CN210691083U

CN210691083U - Machine vision intelligent sensor

Info

Publication number: CN210691083U
Application number: CN201921975554.8U
Authority: CN
Inventors: 刘兵; 关腾腾
Original assignee: Tianjin Xinsong Intelligent Technology Co Ltd
Current assignee: Tianjin Xinsong Intelligent Technology Co Ltd
Priority date: 2019-11-15
Filing date: 2019-11-15
Publication date: 2020-06-05
Anticipated expiration: 2029-11-15

Abstract

The utility model relates to a machine vision intelligent sensor, including sensor housing and camera, be equipped with control module in the sensor housing, control module includes host processor module, coprocessor module, image sensor module, POE module and IO module, the utility model discloses an architecture and design bring network edge end efficient computing power and image processing ability, reduce the heavier and heavier calculation pressure of data center server, also let the edge end realize simultaneously that real-time complicated image processing becomes possible; the modular design of the utility model is convenient for maintenance, and other electrical elements are not easy to be damaged during maintenance; the IO module provides DC12-36V voltage for the coprocessor module to realize wide-voltage power supply; realize external power source and gigabit ethernet and all can give the power supply of main processor module and coprocessor module, guarantee the utility model discloses stable work. Furthermore, the utility model discloses the internal connection realizes the cableless connection, and the working property is more stable.

Description

Machine vision intelligent sensor

Technical Field

The utility model relates to the technical field of sensors, especially, relate to a machine vision intelligent sensor.

Background

The machine vision intelligent sensor is a sensor for realizing real-time image processing and deep learning inference at the network edge end. The sensor is a typical product which deeply fuses a computer technology and a sensor technology and is mainly applied to the fields of industry, traffic, education, security, retail and the like. The industrial application scenes comprise: the method comprises the following steps of defect detection, classification, statistics and auxiliary robot application of grabbing, carrying, moving and the like. The traffic application scene comprises the following steps: vehicle violation identification, driver and pedestrian behavior identification, intersection flow analysis and the like. The educational application scenario includes: intelligent invigilation, teaching and scientific research and the like. The security application scene comprises the following steps: face recognition, dangerous behavior recognition, and the like. The retail application scenario includes: the supermarket intelligent check, the intelligent cash register and the like.

At present, most of visual sensors in the market only complete simple image processing or image generation tasks, and complex image processing algorithms such as deep learning reasoning need to be completed by a USB host with a processor or a network host. The application mode not only brings the load pressure of the host end, but also greatly reduces the real-time performance of image processing due to the time delay brought by network transmission. In addition, the integral structure of the existing vision sensor is inconvenient to maintain, time-consuming and labor-consuming to maintain, and other electrical elements are easily damaged during maintenance.

Disclosure of Invention

The utility model discloses aim at solving prior art not enough, and provide a machine vision intelligent sensor.

The utility model discloses a realize above-mentioned purpose, adopt following technical scheme: a machine vision intelligent sensor comprises a sensor shell and a camera, and is characterized in that the camera is arranged on one side of the top of the sensor shell, a control module is arranged in the sensor shell, the control module comprises a main processor module, a coprocessor module, an image sensor module, a POE module and an IO module, the image sensor module is electrically connected with the camera, the image sensor module is electrically connected with the coprocessor module, the coprocessor module is electrically connected with the main processor module, the IO module is electrically connected with the coprocessor module, the IO module is connected with the POE module through gigabit Ethernet, the main processor module and the coprocessor module are electrically connected with the POE module, the POE module is used for providing electric energy for the main processor module and the coprocessor module, the power of external power supply DC12-36V is supplied power for the coprocessor module through the IO module, and the coprocessor module provides the electric energy of DC5V for the main processor module.

A heat dissipation plate is arranged between the main processor module and the coprocessor module.

The IO module and the coprocessor module are connected through a GPIO port, a UART port and a CAN port to transmit data.

The camera is an Ansenmei PYTHON5000 camera.

The co-processor module provides DC1.8V power to the image sensor module.

The main processor module and the coprocessor module transmit data in two directions through the PCI bus and the SPI bus.

The top surface of the sensor shell is provided with a plurality of radiating grooves, and the bottom surface of the sensor shell is provided with a plurality of radiating fins.

The utility model has the advantages that: the framework and the design of the utility model bring the high-efficiency computing power and the image processing power of the network edge end, reduce the increasingly heavy computing pressure of the data center server, and simultaneously make the edge end realize the real-time complex image processing possible; the modular design of the utility model is convenient for maintenance, and other electrical elements are not easy to be damaged during maintenance; the IO module provides DC12-36V voltage for the coprocessor module to realize wide-voltage power supply; realize external power source and gigabit ethernet and all can give the power supply of main processor module and coprocessor module, guarantee the utility model discloses stable work. Furthermore, the utility model discloses the internal connection realizes the cableless connection, and the working property is more stable.

Drawings

Fig. 1 is a schematic structural view of the present invention;

FIG. 2 is an exploded view of the structure of the present invention;

FIG. 3 is a connection diagram of the control module of the present invention;

FIG. 4 is a functional block diagram of a main processor module;

FIG. 5 is a functional block diagram of a coprocessor module;

in the figure: 1-a sensor housing; 2-a camera; 3-a main processor module; 4-coprocessor module; 5-an image sensor module; 6-POE module; 7-IO module; 8-a heat sink;

the following detailed description will be made in conjunction with embodiments of the present invention with reference to the accompanying drawings.

Detailed Description

The invention will be further explained with reference to the following figures and examples:

as shown in fig. 1-5, a machine vision intelligent sensor comprises a sensor housing 1 and a camera 2, wherein the camera 2 is installed on one side of the top of the sensor housing 1, a control module is arranged in the sensor housing 1, the control module comprises a main processor module 3, a coprocessor module 4, an image sensor module 5, a POE module 6 and an IO module 7, the image sensor module 5 is electrically connected with the camera 2, the image sensor module 5 is electrically connected with the coprocessor module 4, the coprocessor module 4 is electrically connected with the main processor module 3, the IO module 7 is electrically connected with the coprocessor module 4, the IO module 7 is connected with the POE module 6 through a gigabit ethernet, the main processor module 3 and the coprocessor module 4 are both electrically connected with the POE module 6, the POE module 6 is used for providing electric energy for the main processor module 3 and the coprocessor module 4, the power supply of the external power supply DC12-36V supplies power to the coprocessor module 4 through the IO module 7, and the coprocessor module 4 supplies power of DC5V to the main processor module 3.

A heat sink plate 8 is provided between the main processor module 3 and the coprocessor module 4.

The IO module 7 and the coprocessor module 4 are connected through a GPIO port, a UART port and a CAN port to transmit data.

The camera 2 is an Ansenmei PYTHON5000 camera.

Coprocessor module 4 supplies DC1.8V power to image sensor module 5.

The main processor module 3 and the coprocessor module 4 transmit data in two directions through the PCI bus and the SPI bus.

The top surface of the sensor shell 1 is provided with a plurality of heat dissipation grooves, and the bottom surface of the sensor shell 1 is provided with a plurality of heat dissipation fins.

The utility model discloses a design is piled up to the multimode piece, and each module adopts the design of no cable connection, and the heat dissipation form is the nature and leads cold. The main processor module 3 mainly provides a basic environment for software to run, and is a platform for all software to run. The coprocessor module 4 mainly completes input power supply conversion, image preprocessing, deep neural network acceleration and industrial field bus and IO control functions. The image sensor module 5 mainly realizes an imaging function. The POE module 6 realizes a POE + power supply function. The IO module 7 realizes connection of the external interface and the internal module. The specific connection schematic relationship is shown in fig. 3.

The main processor module 3 adopts Intel Apollo Lake E3950 as a core processor, which is a 4-core 64-bit processor, supports a master frequency of 1.6Ghz, integrates a GPU, has rich interfaces and PCIe with up to 6 lanes, supports a maximum 8G DDR3 memory, and has a maximum TDP of 12W. And the main processor module 3 integrates a 64GB onboard SATA hard disk and a 4GB onboard memory, and reserves a Type-C and MiniHDMI interface for manufacturing an operating system and debugging. One path of PCIex1 interface of E3950 is adopted to expand the Intel I210 network card, so that one path of gigabit network is realized. The other path PCIe x1 of E3950 is used for realizing MiniPCie interface for expanding the neural network accelerator VPU. The remaining PCIe x4 of E3950 are all reserved to board-to-board connectors interconnected with the coprocessor for communication with the FPGA on the coprocessor. In addition, the main processor module also supports the function of BIOS configuration FPGA. The functional block diagram of the main processor module 3 is shown in fig. 4.

The coprocessor module 4 adopts Intel CycloneV GT series FPGA as a core processor, and the processor is provided with a PCIe hard core IP, has rich IO and logic resources and supports OpenCL programming. The coprocessor module 4 adopts a hard core PCIe in the FPGA as a communication channel with the main processor module 3, and adopts a Verilog HDL language in the FPGA to realize an LVDS image data channel with the image processor module and CAN, UART and IO interfaces for communicating with external control nodes. Wherein the UART supports RS232 and RS485 protocols. The communication of coprocessor module 4 and external equipment all adopts the isolation mode to protect the utility model discloses. The CAN and UART interface isolation is realized by adopting a digital isolation power supply scheme, and the digital IO interface isolation is realized by adopting an optical coupling isolation scheme. The direct current inlet power supply supports 12-36V wide voltage input, supports automatic selection of a direct current power supply and a POE power supply, and selects the direct current power supply input by default. A functional block diagram of the coprocessor module 4 is shown in fig. 5.

The image processor module 5 adopts a CMOS sensor as a core device, the default parameters are 500 ten thousand pixels, the pixel size is 4.8um, the global shutter is adopted, the frame rate can reach 100FPS at most, 8 LVDS image data channels with the highest rate of 720Mbps are supported, the dynamic range is 60db, and the signal-to-noise ratio is 40 db.

The POE module 6 adopts a power transformer to realize a POE + protocol conforming to the IEEE 802.at standard, and can supply up to 30W of power. The design can simplify power supply wiring in distributed application, and gigabit network data transmission is provided while power supply is met. The POE module 6 is an optional module, and when the module is inserted, power supply through the gigabit ethernet can be realized, and when the module is not inserted, the gigabit ethernet is only used as a communication interface.

IO module 7 mainly adopts 3 12 core M12 connectors to provide the utility model discloses to outer physical interface to the board that adopts two 20 cores is provided to the board connector the utility model discloses external data channel.

The sensor shell 1 adopts a closed waterproof design, is made of aluminum alloy, and has good heat-conducting property and light weight.

The sensor shell 1 is made of aluminum alloy 5A05 light-weight material. The heat dissipation plate 8 in the sensor shell 1 is made of red copper, so that the heat dissipation area of the device is increased, and the heat dissipation is accelerated.

Sealing strips and screws are pressed between the sensor shells 1 to realize sealing and water proofing, and the M12 connector also selects a connector with the water proofing grade of IP 67. The fastening screw of the sensor shell 1 is made of stainless steel 1Cr18Ni9 Ti.

The upper surface, the lower surface, the front surface and the rear surface of the sensor shell 1 can be detached (seen from the M12 connector), so that the installation, debugging and maintenance are convenient.

The utility model discloses the during operation, image sensor module 6 passes through camera 2 and gathers the graphic data, gives coprocessor module 4 with information transfer and carries out the preliminary treatment to information, gives the information transfer after the preliminary treatment again and handles for main processor module 3, accomplishes the processing of information at the edge end, has reduced network bandwidth consumption, also very big must reduce server end pressure.

If in industrial application, the pictures collected by a camera which is deployed on a production line and used for identifying and classifying need to be transmitted to an image processing host computer through a network, after the host computer finishes algorithm identification, information is transmitted to an industrial robot through a field bus to finish grabbing and sorting, and the efficiency of the production line can be reduced due to uncertain delay caused by network transmission in the process, and system oscillation is easily caused. The utility model discloses because gather image data in real time to integrated image processing algorithm and industrial control field bus, consequently can by the utility model discloses guide the robot in real time and accomplish and snatch the letter sorting. In traffic application, a large number of pictures shot by thousands of traffic violation cameras in a city every day are transmitted back to the traffic management center, are judged by the machine system firstly, and are rechecked manually, so that the process not only occupies a large amount of network bandwidth, but also aggravates the storage and calculation pressure of the traffic management center server, and simultaneously brings a large amount of manual rechecking work. And adopt the technical scheme of the utility model, because each sensor is dispersed to the pressure that will calculate and save, the behavior recognition violating the regulations has been accomplished at the edge end, and what transmit back traffic management center is the data violating the regulations that have the discrimination information, when guaranteeing the data accuracy, has reduced network bandwidth consumption greatly, also very big must reduce server end pressure. In addition in the control of traffic light is used the utility model discloses the scheme is real-time according to the utility model provides a flow analysis algorithm, by the utility model discloses come the switching frequency of automatic control traffic light, reduce the contradiction that the manual control limitation brought promptly, also improved the real-time of control greatly.

The present invention has been described above with reference to the accompanying drawings, and it is obvious that the present invention is not limited by the above embodiments, and various improvements made by the method concept and technical solution of the present invention or directly applied to other occasions without improvement are all within the protection scope of the present invention.

Claims

1. A machine vision intelligent sensor comprises a sensor shell (1) and a camera (2), and is characterized in that the camera (2) is installed on one side of the top of the sensor shell (1), a control module is arranged in the sensor shell (1), the control module comprises a main processor module (3), a coprocessor module (4), an image sensor module (5), a POE module (6) and an IO module (7), the image sensor module (5) is electrically connected with the camera (2), the image sensor module (5) is electrically connected with the coprocessor module (4), the coprocessor module (4) is electrically connected with the main processor module (3), the IO module (7) is electrically connected with the coprocessor module (4), the IO module (7) is connected with the POE module (6) through a gigabit Ethernet, the main processor module (3) and the coprocessor module (4) are both electrically connected with the POE module (6), the POE module (6) is used for providing electric energy for the main processor module (3) and the coprocessor module (4), the power supply of an external power supply DC12-36V supplies power for the coprocessor module (4) through the IO module (7), and the coprocessor module (4) provides electric energy of DC5V for the main processor module (3).

2.A machine vision smart sensor as claimed in claim 1, characterised in that a heat sink (8) is provided between the main processor module (3) and the co-processor module (4).

3. The machine vision intelligent sensor of claim 1, wherein the IO module (7) and the coprocessor module (4) are connected through a GPIO port, a UART port and a CAN port to transmit data.

4. A machine-vision smart sensor as claimed in claim 1, characterized in that the co-processor module (4) supplies DC1.8V power to the image sensor module (5).

5. The machine vision intelligent sensor of claim 1, wherein the main processor module (3) and the coprocessor module (4) transmit data bidirectionally through a PCI bus and an SPI bus.

6. The machine vision intelligent sensor according to claim 1, wherein the top surface of the sensor housing (1) is provided with a plurality of heat dissipation grooves, and the bottom surface of the sensor housing (1) is provided with a plurality of heat dissipation fins.