CN108573247B

CN108573247B - Method for detecting parking space parking state based on edge calculation

Info

Publication number: CN108573247B
Application number: CN201810439574.7A
Authority: CN
Inventors: 秦诗玮; 赵辛; 褚英昊; 王树燚; 黄琛; 赵紫州; 叶丹微; 吴嘉杰
Original assignee: Shenzhen Weiai Intelligent Technology Co ltd
Current assignee: Shenzhen Weiai Intelligent Technology Co ltd
Priority date: 2018-05-09
Filing date: 2018-05-09
Publication date: 2021-09-28
Anticipated expiration: 2038-05-09
Also published as: CN108573247A

Abstract

The invention relates to a method for detecting parking status of a parking space based on edge calculation, which comprises the following steps: performing data characteristic engineering processing on original triaxial data which are acquired by a geomagnetic sensor and used for expressing the parking state of the parking space, and selecting the triaxial data subjected to data characteristic processing to perform value calculation; outputting a pre-judgment result through edge calculation according to the obtained operation value, screening out characteristic data, packaging and uploading the characteristic data to a cloud server; and combining the output result of the edge calculation with cloud processing, identifying interference under various special conditions, and realizing overall optimal judgment. The invention considers the fluctuation of data and dynamically uploads the data, thereby achieving the effect of energy saving; by means of the edge algorithm, data are uploaded under the limited condition of the equipment instead of periodically uploading the data, power consumption of the equipment is reduced, and the working period of the equipment is prolonged.

Description

Method for detecting parking space parking state based on edge calculation

Technical Field

The invention relates to a parking space parking state detection technology, in particular to a method for detecting a parking space parking state based on edge calculation.

Background

At present, the parking space is unsupervised and can be realized mainly through the following technologies: (1) a magnetic induction coil: a magnetic induction coil environment is artificially manufactured, and when an object enters the environment, the original magnetic induction wire is cut to generate induction current. Monitoring the parking space condition based on the information; (2) pattern recognition: the method comprises the following steps of manually setting corresponding modes according to images of vehicles and vehicles which are acquired by a camera, and monitoring the parking space condition in real time based on the corresponding modes; (3) infrared/RFID/bluetooth: based on the electromagnetic wave reflection phenomenon, under the condition that a vehicle is in a parking space or not in the parking space, the vehicle has different effects on electromagnetic waves in the environment, and the parking space condition is monitored by utilizing the difference. However, the above techniques have a common disadvantage in that they are relatively expensive to build and maintain. The cost of single camera and magnetic induction coil and set up one set of infrared, RFID, bluetooth equipment price still higher, in addition when setting up magnetic induction coil, laying the camera or laying equipment such as corresponding infrared, bluetooth, RFID label, can relate to and destroy original road surface, increase and lay corresponding power cord and signal line. Therefore, the total cost of the technologies is quite high due to the addition of the engineering amount and the equipment cost, and if only one function of monitoring the parking space is realized, the total investment is often not matched with the realized function return. Under this background, the technology that utilizes earth magnetism environment to carry out parking stall control is in force.

Based on the earth's own geomagnetic environment, when the vehicle entered into local geomagnetic environment, can cause certain influence to original geomagnetic environment, upload through loRa wireless transmission protocol with real-time geomagnetic data acquisition by geomagnetic sensor. The current parking space condition can be calculated by analyzing the real-time monitored geomagnetic data.

Edge Computing (Edge Computing) refers to providing a nearest-end service nearby by using an open platform with integrated network, Computing, storage and application core capabilities on one side close to an object or a data source. The application program is initiated at the edge side, so that the faster network service response is generated, and the basic requirements of the industry in the aspects of real-time business, application intelligence, safety, privacy protection and the like are met. The edge computation is between the physical entity and the industrial connection, or on top of the physical entity. And the cloud computing still can access the historical data of the edge computing.

Internet-connected devices for cell phones, tablets, sensors, security cameras and vehicles have produced a large amount of data that can be mined and analyzed. Globally, the cloud (cloud) is an indispensable part of this process. Unfortunately, as the distance between the cloud and the user increases, as does the transmission delay, edge computing has proven to be an effective way to bridge the distance between the cloud and the user. Edge computing involves placing a small edge server between the cloud and the user. This approach will help offload some of the workload from the cloud and server shoulders, which will help speed the growth of numerous application edge computing-driven data requiring low latency away from the central data server, providing real-time analysis, while reducing deployment costs. Up to now, edge calculation has been applied to several scenarios:

1. augmented Reality (Augmented Reality): augmented reality technology stacks text and images into a digital representation of a personal environment, and as AR applications proliferate, systems need to know the coordinates and movements of users, platforms can support AR services by providing highly localized data of interest to a particular user. Because the original data volume is large and disordered, the data excessive redundancy often appears, and the platform often does a large amount of useless work, the data processing process is slow, and the information feedback is delayed, so the arrival of edge calculation is equivalent to a process of filtering the original data, and the efficiency of the platform for processing the user information is greatly improved.

2. Virtual Reality (Virtual Reality scenes): the virtual reality technology is an important direction of simulation technology, and mainly comprises aspects of simulating environment, perception, natural skills, sensing equipment and the like. A very important process is called perception feedback, and the edge calculation can assist the VR system to realize a more efficient and vivid simulation effect through the characteristic processing of the surrounding environment data collected by the sensing equipment.

3. Unmanned aerial vehicle (Drones): the use of drones is expanding, including on the one hand in agriculture and on the other hand in the mining industry. While these devices must be able to respond to the data they collect by "the headquarters," edge calculations enable the drone to examine the data and respond to it in real time, for example when the drone identifies a car accident, the device may provide valuable information about the debris to nearby pedestrians.

4. Remote Monitoring of Oil and Gas (Oil & Gas Remote Monitoring): while conventional centralized data analysis infrastructures can determine when a shutdown will occur, they cannot identify the cause of the shutdown in real-time. By edge computing, oil and gas companies can access field data, enabling them to anticipate and protect against disasters.

5. Healthcare (Healthcare): with the rise of the digital age, the medical industry is changing. Devices such as various telehealth tools such as fitbit and glucose monitors are remodeling the healthcare field. The data stored on these devices can be used to update a patient's digital medical records, however existing cloud infrastructures are not able to manage the amount of data they produce. Edge computing will connect these medical devices to provide reliable and up-to-date patient information to hospitals and doctors in emergency situations.

The edge calculation of the current chip written in the geomagnetic equipment still has the following defects or difficulties: (1) due to the limitation of the memory size of geomagnetic equipment and the like, a large amount of operations cannot be directly and completely calculated by an edge algorithm; (2) due to the consideration of the power consumption of the equipment, the equipment cannot continuously upload data to a server (cloud end) in real time; (3) due to noise interference (such as ferrous substances, interference from approaching vehicles, temperature and humidity and the like) originally existing in the geomagnetic environment, the real parking space state information cannot be completely distinguished from the interference factors only through an edge algorithm of the writing device, and the layout of parts (such as the size of a battery and the like) among the devices can also have certain influence.

Disclosure of Invention

The invention aims to accurately judge common scenes before the judgment is processed by a cloud algorithm, realize the judgment accuracy rate of more than 95 percent and provide effective input data processed by characteristic engineering for solving the common complex scenes by the cloud algorithm.

In order to achieve the above object, the present invention provides a method for detecting parking status of a parking space based on edge calculation, comprising the following steps: performing data characteristic engineering processing on original triaxial data which are acquired by a geomagnetic sensor and used for expressing the parking state of the parking space, and selecting the triaxial data subjected to data characteristic processing to perform value calculation; outputting a pre-judgment result through edge calculation according to the obtained operation value, screening out characteristic data, packaging and uploading the characteristic data to a cloud server; the output result of the edge calculation is combined with the cloud processing, the interference under various special conditions is recognized, and the overall optimal judgment is realized.

Drawings

Fig. 1 is a schematic flow chart of a method for detecting a parking status of a parking space based on edge calculation according to an embodiment of the present invention;

FIG. 2 is a diagram of the variation of original triaxial data when a vehicle enters or exits a parking space;

FIG. 3 is a graph of changes in Mod;

FIG. 4 is a graph showing the change of the geomagnetic mod value at a sampling frequency of 2 Hz;

FIG. 5 is a graph showing the change of the geomagnetic mod value at a sampling frequency of 0.5 Hz;

FIG. 6 is a graph showing the change of the geomagnetic mod value at a sampling frequency of 0.2 Hz;

FIG. 7 is a complete data processing flow;

FIG. 8 is a graph illustrating a mean filtering process performed every 8 samples;

FIG. 9 is a schematic diagram of a data transfer process;

fig. 10 is a schematic diagram of a vehicle type a entering and exiting a parking space at a time.

Detailed Description

Other features, characteristics and advantages of the present invention will become more apparent from the following detailed description of embodiments of the present invention, which is to be read in connection with the accompanying drawings.

Fig. 1 is a schematic flow chart of a method for detecting a parking status of a parking space based on edge calculation according to an embodiment of the present invention.

As shown in fig. 1, the implementation of the method of the present invention mainly includes three parts: and performing data characteristic engineering, model building and judgment logic and data content of uploaded data aiming at the original geomagnetic sensor data. Through the model of the embodiment of the invention, the following breakthroughs are realized: (1) most of judgment aiming at the process at present is converted into identification judgment focusing on a steady-state result, so that the influence caused by environmental noise and self interference of equipment can be greatly reduced; (2) the power consumption required by the operation of each algorithm is reduced, and the uploading trigger condition is set to ensure that the change of the magnetic field strength exceeds a certain threshold, so that the energy and the electricity are saved, and the operation period of the equipment is prolonged; (3) a dynamic monitoring method is introduced to regulate and control the uploading frequency, so that the service life of the equipment is obviously prolonged by reducing the power consumption; (4) the data uploaded to the cloud end are processed by the characteristic engineering instead of rough original data, and a basis is provided for good performance of a cloud end algorithm; (5) the judgment accuracy rate of over 95% is realized at the equipment end, the problem is solved nearby, and the method of completely judging by depending on the cloud is not used originally.

First, data feature engineering is introduced

The data characteristic engineering means that original data are characterized through certain operation, and redundant information is filtered out on the premise that most of characteristic information related to specific application scenes is guaranteed. On one hand, according to the requirement of algorithm design, original data can be mapped to the dimension required by a designer, and the characteristic information required by the designer is extracted from huge data information; on the other hand, the work load of later algorithm operation can be greatly reduced, the operation efficiency is improved to the maximum extent, and useful information is processed in the algorithm operation process. Therefore, the design of the data characteristic engineering can directly influence the performance and the accuracy of the algorithm of the embodiment of the invention, and one of the innovation points of the invention is to design a suitable data characteristic engineering method according to the application scene of judging the parking space state.

The original data collected by the geomagnetic sensor is changed into a daily Gaussian value through data cleaning, and the Gaussian value is divided into three-axis data values of an X value, a Y value and a Z value in space. In the edge calculation, the algorithm occupies too much space and the related operation occupies too much memory space because the algorithm is directly burned into the chip of the geomagnetic equipment. On one hand, the power consumption of the device is saved as much as possible, the working life cycle of the device is increased, and on the other hand, more space is reserved as much as possible in a limited chip hardware environment to store geomagnetic data so as to be uploaded to a cloud end at a later period to serve as the input of a cloud algorithm.

In the process of analyzing data in the early period, a large number of experiments prove that the data processed by characteristic engineering such as primary data cleaning has the effect of representing the parking space state process characteristics better than the original X, Y, Z triaxial data. FIG. 2 shows the change of the original three-axis data when the vehicle enters or exits the parking space:

therefore, in the embodiment of the present invention, only one set of data processed by the feature engineering is selected as an input, that is, a Module value (Module) of the triaxial data, and a formula is as follows:

after the modulo operation, the characteristics of the triaxial data are collected into new data, which is shown in fig. 3.

Similar value-taking methods are various, such as a K value model and a Bznom value model:

under a large amount of experimental verification, after comprehensive consideration, the embodiment of the invention selects the Mod value as the input of the edge algorithm.

As shown in the above diagram, it can be seen that the curve presented by the geomagnetic data is not smooth, but has many spikes (local fluctuations), and in this case, the embodiment of the present invention includes an average filtering algorithm (also referred to as a sliding filtering algorithm) after the Mod calculation:

wherein M is_(k)For raw signals collected by the sensor, A_(k)For the filtered signal, N is the length of the sliding filter window.

Meanwhile, when the condition of multiple burrs occurs in the data waveform is processed, in addition to the above-mentioned sliding filtering method, the embodiment of the present invention is implemented by adopting another method, that is, changing the sampling frequency, the sampling frequency adopted by us is 2Hz (500nm) initially, and the sampling frequency adopted by us is 0.5Hz (2s) and 0.2Hz (5s) later, and the experimental results are respectively shown in fig. 4-fig. 6:

as shown in fig. 4 to 6, it can be clearly seen that the effect of the mean filtering can be obtained at the sampling frequencies of 0.5Hz and 0.2Hz, and the fluctuating interference caused by the entrance and exit of the vehicle can be filtered out, so that the oscillogram becomes smooth. A complete data processing procedure is shown in fig. 7:

next, the decision model is introduced

Common geomagnetic applications are mostly used for vehicle statistics and process analysis, so the analysis of the state process is emphasized, however, the method is not directed to the application scene of parking, and the computation amount and the power consumption are large, so the service life of the equipment is shortened. The application scene of the invention is to accurately judge whether a vehicle is in a parking space, so that the invention belongs to the judgment of a state result, but not the judgment of a state process. Therefore, the method has larger tolerance to fluctuation caused by the process that the vehicle enters and exits the parking space, so that steady state values under the vehicle-free and vehicle-in states are needed, and the core of the algorithm is around identifying the steady state at different moments. Generally speaking, a steady state of a vehicle-free state only has one value within a certain time range, but as the running time of equipment increases and magnetic saturation occurs, a reference value of the vehicle-free state has a certain deviation, and for this phenomenon, a real-time updating method is adopted to correct the reference value in real time, and a specific algorithm is as follows:

wherein α is a weighting coefficient, A_(k)Is the sampled value.

The detection of the parking state is complicated because the steady state values of the parking state are different according to the different final parking positions of the vehicle, so that the lowest value needs to be found and a corresponding threshold range is set. For example, the geomagnetic steady-state values of half and half of the occupied parking spaces are greatly different, and the steady-state values of different vehicle types at the same position are also different. By sampling geomagnetic data in real time, when the geomagnetic data is in a steady state within a certain time range, the geomagnetic data is compared with a reference value of a vehicle-free state closest to the moment for judgment.

Finally, introduction is made to dynamic monitoring and data uploading:

in the embodiment of the invention, the judgment of the parking space condition is firstly divided into two situations, namely a general situation and a special situation, in real life, the general situation accounts for about 80-90 & so of the total situation, and under the situation, the influence factor is single. Therefore, the judgment with high accuracy can be realized through edge calculation, in addition, the original data under other conditions can be subjected to feature processing through the edge calculation, corresponding feature information is extracted and packaged together with the judgment result to be uploaded, and a foundation is laid for the judgment under the special condition of cloud solution later.

The upload data contains the following: (1) uploading conditions; (2) an uploading form; (3) uploading the content; the upload condition is the trigger condition we usually say. In the method, judgment is carried out through the threshold value, so that the uploading frequency of the equipment end can be effectively reduced, the power consumption is saved, and the uploading effectiveness can be ensured. Before this, we first perform filtering processing once, and process the original data once every N times, and then obtain corresponding a (k), which is the real-time sampling data after the sliding filtering processing. The process is shown in fig. 8.

The upload (trigger) conditions are as follows: (A (k) is the real-time magnetic field value after filtering treatment, M is a set threshold value)

The uploading is in the form of how to send the local data to the cloud server, and the process is shown in fig. 9.

The uploaded content comprises a result judged by a local edge algorithm, data when uploading conditions are triggered, previous 20 groups of historical data and a one-time no-vehicle state reference value closest to the current state.

Effective effects of embodiments of the present invention

Most of earth magnetism equipment can have the judgement of fine performance and higher rate of accuracy under single environment or single scene, in case meet comparatively complicated condition, for example the business turn over parking stall of different angles, different speeds of a motor vehicle etc. then can decline to some extent in judging the rate of accuracy. On the other hand, most geomagnetic devices need to upload a large amount of related data to the cloud end due to low accuracy of local algorithm judgment, so as to make up for the accuracy. In the invention, the designed edge calculation can normally operate in more complex scenes than before. Meanwhile, the data volume uploaded to the cloud end can be effectively reduced through the achieved accuracy, and the sampling frequency of the device is reduced. Specifically, the data is uploaded intermittently, for example, sampled data is uploaded every 5s, 720 times are uploaded within 1 hour, and if data is uploaded every time, current of 30 μ a is consumed, current of 2.16 × 10-2A is consumed in one hour, and power consumption is very high; if the model is adopted, assuming that the parking time of one vehicle is 10min on average under the condition of real-time saturation of the parking space, the data are uploaded for 12 times in one hour, 3.6 x 10-4A of current is consumed totally, and compared with the prior method, the method saves about 98% of power consumption, and if the parking space is not under the condition of full saturation, the power consumption is saved more. Therefore, the service life of the equipment can be really prolonged from the actual situation through the model, and the performance of the equipment is improved.

According to the conditions of different vehicle types, different parking space entering and exiting modes, different parking scenes and the like, the results of the edge algorithm identification and judgment are as follows:

single-pass parking place for vehicles

Vehicle model	Number of times of measurement of X	X ground determination accuracy	Number of measurements of Y	Accuracy of Y ground determination
					A	35	99％	27	99％
B	32	99％	26	99％

Parking space for vehicles to go in and out at different angles

Parking space for vehicles to go in and out at different speeds

Vehicle model

Vehicle speed

Number of measurements

Rate of accuracy of determination

Vehicle speed

Number of measurements

Rate of accuracy of determination

A

Fast-acting toy

26

99％

Slow

28

99％

Parking space for vehicles to go in and out in complex condition

Vehicle model	Parking scene	Number of measurements	Rate of accuracy of determination
				B
	2 times of feeding and 2 times of discharging	24	99％
				B
	2 times out and 2 times in	28	99％
				A
	2 times of feeding and 1 time of discharging	26	99％

The embodiment of the invention can improve the edge calculation performance of geomagnetic parking recognition, and when edge calculation is used, data are collected in real time through a geomagnetic sensor, a group of data with the most representatives is selected for processing after characteristic engineering processing, then the pre-judgment result is output through edge calculation, and the characteristic data is screened out and packed together for uploading, so that the effectiveness of uploading data can be greatly improved, and finally the interference under various special conditions is recognized by combining the output result of the edge calculation with cloud processing, and the optimal judgment of the whole system is realized.

It should be noted that the above embodiments are only used for illustrating the structure and the working effect of the present invention, and are not used for limiting the protection scope of the present invention. Modifications and adaptations to the above-described embodiments may occur to one skilled in the art without departing from the spirit and scope of the present invention and are intended to be covered by the following claims.

Claims

1. The method for detecting the parking state of the parking space based on the edge calculation is characterized by comprising the following steps of:

performing data characteristic engineering processing on original triaxial data which are acquired by a geomagnetic sensor and used for expressing the parking state of the parking space, and selecting the triaxial data subjected to data characteristic processing to perform value calculation;

outputting a pre-judgment result through edge calculation according to the obtained operation value, screening out characteristic data, packaging and uploading the characteristic data to a cloud server;

the output result of the edge calculation is combined with the cloud processing, the interference under various special conditions is recognized, and the integral optimal judgment is realized;

the selection of the three-axis data subjected to data characterization processing for value calculation comprises the following steps:

selecting a group of three-axis data subjected to characterization processing, and taking a module value through the following formula to obtain the module value as the input of edge calculation;

wherein x, y and z are three-axis data.

2. The method of claim 1, wherein the data waveform is processed in the presence of multiple glitches by incorporating a mean filtering algorithm after the modulo calculation, which is expressed by the following equation:

3. The method of claim 1, wherein the sampling frequency of the data waveform is 2Hz initially and 0.5Hz and 0.2Hz later when the data waveform is processed, so as to achieve the effect of mean filtering and minimize the influence of noise.

4. The method according to claim 1, wherein the uploading condition for uploading the data to the cloud server is judged by a threshold value, so that the uploading frequency of the device side can be effectively reduced, the power consumption can be saved, and the uploading effectiveness can be ensured;

the threshold is set by the following formula:

wherein, A (k) is the real-time magnetic field value after filtering, and M is the set threshold.

5. The method of claim 4, wherein a filtering process is performed before the data uploading, and the corresponding A (k) is obtained after each N times of raw data processing, namely the real-time sampled data after the sliding filtering process.

6. The method of claim 1, wherein the step of packaging and uploading the screened feature data together to a cloud server comprises:

and uploading the data when the condition is triggered and the previous 20 groups of historical data according to the judgment result of the local edge algorithm, and the reference value of the one-time no-vehicle state closest to the current state.

7. The method according to claim 1, wherein the reference value of the vehicle-free state is corrected in real time by a real-time updating method, and the specific formula is as follows:

wherein α is a weighting coefficient, A_(k)Is the sampled value.

8. The method of claim 1, wherein a K value model is used for value calculation; the K value model and the mod value operation are in a parallel relation, and meanwhile, a data characteristic processing mode is adopted.

9. The method of claim 1, wherein a Bznom value model is used for value calculation; the Bznom value model and the mod value operation are in a parallel relation, and meanwhile, a data characteristic processing mode is adopted.