CN113470416A - System, method and storage medium for realizing parking space detection by using embedded system - Google Patents
System, method and storage medium for realizing parking space detection by using embedded system Download PDFInfo
- Publication number
- CN113470416A CN113470416A CN202010242254.XA CN202010242254A CN113470416A CN 113470416 A CN113470416 A CN 113470416A CN 202010242254 A CN202010242254 A CN 202010242254A CN 113470416 A CN113470416 A CN 113470416A
- Authority
- CN
- China
- Prior art keywords
- model
- dlod
- parking space
- module
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/14—Traffic control systems for road vehicles indicating individual free spaces in parking areas
- G08G1/145—Traffic control systems for road vehicles indicating individual free spaces in parking areas where the indication depends on the parking areas
- G08G1/146—Traffic control systems for road vehicles indicating individual free spaces in parking areas where the indication depends on the parking areas where the parking area is a limited parking space, e.g. parking garage, restricted space
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/14—Traffic control systems for road vehicles indicating individual free spaces in parking areas
- G08G1/145—Traffic control systems for road vehicles indicating individual free spaces in parking areas where the indication depends on the parking areas
- G08G1/148—Management of a network of parking areas
Abstract
The invention relates to a system, a method and a storage medium for realizing parking space detection by utilizing an embedded system. The system for realizing parking space detection by utilizing the embedded system comprises: a Deep Learning Object Detection (DLOD) model building module configured to generate a first DLOD model and convert the first DLOD model into a second DLOD model suitable for a board end; the system comprises a case construction and compiling module, a parking space detection and processing module and a storage module, wherein the case construction and compiling module is configured to construct a parking space detection case and compile the case to generate a system file for an embedded system; the video preprocessing module is configured to preprocess the acquired parking space video to generate parking space data; and the embedded system operation module is configured to realize parking space detection by the embedded system based on the second DLOD model, the system file and the parking space data.
Description
Technical Field
The present invention relates to the field of automatic parking. Specifically, the invention relates to a parking space detection system and method.
Background
With the rapid development of automobile intellectualization, more and more automobiles are configured with different Advanced Driving Assistance Systems (ADAS), wherein an Auto Park Assist (APA) System can realize automatic Parking space detection and Parking, and effectively solve the problems of difficult Parking and difficult Parking. At present, a lot of vehicles are equipped with ultrasonic radar sensors to identify parking spaces to realize a semi-automatic parking function, but due to the hardware defects of the ultrasonic radar, the problems of detection blind areas, obstacle interference and the like exist, and the parking efficiency is low. Therefore, more and more vehicles are additionally provided with a plurality of camera sensors, parking spaces are detected by a method based on machine vision and deep learning, the parking space detection precision is improved by training a complex neural network detection model, and meanwhile, great requirements are provided for a hardware computing platform. One solution to this problem is to use a cloud, and if the edge calculation is performed at the cloud and then the terminal is returned, there is a delay, which results in low security. The other solution is to adopt an industrial personal computer as a hardware computing platform, but the cost for installing the industrial personal computer on the vehicle is too high, and the industrial personal computer cannot be produced in mass.
Aiming at the field of automatic parking auxiliary systems in the current automobile industry, the following limitations are summarized:
1) the parking scheme based on the ultrasonic radar in the market at present cannot detect the parking spaces without vehicles on two sides, and has the problems of detection blind areas, environmental interference and the like, so that the parking efficiency is poor;
2) with the camera image-based recognition approach, high-performance hardware computing platforms are required, or computing is performed in the cloud. If the data are calculated in the cloud, the data need to be acquired by the equipment terminal and then uploaded, and the data are returned to the terminal after the calculation is finished, so that a certain time delay is brought, and the safety of an automobile system is low;
3) the industrial personal computer hardware platform is limited by installation space and too high in cost, and mass production cannot be realized.
Disclosure of Invention
Therefore, how to realize accurate, efficient, safe and low-cost automatic parking is an urgent problem to be solved by the current development of an APA parking system.
Therefore, a system and a method for detecting a parking space based on deep learning, which can be implemented on a low-cost embedded hardware platform, are needed.
To achieve one or more of the above objects, the present invention provides the following technical solutions.
According to a first aspect of the present invention, there is provided a system for implementing parking space detection by using an embedded system, comprising: a Deep Learning Object Detection (DLOD) model building module configured to generate a first DLOD model and convert the first DLOD model into a second DLOD model suitable for a plate end; the use case constructing and compiling module is configured to construct a parking space detection use case and compile the use case to generate a system file for the embedded system; the video preprocessing module is configured to preprocess the acquired parking space video to generate parking space data; and the embedded system operation module is configured to realize the parking space detection by the embedded system based on the second DLOD model, the system file and the parking space data.
According to an embodiment of the present invention, the system for implementing parking space detection by using an embedded system, wherein the DLOD model building module further includes: a model design module configured to generate a preliminary DLOD model; a model training module configured to train the preliminary DLOD model with training data to obtain a trained DLOD model; a model testing module configured to test the trained DLOD model with test data; and a model conversion module configured to receive the first DLOD model from the model test module and convert the first DLOD model to the second DLOD model.
According to another embodiment of the present invention or any one of the above embodiments, the system for implementing parking space detection using an embedded system, wherein the model testing module is further configured to: sending the trained DLOD model that passes the test as the first DLOD model to the model conversion module; and sending the trained DLOD model that fails the test to the model design module.
According to another embodiment of the present invention or any one of the above embodiments, the system for implementing parking space detection by using an embedded system, wherein the embedded system operation module includes: the system comprises a starting system and an on-board operation module, wherein the starting system carries out the parking space detection by loading the second DLOD model, the system file and the parking space data on the on-board operation module.
According to another embodiment of the present invention or any one of the above embodiments, the system for implementing parking space detection by using an embedded system further includes: a dataflow graph design module configured to generate operating logic for performing the parking space detection for use by the use case construction and compilation module.
According to another embodiment of the present invention or any one of the above embodiments, the system for implementing parking space detection by using an embedded system, wherein the system file generated by the use case constructing and compiling module includes a parking space detection mirror image file and a system starting file.
According to another embodiment of the present invention or any one of the above embodiments, the system for implementing parking space detection using an embedded system, wherein the model design module is further configured to: and generating the preliminary DLOD model by adopting a Caffe deep learning architecture and an SSD network architecture.
According to a second aspect of the present invention, there is provided a method for implementing parking space detection by using an embedded system, comprising the following steps: generating a first DLOD model, and converting the first DLOD model into a second DLOD model suitable for a plate end; constructing a parking space detection case, and compiling the case to generate a system file for the embedded system; preprocessing the acquired parking space video to generate parking space data; and realizing the parking space detection by the embedded system based on the second DLOD model, the system file and the parking space data.
According to an embodiment of the present invention, the method for detecting a parking space by using an embedded system, wherein the step of generating the first DLOD model further includes: generating a primary DLOD model; training the preliminary DLOD model by using training data to obtain a trained DLOD model; and testing the trained DLOD model using test data.
According to another embodiment of the present invention or any one of the above embodiments, the method for detecting a parking space using an embedded system, wherein the step of testing the trained DLOD model further includes: sending the trained DLOD model that passes the test as the first DLOD model to a model conversion module; and sending the trained DLOD model that fails the test to a model design module.
According to another embodiment of the present invention or any one of the above embodiments, the method for detecting a parking space using an embedded system, wherein the step of using the embedded system to detect the parking space includes: and carrying out the parking space detection by a starting system through loading the second DLOD model, the system file and the parking space data on an on-board operation module.
According to another embodiment of the present invention or any one of the above embodiments, the method for detecting a parking space using an embedded system further includes: and generating operation logic for executing the parking space detection.
According to another embodiment of the present invention or any one of the above embodiments, the method for implementing parking space detection by using an embedded system, wherein the system file includes a parking space detection image file and a system start file.
According to another embodiment of the present invention or any one of the above embodiments, the method for detecting a parking space using an embedded system, wherein the step of generating the preliminary DLOD model further includes: and generating the preliminary DLOD model by adopting a Caffe deep learning architecture and an SSD network architecture.
According to a third aspect of the present invention, there is provided a non-transitory computer readable storage medium storing computer program instructions which, when executed, implement the method for implementing parking space detection using an embedded system according to the second aspect of the present invention.
The parking space detection system and method according to the invention eliminate or at least reduce the limitation of the parking space detection link in the automatic parking system in the prior art. According to the embedded design scheme for the parking space detection of the automatic parking system, a deep learning parking space detection algorithm can be operated on a board-end embedded system, and real-time detection and positioning of parking spaces in a complex and extreme environment are realized, so that the parking space detection accuracy and the generalization capability of parking space detection scenes are improved, and a foundation is laid for further realizing development of the automatic parking system.
Drawings
The above and/or other aspects and advantages of the present invention will become more apparent and more readily appreciated from the following description of the various aspects taken in conjunction with the accompanying drawings, in which like or similar elements are designated with like reference numerals. The drawings comprise:
FIG. 1 is a schematic block diagram of a system for implementing parking space detection using an embedded system according to an embodiment of the present invention;
FIG. 2 is a schematic flow diagram of deep learning based generation of a DLOD model according to an embodiment of the present invention;
fig. 3 is a flow chart of DLOD parking space detection data flow design according to an embodiment of the present invention; and
fig. 4 is a schematic flowchart of a method for implementing parking space detection by using an embedded system according to an embodiment of the present invention.
Detailed Description
In this specification, the invention is described more fully with reference to the accompanying drawings, in which exemplary embodiments of the invention are shown. This invention may, however, be embodied in different forms and should not be construed as limited to the embodiments set forth herein. The embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Words such as "comprising" and "comprises" mean that, in addition to having elements or steps which are directly and unequivocally stated in the description and the claims, the solution of the invention does not exclude other elements or steps which are not directly or unequivocally stated. Terms such as "first" and "second" do not denote an order of the elements in time, space, size, etc., but rather are used to distinguish one element from another.
The present invention is described below with reference to flowchart illustrations, block diagrams, and/or flow diagrams of methods and systems according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block and/or flow diagram block or blocks.
These computer program instructions may be loaded onto a computer or other programmable data processor to cause a series of operational steps to be performed on the computer or other programmable processor to produce a computer implemented process such that the instructions which execute on the computer or other programmable processor provide steps for implementing the functions or acts specified in the flowchart and/or block diagram block or blocks. It should also be noted that, in some alternative implementations, the functions/acts noted in the blocks may occur out of the order noted in the flowcharts. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
Referring to fig. 1, a system 100 for implementing parking space detection by using an embedded system according to an embodiment of the present invention is described. The system 100 includes a DLOD (deep learning object detection) model building module 110, a use case building and compiling module 120, a video preprocessing module 130, and an embedded system running module 140.
The DLOD model building module 110 is configured to generate a first DLOD model and convert the first DLOD model into a second DLOD model suitable for the plate end. For example, a deep learning-based parking space target detection algorithm model is converted into a DLOD model which can be operated by a board end. In one embodiment, the DLOD model building module 110 may include: a model design module 1102, a model training module 1104, a model testing module 1106, and a model transformation module 1108.
At present, a target detection algorithm for traditional machine learning is complex, wherein manually designed features are poor in robustness, and a sliding frame is inaccurate in positioning, so that the parking space recognition rate is low. In the present invention, the model design module 1102 of the DLOD model building module 110 of the system 100 may adopt a parking space recognition algorithm based on a deep convolutional network, and design a parking space detection algorithm for an automatic parking system by using a Caffe deep learning architecture and an SSD network architecture, so as to generate a preliminary DLOD model (i.e., model design). By adopting the SSD parking space detection algorithm based on deep learning, the end-to-end parking space detection is realized, the running speed and the detection precision are improved, and the detection algorithm has strong adaptability under various environments. At present, in consideration of limited hardware computing capacity and difficulty of embedded operation, most automatic parking systems adopt an ultrasonic radar or a traditional vision method to detect parking spaces, and have the problems of detection blind areas, poor environmental adaptability and the like.
The model conversion module 1108 may receive the first DLOD model from the model test module 1106 and convert the first DLOD model to a second DLOD model. For example, model conversion module 1108 may convert the first DLOD model into a second DLOD model that is board-end operable and that includes two executable files, a parking space detection model structure (e.g., OD _ net.
Based on the DLOD model building module 110, end-to-end parking space detection can be realized, and the accuracy and the environmental adaptivity of parking space detection are improved by training and optimizing the model through a big data parking space data set.
A use case construction and compilation module 120 configured to construct a parking space detection use case and compile the use case to generate a system file for the embedded system. The use case constructing module 120 may be used as a core for embedded implementation of parking space detection, and on the basis of an input/output operating system (BIOS), as an interface driver between a development board end and a parking space detection algorithm model, and completes software development of a parking space detection use case by calling a data stream program. Then, the use case is compiled to generate system files (e.g., a parking space detection image file and a system start file) for use by the embedded system running module 140 later. Specifically, an operation platform is constructed according to embedded hardware, an input/output BIOS system is used as a real-time operation system for the operation of the parking space detection use case, and a CPU (IPU/MPU/EVE/DSP) for the operation of the use case is specified. And then developing an interface driving program between the board end and the algorithm to construct a parking space detection application program, and finally generating a parking space detection mirror image file and a system starting file through compiling. The parking space detection mirror image file can be generated as an application program file. In one embodiment, the use case building and compiling module 120 may include a use case building module and a use case compiling module to respectively perform the corresponding functions described above.
The video pre-processing module 130 is configured to pre-process the acquired parking space video to generate parking space data. The video pre-processing module 130 may pre-process the parking space video stream (e.g., obtained by an onboard sensor such as a camera) by cropping it, convert it into a video data format (e.g., InData and InHeader) that can be recognized by the parking space detection algorithm model, and use it as an input to the parking space detection algorithm model. Specifically, the parking space video stream is firstly subjected to preprocessing such as cutting, so that the pixel size meets the requirements of a parking space detection algorithm model. The video stream is then converted to a specific data frame format for recognition and processing by the parking space detection algorithm.
The embedded system operating module 140 is configured to implement the parking space detection with the embedded system based on the second DLOD model, the system file, and the parking space data. The embedded system execution module 140 includes: a startup system 1402, and an on-board run module 1404. Wherein the start system 1402 performs the parking space detection by loading the second DLOD model, the system file, and the parking space data onto the onboard operation module 1404. In one embodiment, the startup system 1402 and the onboard operation module 1404 start the embedded system by writing into the SD card, which includes a second DLOD model (e.g., a parking space detection DLOD model), a use case image file, a system startup file, a system configuration file, and the like, thereby implementing embedded operation of parking space detection. Specifically, an SD card starting file is manufactured, and the parking spot target detection is realized on the embedded system on chip. The method comprises the steps of firstly manufacturing an SD card, integrating files such as OD _ Net.bin, OD _ Prm.bin, vehicle-mounted detection mirror image files generated by compiling by the case construction and compiling module 120 and system starting files and the like in the second DLOD model to start an embedded system, setting hardware and electrifying to operate, and finally realizing embedded detection of the parking space.
In one embodiment, the system 100 further includes a dataflow graph design module 150 configured to generate operating logic for performing the parking space detection for use by the use case construction and compilation module. The dataflow graph design module 150 may be a "link and link" software architecture for designing parking space detection. That is, the data flow diagram design module 150 may design the operation logic of the parking space video stream and the algorithm, and form a link by specifying the link modules required to be used for parking space detection and the cores of each link operation and connecting the link modules according to the operation logic, so as to be called by the use case construction module 120 (e.g., target detection). Specifically, the data flow graph design module 150 may design a parking space detection data flow graph and determine a video processing and algorithm operation core. The links are formed by establishing the connection of the links, and the SOC (System-on-a-Chip) core of each link is appointed to operate, so that the operation logic of the video stream and the parking space detection algorithm is realized. As shown in fig. 3, the operations related to reading, decoding, and video Processing of the video stream may be performed in an ipu (image Processing Unit) core, then an MPU (Micro-Processing Unit) is used to perform preprocessing and scheduling, and the video is distributed to each EVEs (embedded Vision engine) core to be processed, and the parking space feature extraction needs to perform a large amount of convolution calculation, and thus is performed in the EVEs core. The vehicle position point detection layer is used as the last layer and is processed on the DSP core, so that the detection efficiency can be improved. Finally, the coordinates of the parking space points are obtained for the planning decision and control module of the automatic parking system to use.
A method 200 for implementing parking space detection using an embedded system according to an embodiment of the present invention is described with reference to fig. 2-4. As shown in fig. 4, the method 200 includes: generating a first DLOD model, and converting the first DLOD model into a second DLOD model suitable for the plate end (S210); constructing a parking space detection use case, and compiling the use case to generate a system file for the embedded system (S220); preprocessing the acquired parking space video to generate parking space data (S230); and realizing parking space detection by an embedded system based on the second DLOD model, the system file and the parking space data (S240).
In step S210: generating a first DLOD model, and converting the first DLOD model into a second DLOD model suitable for the plate end. For example, a deep learning-based parking space target detection algorithm model is converted into a DLOD model which can be operated by a board end. The step of generating the first DLOD model and the second DLOD model may include: model design S2102, model training S2104, model testing S2106, and model transformation S2108, as shown in fig. 2.
At present, a target detection algorithm for traditional machine learning is complex, wherein manually designed features are poor in robustness, and a sliding frame is inaccurate in positioning, so that the parking space recognition rate is low. In the method 200, in step S2102, a parking space recognition algorithm based on a deep convolutional network may be adopted, and a Caffe deep learning architecture and an SSD network architecture are utilized to design a parking space detection algorithm for an automatic parking system, so as to generate a preliminary DLOD model (i.e., model design). Therefore, the SSD parking space detection algorithm based on deep learning is adopted, end-to-end parking space detection is achieved, meanwhile, the running speed and the detection precision are improved, and the detection algorithm is high in adaptability under various environments. At present, in consideration of limited hardware computing capacity and difficulty of embedded operation, most automatic parking systems adopt an ultrasonic radar or a traditional vision method to detect parking spaces, and have the problems of detection blind areas, poor environmental adaptability and the like.
In step S2104, the preliminary DLOD model may be trained using a preferably large amount of training data to arrive at a trained DLOD model.
In step S2106, the trained DLOD model may be tested with test data, where the test data may be test data sets from different types of parking spaces. In step S2107, a detection accuracy test may be performed on the trained DLOD model, and the trained DLOD model that passes the test is sent to the model conversion module 1108 as the first DLOD model; and send the trained DLOD model that failed the test to the model design module 1102. For example, for a DLOD model trained because the detection accuracy does not meet the mass production standard, the DLOD model may be fed back to the model design module 1102 to continuously perfect the structure of the network model until the parking space detection accuracy of the DLOD model trained reaches the test standard.
In step S2108, a first DLOD model may be received and converted into a second DLOD model. For example, the first DLOD model may be converted into a second DLOD model that is executable on the board end, and the second DLOD model includes two executable files of a parking space detection model structure (e.g., OD _ net.
Based on the above step S210 for generating the first and second DLOD models, end-to-end parking space detection can be achieved, and the model is trained and optimized through the big data parking space data set to improve the accuracy of parking space detection and environmental adaptivity.
In step S220: and constructing a parking space detection use case, and compiling the use case to generate a system file for the embedded system. An interface driving program between a development board end and a parking space detection algorithm model can be compiled on the basis of an input/output operating system (BIOS), and software development of a parking space detection case is completed by calling a data flow program. Then, the use case is compiled to generate system files (e.g., a parking space detection image file and a system start file) for use by the embedded system running module 140 later. Specifically, an operation platform is constructed according to embedded hardware, an input/output BIOS system is used as a real-time operation system for the operation of the parking space detection use case, and a CPU (IPU/MPU/EVE/DSP) for the operation of the use case is specified. And then developing an interface driving program between the board end and the algorithm to construct a parking space detection application program, and finally generating a parking space detection mirror image file and a system starting file through compiling. The parking space detection mirror image file can be generated as an application program file.
In step S230, the acquired parking space video is preprocessed to generate parking space data. The parking space video stream (e.g., obtained by an onboard sensor such as a camera) may be pre-processed, such as cut, converted into a video data format (e.g., InData and InHeader) that may be recognized by the parking space detection algorithm model, and used as an input to the parking space detection algorithm model. Specifically, the parking space video stream is firstly subjected to preprocessing such as cutting, so that the pixel size meets the requirements of a parking space detection algorithm model. The video stream is then converted to a specific data frame format for recognition and processing by the parking space detection algorithm.
In step S240, the parking space detection is implemented by the embedded system based on the second DLOD model, the system file, and the parking space data. The parking space detection may be performed by loading the second DLOD model, the system file, and the parking space data onto the onboard operation module 1404. In one embodiment, the embedded system may be started by writing into the SD card, and includes a second DLOD model (e.g., a parking space detection DLOD model), a use case image file, a system start file, a system configuration file, and the like, so as to implement embedded operation of parking space detection. Specifically, an SD card starting file is manufactured, and the parking spot target detection is realized on the embedded system on chip. Firstly, an SD card is manufactured, files such as OD _ Net.bin and OD _ Prm.bin included in a second DLOD model, vehicle-mounted detection mirror image files generated through compiling, system starting files and the like are collected to be used for starting an embedded system, meanwhile, hardware is set and powered on to operate, and finally, embedded detection of a parking space is achieved.
In one embodiment, the method may further comprise generating operating logic for performing the parking space detection. The operation logic of the parking space video stream and the algorithm may be designed, and the link modules and the cores of the link operations that need to be adopted for parking space detection are designated and connected together according to the operation logic to form a link, so as to be called in step S220 of (for example, target detection) building a parking space detection case. Specifically, a parking space detection data flow graph can be designed and a video processing and algorithm operation core can be determined. The links are formed by establishing the connection of the links, and the SOC (System-on-a-Chip) core of each link is appointed to operate, so that the operation logic of the video stream and the parking space detection algorithm is realized. As shown in fig. 3, the operations related to reading, decoding, and video Processing of the video stream may be performed in an ipu (image Processing Unit) core, then an MPU (Micro-Processing Unit) is used to perform preprocessing and scheduling, and the video is distributed to each EVEs (embedded Vision engine) core to be processed, and the parking space feature extraction needs to perform a large amount of convolution calculation, and thus is performed in the EVEs core. The vehicle position point detection layer is used as the last layer and is processed on the DSP core, so that the detection efficiency can be improved. Finally, the coordinates of the parking space points are obtained for the planning decision and control module of the automatic parking system to use.
According to a third aspect of the present invention, a non-transitory computer-readable storage medium is provided storing computer program instructions that, when executed, implement the method 200 for implementing parking space detection using an embedded system according to the present invention.
According to the system and the method for realizing the parking space detection by utilizing the embedded system, the scheme that each module for the parking space detection runs on the processing cores of different Systems On Chip (SOC) is designed, the advantages of hardware and an algorithm are fully exerted, and the running speed of the algorithm at a terminal is improved; and each module of the parking space detection algorithm is operated according to the characteristics of the processing core, so that good resources and computing capacity are fully utilized, the frame rate of parking space detection is improved, and further automatic parking is guaranteed.
The parking space detection system and method according to the invention eliminate or at least reduce the limitation of the parking space detection link in the automatic parking system in the prior art. According to the embedded design scheme for the parking space detection of the automatic parking system, a deep learning parking space detection algorithm can be operated on a board-end embedded system, so that on one hand, real-time detection and positioning of parking spaces in complex and extreme environments are realized, the parking space detection accuracy and the generalization capability of parking space detection scenes are improved, and a foundation is laid for further realizing the development of the automatic parking system; on the other hand, information interaction with a cloud end is not needed, and the real-time performance and the safety are good.
It is noted that the present invention generally belongs to the field of embedded development of automatic parking systems for intelligent automobiles, but can be extended to the field of unmanned and embedded artificial intelligence without departing from the true intention of the present invention.
The embodiments and examples set forth herein are presented to best explain the embodiments in accordance with the present technology and its particular application and to thereby enable those skilled in the art to make and utilize the invention. However, those skilled in the art will recognize that the foregoing description and examples have been presented for the purpose of illustration and example only. The description as set forth is not intended to cover all aspects of the invention or to limit the invention to the precise form disclosed.
Claims (15)
1. The utility model provides a system for utilize embedded system to realize parking stall detection, includes:
a Deep Learning Object Detection (DLOD) model building module configured to generate a first DLOD model and convert the first DLOD model into a second DLOD model suitable for a plate end;
the use case constructing and compiling module is configured to construct a parking space detection use case and compile the use case to generate a system file for the embedded system;
the video preprocessing module is configured to preprocess the acquired parking space video to generate parking space data; and
an embedded system operation module configured to implement the parking space detection with the embedded system based on the second DLOD model, the system file, and the parking space data.
2. The system of claim 1, wherein the DLOD model building module further comprises:
a model design module configured to generate a preliminary DLOD model;
a model training module configured to train the preliminary DLOD model with training data to obtain a trained DLOD model;
a model testing module configured to test the trained DLOD model with test data; and
a model conversion module configured to receive the first DLOD model from the model test module and convert the first DLOD model to the second DLOD model.
3. The system of claim 2, wherein the model testing module is further configured to:
sending the trained DLOD model that passes the test as the first DLOD model to the model conversion module; and
sending the trained DLOD model that fails the test to the model design module.
4. The system of claim 1, wherein the embedded system runtime module comprises:
the system comprises a starting system and an on-board operation module, wherein the starting system carries out the parking space detection by loading the second DLOD model, the system file and the parking space data on the on-board operation module.
5. The system of claim 1, further comprising:
a dataflow graph design module configured to generate operating logic for performing the parking space detection for use by the use case construction and compilation module.
6. The system of claim 1, wherein the system files generated by the use case building and compiling module include a parking space detection image file and a system start file.
7. The system of claim 2, wherein the model design module is further configured to:
and generating the preliminary DLOD model by adopting a Caffe deep learning architecture and an SSD network architecture.
8. A method for realizing parking space detection by using an embedded system comprises the following steps:
generating a first DLOD model, and converting the first DLOD model into a second DLOD model suitable for a plate end;
constructing a parking space detection case, and compiling the case to generate a system file for the embedded system;
preprocessing the acquired parking space video to generate parking space data; and
and realizing the parking space detection by the embedded system based on the second DLOD model, the system file and the parking space data.
9. The method of claim 8, wherein the step of generating a first DLOD model further comprises:
generating a primary DLOD model;
training the preliminary DLOD model by using training data to obtain a trained DLOD model; and
the trained DLOD model was tested using test data.
10. The method of claim 9, wherein the step of testing the trained DLOD model further comprises:
sending the trained DLOD model that passes the test as the first DLOD model to a model conversion module; and
sending the trained DLOD model that fails the test to a model design module.
11. The method of claim 8, wherein the step of implementing the parking spot detection with the embedded system comprises:
and carrying out the parking space detection by a starting system through loading the second DLOD model, the system file and the parking space data on an on-board operation module.
12. The method of claim 8, further comprising:
and generating operation logic for executing the parking space detection.
13. The method of claim 8, wherein the system files include a parking space detection image file and a system start file.
14. The method of claim 9, wherein the step of generating a preliminary DLOD model further comprises:
and generating the preliminary DLOD model by adopting a Caffe deep learning architecture and an SSD network architecture.
15. A non-transitory computer readable storage medium storing computer program instructions which, when executed, implement a method of implementing parking spot detection with an embedded system according to claims 8-14.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010242254.XA CN113470416B (en) | 2020-03-31 | 2020-03-31 | System, method and storage medium for realizing parking space detection by using embedded system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010242254.XA CN113470416B (en) | 2020-03-31 | 2020-03-31 | System, method and storage medium for realizing parking space detection by using embedded system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113470416A true CN113470416A (en) | 2021-10-01 |
| CN113470416B CN113470416B (en) | 2023-02-17 |
Family
ID=77865218
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010242254.XA Active CN113470416B (en) | 2020-03-31 | 2020-03-31 | System, method and storage medium for realizing parking space detection by using embedded system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113470416B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1725177A (en) * | 2004-07-21 | 2006-01-25 | 中国科学院软件研究所 | Built-in equipment integrated development system and using method thereof |
| CN1928816A (en) * | 2006-09-26 | 2007-03-14 | 武汉大学 | Model drive for embedded system software and component development method |
| CN104391791A (en) * | 2014-11-21 | 2015-03-04 | 北京经纬恒润科技有限公司 | Method and device for testing embedded control algorithm |
| CN106873566A (en) * | 2017-03-14 | 2017-06-20 | 东北大学 | A kind of unmanned logistic car based on deep learning |
| US20170193987A1 (en) * | 2015-12-30 | 2017-07-06 | Le Holdings (Beijing) Co., Ltd. | Speech recognition method and device |
| CN107016341A (en) * | 2017-03-03 | 2017-08-04 | 西安交通大学 | A kind of embedded real-time face recognition methods |
| CN108805070A (en) * | 2018-06-05 | 2018-11-13 | 合肥湛达智能科技有限公司 | A kind of deep learning pedestrian detection method based on built-in terminal |
| CN109086708A (en) * | 2018-07-25 | 2018-12-25 | 深圳大学 | A kind of parking space detection method and system based on deep learning |
| CN110175636A (en) * | 2019-05-08 | 2019-08-27 | 深圳欧翼思特科技有限公司 | A kind of Internet of Things deep neural network distribution differentiation inference system and method |
-
2020
- 2020-03-31 CN CN202010242254.XA patent/CN113470416B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1725177A (en) * | 2004-07-21 | 2006-01-25 | 中国科学院软件研究所 | Built-in equipment integrated development system and using method thereof |
| CN1928816A (en) * | 2006-09-26 | 2007-03-14 | 武汉大学 | Model drive for embedded system software and component development method |
| CN104391791A (en) * | 2014-11-21 | 2015-03-04 | 北京经纬恒润科技有限公司 | Method and device for testing embedded control algorithm |
| US20170193987A1 (en) * | 2015-12-30 | 2017-07-06 | Le Holdings (Beijing) Co., Ltd. | Speech recognition method and device |
| CN107016341A (en) * | 2017-03-03 | 2017-08-04 | 西安交通大学 | A kind of embedded real-time face recognition methods |
| CN106873566A (en) * | 2017-03-14 | 2017-06-20 | 东北大学 | A kind of unmanned logistic car based on deep learning |
| CN108805070A (en) * | 2018-06-05 | 2018-11-13 | 合肥湛达智能科技有限公司 | A kind of deep learning pedestrian detection method based on built-in terminal |
| CN109086708A (en) * | 2018-07-25 | 2018-12-25 | 深圳大学 | A kind of parking space detection method and system based on deep learning |
| CN110175636A (en) * | 2019-05-08 | 2019-08-27 | 深圳欧翼思特科技有限公司 | A kind of Internet of Things deep neural network distribution differentiation inference system and method |
Non-Patent Citations (1)
| Title |
|---|
| 储泽楠等: "基于嵌入式GPU的车流量检测系统", 《河南科技》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113470416B (en) | 2023-02-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6923960B2 (en) | A learning method and a learning device for detecting a parking space by utilizing the relationship between decision points and the regression result for the decision point in order to provide an automatic parking system, and a testing method and a testing device using the learning method and the learning device. | |
| CN111142402B (en) | Simulation scene construction method, device and terminal | |
| CN113470416B (en) | System, method and storage medium for realizing parking space detection by using embedded system | |
| US11703874B2 (en) | System and method for collection of performance data by a vehicle | |
| CN110789515B (en) | System and method for hardware validation in a motor vehicle | |
| CN111583335B (en) | Positioning system, positioning method, and non-transitory computer readable storage medium | |
| CN116626670B (en) | Automatic driving model generation method and device, vehicle and storage medium | |
| CN104268884A (en) | Lane departure early warning calibration system and method based on vehicle networking | |
| CN111210411B (en) | Method for detecting vanishing points in image, method for training detection model and electronic equipment | |
| CN115909815B (en) | Fusion detection method, device, equipment and storage medium based on multivariate data | |
| CN115610435B (en) | Method and device for predicting object driving intention, storage medium and electronic device | |
| CN117002530A (en) | Method and device for predicting future motion trail of vehicle and unmanned equipment | |
| CN116258174A (en) | System and method for detecting inference quality of deep neural network | |
| CN111177878B (en) | Derived simulation scene screening method, device and terminal | |
| CN115691109A (en) | Intelligent internet motorcade collaborative lane changing decision-making method | |
| CN114089677A (en) | Vehicle movement control method and device, vehicle and computer readable storage medium | |
| CN114708565A (en) | Intelligent driving scene recognition model creating method, device, equipment and storage medium | |
| AU2019206628B2 (en) | Monitoring system for autonomous vehicle operation | |
| CN112498337A (en) | Vehicle safety management method, device, system and storage medium | |
| CN111580499A (en) | Verification method and device for automatic navigation function and electronic equipment | |
| CN116968726B (en) | Memory parking method and device, vehicle and computer readable storage medium | |
| CN117668136A (en) | Highway infrastructure data processing method and device | |
| CN117197765A (en) | Image recognition method, device, electronic equipment and medium | |
| Lu et al. | An Improved YOLOv5s for Lane Line Detection | |
| US20230138969A1 (en) | Method and system for predicting the presence of environmental conditions in image data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |