Beidou edge computing equipment and method for acquiring inspection facilities based on equipment
Technical Field
The invention relates to the technical field of road inspection, in particular to Beidou edge computing equipment and a method for acquiring inspection facilities based on the equipment.
Background
The new generation technology innovation represented by unmanned is that the quality and the performance requirements of the space data represented by the high-precision map are greatly improved, and the requirements of the space data acquisition accuracy, the update frequency and the like are improved by several grades.
The collection of spatial data is heavily dependent on scale effects, and a systematic answer to cost questions is required for how new national infrastructure maps are provided. At present, data acquisition represented by map field industry and interior industry still takes manual acquisition as a main means, and the large-scale dependence on manual acquisition becomes a brake of the spatial information industry.
The space information data is acquired by using artificial intelligence and becomes the industry leading direction, most of using modes in the market are that video data terminals are used for acquiring and uploading the space information data to an AI server in an offline mode, data identification is carried out in a server background, and due to the limitation of the current mobile network conditions, real-time identification and real-time update cannot be achieved for information acquisition, so that the requirements on the situation of map data are poor.
For this reason, there is an urgent need in the art to develop a device capable of overcoming the above technical problems existing in the prior art.
Disclosure of Invention
The invention aims to provide Beidou edge computing equipment, which can solve the technical problems of high cost, poor instantaneity and poor normalization of the existing road inspection by mainly identifying and formulating targets through manual training.
The invention aims to provide a method for acquiring inspection facilities based on Beidou edge computing equipment, which can solve the technical problems of high cost, poor real-time performance and poor standardization of the existing road inspection by mainly identifying and formulating targets through manual training.
The invention provides Beidou edge computing equipment which comprises an AI visual unit, a Beidou positioning module, an NPU processing unit, a CPU processing unit and a data transmission unit,
the AI visual unit is used for acquiring standard video stream data in real time and transmitting the standard video stream data to the NPU processing unit;
the NPU processing unit is used for processing the standard video stream data by running different neural networks so as to realize the identification and extraction of spatial data information in vision and transmitting the identified spatial data information to the CPU processing unit;
the Beidou positioning module is used for receiving the space position information in real time;
the CPU processing unit is used for receiving the space position information transmitted by the Beidou positioning module in real time, and matching the space data information with the space position information to form real-time normalized space data information;
the data transmission unit is used for transmitting the normalized spatial data information in the CPU processing unit to the background server.
Preferably, the NPU processing unit acquires and identifies spatial data information in units of milliseconds, and the identification frame number is smaller than the input video stream frame number.
Preferably, the CPU processing unit acquires the spatial position information of the spatial data image of each frame based on a linear difference algorithm.
Preferably, the AI vision unit comprises a vision sensor and ISP processing unit,
the visual sensor is used for acquiring original video stream data;
the ISP processing unit is used for preprocessing images in the video stream acquired by the visual sensor to acquire standard video stream data.
Preferably, the ISP processing unit performs anti-shake and light intensity adjustment processing on the image in the video stream acquired by the vision sensor.
The invention also provides a method for acquiring the inspection facility based on the Beidou edge computing device, which comprises the following steps:
the AI visual unit acquires the standard video stream data in real time and transmits the standard video stream data to the NPU processing unit;
the NPU processing unit processes the standard video stream data by running different neural networks so as to realize the identification and extraction of spatial data information in vision, and transmits the identified spatial data information to the CPU processing unit;
the Beidou positioning module is used for receiving the space position information in real time;
the CPU processing unit receives the spatial position information transmitted by the Beidou positioning module in real time, and matches the spatial data information with the spatial position information to form real-time normalized spatial data information;
and the normalized spatial data information is returned to the background server in real time through a wireless network.
Preferably, the NPU processing unit processes the standard video stream data by running different neural networks, so as to implement identification and extraction of spatial data information in vision, and transmits the identified spatial data information to the CPU processing unit, and the NPU processing unit includes:
acquiring real-time standard video stream data information;
the standard video stream data are identified and tracked by running different neural networks, so that continuous frame-by-frame tracking of the inspection facility in the identification process of the space data information is realized;
acquiring space data information;
the spatial data information is transmitted to the CPU processing unit.
Preferably, the CPU processing unit obtains the spatial position information through a linear difference algorithm, locates the spatial position of the inspection facility in each frame of spatial data image, and matches the spatial position information processed through the linear difference algorithm with the spatial data information to form real-time normalized spatial data information.
Preferably, the AI visual unit is provided with an HTTP request interface for implementing decoupling of the beidou edge computing device acquisition device function mechanism.
Preferably, the AI visual unit obtains video stream data with a resolution of 1080P at an efficiency of 30 frames per second or more and performs real-time transmission;
the NPU processing unit keeps the processing speed of more than 5 frames per second for video stream data processing.
Compared with the prior art, the Beidou edge computing device and the method for acquiring the inspection facility based on the device have the following beneficial effects:
according to the invention, through Beidou positioning and geofence technologies, images of the inspection points needing to be mainly confirmed in road inspection are extracted efficiently at low cost, and the images are uploaded to the background in real time to carry out remote road inspection identification in a manual auxiliary mode through an AI image identification technology.
According to the invention, the recognition result is textized and normalized through the edge calculation method and the real-time recognition of the Beidou positioning at the field production side, the processed result data is small in size and low in dependence on the bandwidth of the mobile digital network, and the processed result data can be automatically uploaded to a background server in real time.
According to the invention, new infrastructure technologies such as Beidou high-precision positioning, artificial intelligence and 5G communication are used for replacing an artificial part in traditional space data acquisition, so that space data acquisition is thoroughly automated and AI is realized, a large number of non-professional personnel can complete professional space data acquisition in a crowdsourcing mode without training, and a mode for realizing digitalization of national road facilities in a low-cost high-quality large-scale mode is systematically provided.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a frame of a Beidou edge computing device of the present invention;
fig. 2 is a flow chart of a method for acquiring a patrol facility based on the Beidou edge computing device.
Summarizing the reference numerals:
1. AI visual unit 2, NPU processing unit 3, big dipper positioning module
4. CPU processing unit 5, data transmission unit 6, and background server
Detailed Description
The objects, technical solutions and advantages of the present invention will become more apparent by the following detailed description of the present invention with reference to the accompanying drawings. It should be understood that the description is only illustrative and is not intended to limit the scope of the invention. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.
As shown in fig. 1, the Beidou edge computing device provided by the invention comprises an AI visual unit 1, a Beidou positioning module 3, an NPU processing unit 2, a CPU processing unit 4 and a data transmission unit 5,
the AI visual unit 1 is used for acquiring standard video stream data in real time and transmitting the standard video stream data to the NPU processing unit 2;
the NPU processing unit 2 is configured to process the standard video stream data by running different neural networks, so as to implement identification and extraction of spatial data information in vision, and transmit the identified spatial data information to the CPU processing unit 4;
the Beidou positioning module 3 is used for receiving the space position information in real time;
the CPU processing unit 4 is used for receiving the space position information transmitted by the Beidou positioning module 3 in real time, acquiring the space position information of each frame of space data image based on a linear difference algorithm, and matching the space data information with the space position information to form real-time normalized space data information;
the data transmission unit 5 is used for transmitting normalized spatial data information (structured text) in the CPU processing unit 4 to the background server 6.
The Beidou edge computing device is integrated, the hardware part of the Beidou edge computing device has no special requirement on the installation of a vehicle, the central control panel of the vehicle is not required to be changed, and the Beidou edge computing device can be executed only by supplying power to a cigar lighter. Or if the electric vehicle is adopted for road facility inspection, the Beidou edge computing device can be powered through the charger, so that portable installation and use are realized, a wireless network of the electric vehicle can be connected with a mobile phone WiFi, a hotspot or Bluetooth, namely, the existing mobile device is fully utilized to realize professional space data acquisition in a crowdsourcing mode, and a mode of realizing national road facility digitization in a low-cost high-quality large-scale mode is systematically provided.
The Beidou edge computing device is a simplified design capable of realizing necessary functions, so that the AI visual unit 1 has a series of advantages in price (thousands of costs are far lower than tens of thousands of costs of the same class in the industry), stability, power consumption (7 hours for single charger), weight (200 g) and the like. On the connection of a transmission layer, the AI visual unit 1 realizes data transmission communication by connecting with a hot spot of an Android central control/mobile phone, and on a software protocol, the AI visual unit 1 provides an HTTP request interface, and the Android central control/mobile phone initiates a Get HTTP request to realize equipment decoupling to the greatest extent.
In the running process of a vehicle, an AI visual unit 1 in the Beidou edge computing device obtains video stream data with the resolution of 1080P with the efficiency of more than 30 frames per second and transmits the video stream data in real time; meanwhile, the NPU processing unit 2 holds a processing speed of 5 frames per second or more for video stream data processing.
Preferably, the NPU processing unit 2 acquires and recognizes spatial data information in units of milliseconds, and the recognition frame number is smaller than the input video stream frame number.
In a further embodiment of the present invention, the AI visual unit 1 comprises a visual sensor for acquiring raw video stream data and an ISP processing unit; the ISP processing unit is used for preprocessing images in the video stream acquired by the vision sensor to acquire standard video stream data.
Preferably, the ISP processing unit performs anti-shake and light intensity adjustment processing on the image in the video stream acquired by the video sensor.
As shown in fig. 2, the invention further provides a method for acquiring the inspection facility based on the Beidou edge computing device, which comprises the following steps:
the AI visual unit 1 acquires the standard video stream data in real time and transmits the standard video stream data to the NPU processing unit 2;
the NPU processing unit 2 processes the standard video stream data by running different neural networks to realize the identification and extraction of spatial data information in the vision, and transmits the identified spatial data information to the CPU processing unit 4;
the Beidou positioning module 3 is used for receiving the space position information in real time;
the CPU processing unit 4 receives the space position information transmitted by the Beidou positioning module 3 in real time, and matches the space data information with the space position information to form real-time normalized space data information;
normalized spatial data information is returned to the background server 6 in real time through the wireless network.
According to the invention, through an edge calculation method, an AI image recognition technology and Beidou positioning and field real-time recognition at the field production side, recognition results are textified and normalized, the processed result data is small in size and low in dependence on mobile digital network bandwidth, and can be automatically uploaded to the background server 6 in real time, so that remote road inspection facility recognition is realized, and a mode for realizing national road facility digitization in a low-cost high-quality large-scale manner is systematically provided.
In a further embodiment of the present invention, the NPU processing unit 2 processes the standard video stream data by running different neural networks to implement identification and extraction of spatial data information in the vision, and transmits the identified spatial data information to the CPU processing unit 4, which includes:
acquiring real-time standard video stream data information;
the standard video stream data are identified and tracked by running different neural networks, so that continuous frame-by-frame tracking of the inspection facility in the identification process of the spatial data information is realized;
acquiring space data information;
the spatial data information is transmitted to the CPU processing unit 4.
Considering that the video stream acquired by the AI visual unit 1 is continuously input, the NPU processing unit 2 loaded on the beidou edge computing device needs to take into account both the functions of "recognition" and "tracking". The identification function uses yolo V3 neural network, the tracking function uses twin neural network, and the two neural networks run simultaneously to realize continuous tracking of the identification object frame by frame in the identification process.
The AI visual unit 1 is provided with an HTTP request interface for realizing decoupling of a Beidou edge computing device acquisition device function mechanism. This arrangement can minimize the dependence of the AI visual unit 1 on other hardware modules, and thus can focus on implementing the visual algorithm.
The CPU processing unit 4 obtains the spatial position information through the linear difference algorithm, locates the spatial position of the inspection facility in each frame of spatial data image, and matches the spatial position information processed through the linear difference algorithm with the spatial data information to form real-time normalized spatial data information.
Specifically, the CPU processing unit 4 may employ a 2TOPS power AI chip module, where 2TOPS is the highest power achievable by the current mainstream single chip, and the tailored mainstream vision-related neural network may be already implemented in a commercialized manner.
Due to the implementation of the tracking algorithm of the video stream, the Beidou edge computing device can continuously track the identification objects (patrol facilities). The time when the vehicle passes the identifier can be deduced from the time when the identifier has disappeared. Meanwhile, the equipment can be adjusted, the stability of the visual input frame number of the equipment is ensured, the spatial position of the equipment when each frame of spatial data image is obtained by a linear interpolation mode on the premise of continuously obtaining Beidou signals, and then the corresponding spatial position of the artificial intelligent recognition result is obtained. Because the decimeter level Beidou signals are generated every second, the vision acquisition frame position after the linear difference value can also reach the decimeter level precision under most conditions. The method has the advantage of meeting business requirements on the basis of low cost without relying on high-cost laser radars and the like.
According to the invention, the artificial intelligence, the edge calculation and the Beidou are positioned on a hardware layer to be integrated, so that a universal platform for collecting the next-generation space data is produced. Under the same hardware architecture, the method covers the generalized space information automatic acquisition fields of high-precision maps, road inspection, grid member inspection, traffic police law enforcement, urban management law enforcement and the like through continuous iteration of algorithms. The identification result is textized and normalized by combining the AI identification technology with Beidou high-precision positioning, the processed result data has small size and low dependence on the bandwidth of the mobile digital network, and the processed result data can be automatically uploaded to the background server 6 in real time, so that the identification of remote road inspection facilities is realized.
It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explanation of the principles of the present invention and are in no way limiting of the invention. Accordingly, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention should be included in the scope of the present invention. Furthermore, the appended claims are intended to cover all such changes and modifications that fall within the scope and boundary of the appended claims, or equivalents of such scope and boundary.