodiment, based on the preset CSI sampling frequency, the vehicle CSI perception data is sorted to obtain CSI information of several time segments; the CSI information of several time segments is merged into continuous CSI time series information; the continuous CSI time series information is input into the inference perception model, and the vehicle's posture evaluation and time series prediction are performed to obtain the vehicle's spatiotemporal posture and predicted trajectory.

[0075] A specific implementation method is to use the routing unit of the parking lot to organize the signal-conditioned CSI information into multiple continuous time segments, upload them to the cloud for perception and reasoning, and sort the data packets through the data upload channel according to the CSI sampling frequency of the terminal device. Taking ESP32 as an example, the sampling frequency is 100 frames / second, corresponding to the camera video 25 frames / second (that is, each video frame corresponds to 4 frames of CSI tensors). The complete motion trajectory and predicted path of each vehicle can be calculated, and multi-vehicle recognition can be achieved. In cloud data processing, the CSI packets of multiple time segments will be merged into continuous CSI time series information, and the amplitude, phase and other parameters in the CSI data will be added with time dimensions, converted into a unified tensor parameter scale through encoding and decoding, and then input into the inference model to complete the inference perception.

[0076] Step 105: Based on the spatiotemporal position and predicted trajectory of the vehicle, control the vehicle to park in a parking space.

[0077] As an example of this embodiment, based on the spatiotemporal position and predicted trajectory of the vehicle, a vehicle driving boundary is obtained; based on the vehicle driving boundary, a vehicle control signal is obtained; and the vehicle is controlled to park in a parking space through the vehicle control signal.

[0078] For a specific implementation method, see Figure 5 , Figure 5 A schematic diagram of a vehicle-road-cloud three-terminal perception decision signal transmission of a CSI-based automatic parking method provided in an embodiment of the present invention; Figure 5 As described above, through the above steps 101 to 104, the parking lot end processes the CSI signal and the visual signal, and the cloud end respectively acquires the spatial position signal, models the vehicle's dynamic posture, and constructs a map of the key points in advance. Finally, the vehicle end processes the GPS position signal, processes the IMU posture signal, and constructs a real-time map of the vision / radar. After verification with the cloud end, a control decision signal for determining the spatiotemporal posture of the vehicle is obtained. Based on the spatiotemporal posture and predicted trajectory of the vehicle, combined with the control decision signal of the spatiotemporal posture of the vehicle, a driving boundary is planned that enables the vehicle to park smoothly and safely in the parking space. This includes two aspects: path planning and speed planning. Path planning determines the spatial driving route of the vehicle, and speed planning determines the driving speed of the vehicle at different positions. According to the planned trajectory, the vehicle control system generates corresponding control signals, including steering signals, throttle and brake signals, etc., to finally control the vehicle to park in the parking space.

[0079] An automatic parking method based on CSI proposed in an embodiment of the present invention obtains the posture data of the vehicle by matching the relative position relationship between the vehicle and the parking space, thereby providing an accurate data basis for subsequent perception reasoning. By flexibly deploying the wifi-CSI perception antenna matrix, it does not occupy the resources in the vehicle, enhances the penetration and diffraction capabilities of obstacles, reduces the interference of complex occlusion scenes and environments on vehicle control, and can output richer tensor parameter information to improve the accuracy of the vehicle direction control during parking. By performing transfer learning on the pre-built parking space model, the model can quickly adapt to the complex and changeable environment, improve the response speed of the model, and then construct an inference perception model to infer and predict the spatiotemporal posture and predicted trajectory of the vehicle. Through posture evaluation and time series prediction, the generalization ability of the model for vehicle recognition is improved, thereby reducing the influence of multipath interference images and improving the accuracy of the vehicle direction control during parking.

[0080] Example 2

[0081] See also Figure 6 , Figure 6 A schematic diagram of the module structure of an automatic parking system based on CSI is provided in one embodiment of the present invention. Figure 6 As shown, the embodiment of the present invention proposes an automatic parking system based on CSI, including: a position acquisition module 601, a CSI perception data acquisition module 602, a reasoning perception model construction module 603, a reasoning prediction module 604 and a vehicle control module 605;

[0082] The position acquisition module 601 is used to match the relative position relationship between the vehicle and the berth based on a first preset algorithm to obtain vehicle posture data;

[0083] The CSI sensing data acquisition module 602 is used to acquire CSI sensing data based on the wifi-CSI sensing antenna matrix;

[0084] As an example of this embodiment, see Figure 7 , Figure 7 A schematic diagram of a CSI sensing data acquisition module structure of a CSI-based automatic parking system provided in an embodiment of the present invention. Figure 7 As shown, the CSI sensing data acquisition module 602 is used to obtain CSI sensing data based on the wifi-CSI sensing antenna matrix, including:

[0085] A first antenna group building module 701, a second antenna group building module 702, a CSI data receiving module 703 and a CSI data processing module 704;

[0086] The first antenna group building module 701 is used to build a flat antenna group based on a first preset group of transmitting antennas and a second preset group of receiving antennas;

[0087] The second antenna group construction module 702 is used to construct a wifi-CSI sensing antenna matrix based on the flat antenna group by orthogonally deploying a third preset group of receiving antennas;

[0088] The CSI data receiving module 703 is used to obtain CSI data of each receiving antenna based on the wifi-CSI sensing antenna matrix;

[0089] The CSI data processing module 704 is used to perform quotient and signal transformation on the CSI data of each receiving antenna based on a second preset algorithm to obtain CSI perception data.

[0090] The inference perception model building module 603 is used to perform transfer learning on the pre-built parking space model through a second preset algorithm based on the CSI perception data and the vehicle posture data to obtain an inference perception model;

[0091] The inference prediction module 604 is used to perform posture evaluation and time series prediction on the vehicle based on the inference perception model to obtain the spatiotemporal posture and predicted trajectory of the vehicle;

[0092] The vehicle control module 605 is used to control the vehicle to park in a parking space based on the spatiotemporal position and predicted trajectory of the vehicle.

[0093] An automatic parking system based on CSI proposed in an embodiment of the present invention matches the relative position relationship between a vehicle and a parking space through a position acquisition module, obtains the posture data of the vehicle, and provides an accurate data basis for subsequent perception reasoning. The wifi-CSI perception antenna matrix is ​​flexibly deployed through the CSI perception data acquisition module, which does not occupy the resources in the vehicle, enhances the penetration and diffraction capabilities of obstacles, and reduces the interference of complex occlusion scenes and environments on vehicle control. It can output richer tensor parameter information to improve the accuracy of the vehicle direction control during parking; by performing transfer learning on the pre-built parking space model, the model can quickly adapt to the complex and changeable environment, and improve the response speed of the model, and then the reasoning perception model construction module constructs the reasoning perception model to reason and predict the spatiotemporal posture and predicted trajectory of the vehicle. Through the posture evaluation and time series prediction of the reasoning prediction module, the generalization ability of the model for vehicle recognition is improved, thereby reducing the influence of multipath interference images, so that the control signal obtained by the vehicle control module to control the vehicle to park in the parking space is more accurate, thereby improving the accuracy of the vehicle direction control during parking.

[0094] A CSI-based automatic parking terminal includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the CSI-based automatic parking method is implemented.

[0095] A CSI-based automatic parking storage medium, the storage medium comprising: a computer-readable storage medium and a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the CSI-based automatic parking method.

[0096] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0097] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the described specific features, structures, materials or characteristics may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0098] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

Claims

1. An automatic parking method based on CSI, characterized in that: include: Based on a first preset algorithm, the relative position relationship between the vehicle and the berth is matched to obtain vehicle posture data; Based on the wifi-CSI sensing antenna matrix, obtain CSI sensing data; Based on the CSI perception data and the vehicle posture data, transfer learning is performed on the pre-built parking space model through a second preset algorithm to obtain an inference perception model; Based on the inference perception model, the vehicle's position and posture are evaluated and time series predicted to obtain the vehicle's spatiotemporal position and predicted trajectory; Based on the spatiotemporal position and predicted trajectory of the vehicle, the vehicle is controlled to park in a parking space.

2. The CSI-based automatic parking method according to claim 1, characterized in that: The method of matching the relative position relationship between the vehicle and the berth based on the first preset algorithm to obtain vehicle posture data includes: Acquire a vehicle image and a parking space image; Preprocessing the vehicle image and the parking space image to extract vehicle feature points and parking space feature points; Based on the first preset algorithm, the relative position relationship between the vehicle feature points and the feature points of the parking space to be parked is matched to obtain vehicle posture data.

3. The CSI-based automatic parking method according to claim 1, characterized in that: The obtaining of CSI sensing data based on the wifi-CSI sensing antenna matrix includes: Constructing a flat panel antenna group based on a first preset group of transmitting antennas and a second preset group of receiving antennas; Based on the flat antenna group, a wifi-CSI sensing antenna matrix is ​​constructed by orthogonally deploying a third preset group of receiving antennas; Based on the wifi-CSI sensing antenna matrix, obtaining CSI data of each receiving antenna; Based on the second preset algorithm, the CSI data of each receiving antenna is quotiented and signal transformed to obtain CSI perception data.

4. The CSI-based automatic parking method according to claim 1, characterized in that: Based on the CSI perception data and vehicle posture data, transfer learning is performed on the pre-built parking space model to obtain an inference perception model, including: Perform modal fusion encoding on CSI perception data and vehicle posture data to construct a comprehensive feature vector; Performing feature extraction on the comprehensive feature vector to obtain CSI tensor parameters; Based on the CSI tensor parameters, transfer learning is performed on the pre-built parking space model to obtain an inference perception model.

5. The CSI-based automatic parking method according to claim 4, characterized in that: Based on the inference perception model, the vehicle posture evaluation and time series prediction are performed to obtain the vehicle's spatiotemporal posture and predicted trajectory, including: Based on the preset CSI sampling frequency, the vehicle CSI perception data is sorted to obtain CSI information of several time segments; Combine CSI information of several time segments into continuous CSI time series information; The continuous CSI time series information is input into the reasoning perception model, and the vehicle posture evaluation and time series prediction are performed to obtain the vehicle's spatiotemporal posture and predicted trajectory.

6. The CSI-based automatic parking method according to claim 1, characterized in that: Based on the spatiotemporal position and predicted trajectory of the vehicle, controlling the vehicle to park in a parking space includes: Based on the vehicle's spatiotemporal position and predicted trajectory, the vehicle's driving boundary is obtained; Based on the vehicle driving boundary, a vehicle control signal is obtained; The vehicle is parked in the parking space by means of vehicle control signals.

7. An automatic parking system based on CSI, characterized in that: include: Position acquisition module, CSI perception data acquisition module, reasoning perception model building module, reasoning prediction module and vehicle control module; The position acquisition module is used to match the relative position relationship between the vehicle and the berth based on a first preset algorithm to obtain vehicle posture data; The CSI sensing data acquisition module is used to acquire CSI sensing data based on the wifi-CSI sensing antenna matrix; The inference perception model building module is used to perform transfer learning on the pre-built parking space model through a second preset algorithm based on the CSI perception data and the vehicle posture data to obtain an inference perception model; The inference prediction module is used to perform posture evaluation and time series prediction on the vehicle based on the inference perception model to obtain the spatiotemporal posture and predicted trajectory of the vehicle; The vehicle control module is used to control the vehicle to park in a parking space based on the spatiotemporal position and predicted trajectory of the vehicle.

8. The CSI-based automatic parking system according to claim 7, characterized in that: The CSI sensing data acquisition module is used to acquire CSI sensing data based on the wifi-CSI sensing antenna matrix, including: A first antenna group construction module, a second antenna group construction module, a CSI data receiving module and a CSI data processing module; The first antenna group building module is used to build a flat antenna group based on a first preset group of transmitting antennas and a second preset group of receiving antennas; The second antenna group construction module is used to construct a wifi-CSI sensing antenna matrix based on the flat antenna group by orthogonally deploying a third preset group of receiving antennas; The CSI data receiving module is used to obtain CSI data of each receiving antenna based on the wifi-CSI sensing antenna matrix; The CSI data processing module is used to perform quotient and signal transformation on the CSI data of each receiving antenna based on a second preset algorithm to obtain CSI perception data.

9. An automatic parking terminal based on CSI, characterized in that: The invention comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a CSI-based automatic parking method as claimed in any one of claims 1 to 6 is implemented.

10. An automatic parking storage medium based on CSI, characterized in that: The storage medium includes: a computer-readable storage medium and a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a CSI-based automatic parking method as described in any one of claims 1 to 6.

Images