CN209993011U - Low-power-consumption online thermal infrared target recognition device - Google Patents
Low-power-consumption online thermal infrared target recognition device Download PDFInfo
- Publication number
- CN209993011U CN209993011U CN201921324178.6U CN201921324178U CN209993011U CN 209993011 U CN209993011 U CN 209993011U CN 201921324178 U CN201921324178 U CN 201921324178U CN 209993011 U CN209993011 U CN 209993011U
- Authority
- CN
- China
- Prior art keywords
- unit
- thermal infrared
- module
- image
- preprocessing unit
- 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.)
- Active
Links
Images
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The utility model relates to a thermal infrared night vision control technical field, in particular to thermal infrared target identification device. A low-power consumption on-line thermal infrared target recognition device comprises: the system comprises a thermal infrared detector, a preprocessing unit, an identification unit, a battery unit, a wireless transmission unit and a rear-end receiving device; the utility model collects the infrared image by the thermal infrared detector, processes the infrared image by the preprocessing unit, when the preprocessing unit does not detect the moving object, the identification unit is in a dormant state, and after the moving object is detected, the identification unit is awakened to provide enough computing resources for identifying the object type and the position; the utility model discloses furthest reduces the equipment consumption, has realized the full autonomy artificial intelligence multi-target type's of low-cost low-power consumption discernment and local storage, video transmission function.
Description
Technical Field
The utility model relates to a thermal infrared night vision control technical field, in particular to thermal infrared target identification device.
Background
The infrared thermal imaging is used for imaging objects in a fully passive mode, has high penetration capacity to rain and fog, can monitor various targets such as people, vehicles, wild animals and the like day and night, and is widely applied to the fields of fire prevention, night vision and security.
The infrared thermal imager uses an infrared detector and an optical imaging objective lens to receive an infrared radiation energy distribution pattern of a detected target and reflect the infrared radiation energy distribution pattern to a photosensitive element of the infrared detector, so as to obtain an infrared thermal image. The existing thermal infrared imager can only meet the imaging requirement and does not have the intelligent self-identification function.
With the rapid development of computer vision, the target detection technology is widely applied in many fields, but the traditional algorithm has high error report and cannot meet the actual accurate identification requirement; the deep learning algorithm has extremely high requirements on computing power, a platform supported by the current deep learning algorithm mainly comprises a GPU of companies such as Invitta and the like, an ARM + Mali GPU of Ruihe-Ching micro company, an FPGA of Xilinx company and the like, and the chip has the defects of high power consumption, large chip volume, high price and the like.
SUMMERY OF THE UTILITY MODEL
The utility model aims at: aiming at the defects of the prior art, the low-power-consumption online thermal infrared target recognition device is provided.
The utility model discloses a technical scheme is: a low-power consumption on-line thermal infrared target recognition device comprises: the system comprises a thermal infrared detector, a preprocessing unit, an identification unit, a battery unit, a wireless transmission unit and a rear-end receiving device;
the recognition unit includes: the target identification chip and the power management circuit are connected with each other;
the thermal infrared detector is in signal connection with the preprocessing unit, the preprocessing unit is in signal connection with the target identification chip and the power management circuit respectively, the preprocessing unit is in signal connection with the rear-end receiving device through the wireless transmission unit, and the battery unit is in signal connection with the thermal infrared detector, the preprocessing unit and the power management circuit.
In the scheme, the thermal infrared detector is used for collecting an original image; the preprocessing unit is used for preprocessing an original image and identifying whether a moving target appears in the image, when the moving target appears, the preprocessing unit sends a wake-up signal to the power management circuit, and the power management circuit is connected with the battery unit and the target identification chip; the target identification chip is used for identifying the specific category and position of the moving target in the image; the wireless transmission unit is used for transmitting the specific category, the position information and the image of the moving target to the rear-end receiving equipment; the rear-end receiving device is used for displaying the image which is superposed with the specific category and position information of the moving object.
Further, the preprocessing unit specifically includes: the device comprises an AD acquisition module, a storage module, an image recognition module and a communication interface module; thermal infrared detector and AD acquisition module), the AD acquisition module and the image identification module are in signal connection, the image identification module, the storage module and the communication interface module are in signal connection respectively, the storage module and the communication interface module are in signal connection, and the communication interface module, the target identification chip and the power management circuit are in signal connection respectively.
Furthermore, the preprocessing unit adopts an ARM chip, and the target identification chip adopts an NPU chip; the target identification part in the device adopts a double-chip scheme of combining an ARM core chip and a NPU chip special for deep learning reasoning, when a moving object is not detected, the NPU chip is always in a low power consumption state, and the power consumption of the whole machine is less than 1.2W. After the ARM chip detects the moving object, the NPU chip is awakened to identify the object type and the position of the moving object, and the power consumption of the whole machine is 3W. The NPU processing part can provide the deep learning reasoning capability of 2TOPS and meet the computational resources of online real-time object recognition.
Furthermore, in order to meet the requirement of autonomously executing an object identification task in the field for a long time, the preprocessing unit, the identification unit, the battery unit and the wireless transmission unit are arranged in the waterproof shell in a connector assembly mode, and the thermal infrared detector and the waterproof shell are sealed by a rubber ring; the battery unit is powered by a battery with 20000 mAh.
The utility model discloses a another technical scheme is: the working method of the low-power-consumption online thermal infrared target recognition device is based on the recognition device and comprises the following steps:
A. the battery unit supplies power to the thermal infrared detector, the preprocessing unit and the power management circuit;
B. the method comprises the following steps that a thermal infrared detector collects original images and sends the original images to a preprocessing unit;
C. the preprocessing unit preprocesses the original image to obtain an infrared image with obvious detail information, identifies whether a moving object appears in the infrared image with the obvious detail information, and sends a wake-up signal to the power management circuit if the moving object appears;
D. after receiving the wake-up signal, the power management circuit is connected with the target identification chip and the battery unit, the target identification chip enters a target identification mode, and the infrared image with obvious detail information in the preprocessing unit is called to identify the type and the position of the moving object; after identification, the target identification chip sends the type and position information of the moving object to the preprocessing unit;
E. the preprocessing unit matches and superposes the received type and position information of the moving object with the infrared image with obvious detail information, and sends the infrared image with obvious detail information superposed with the type and position information of the moving object to the rear-end receiving equipment through the wireless transmission unit;
F. the rear-end receiving device displays an infrared image in which the type of the moving object and the detailed information of the position information are recognized are obvious.
Specifically, the method comprises the following steps: in the step C, the method for preprocessing the original image comprises the following steps: firstly, format conversion is carried out on an original infrared image, non-uniform correction and blind pixel compensation are carried out on the infrared image after format conversion to obtain a corrected image, and then super-resolution image enhancement is carried out on the corrected image, so that the infrared image with obvious detail information is obtained.
And step C, separating the moving object in the infrared image with obvious detail information from the background by adopting a background dynamic modeling mode, thereby realizing the identification of whether the moving object appears.
And D, identifying the type and the position of the moving object in the infrared image with obvious detail information by adopting a depth separable convolution neural network inference weight and infrared Hog characteristic weighted summation mode input by a monochromatic two-dimensional matrix.
Further, when the preprocessing unit includes: AD collection module, storage module, image recognition module and communication interface module:
the step C comprises the following steps:
C1. the AD acquisition module performs format conversion on an original image sent by the thermal infrared detector;
C2. the image recognition module preprocesses the original image after format conversion to obtain an infrared image with obvious detail information;
C3. the image identification module identifies whether a moving object appears in two or more adjacent frames of images in the infrared image sequence with obvious detail information, if the moving object appears, the image identification module sends the two or more adjacent frames of images containing the moving object to the storage module for storage, and simultaneously sends a wake-up signal to the power management circuit through the communication interface module;
the step D comprises the following steps:
D1. after receiving the wake-up signal, the power management circuit switches on the target identification chip and the battery unit, and the target identification chip enters a target identification mode;
D2. the target identification chip calls the infrared image with obvious detail information stored in the storage module through the communication interface module, identifies the type and the position of the moving object in the image and obtains the type and the position information of the moving object;
D3. the target identification chip sends the type and the position information of the moving object to the storage module through the communication interface module for storage;
the step E comprises the following steps:
E1. the image identification module calls the infrared image with obvious type, position information and detail information of the moving object in the storage module for matching, and superposes the type and position information of the moving object and the infrared image with obvious corresponding detail information;
E2. the image identification module sends the infrared image with the obvious detail information of the type and the position information of the moving object superimposed to the rear-end receiving equipment through the communication interface module and the wireless transmission unit.
Has the advantages that: the utility model adopts the design mode of the heterogeneous processing platform, when the preprocessing unit does not detect the moving object, the identification unit is in a dormant state, and after the moving object is detected, the identification unit is awakened to provide enough computing resources for identifying the object type and the position; the utility model discloses furthest reduces the equipment consumption, has realized the full autonomy artificial intelligence multi-target type's of low-cost low-power consumption discernment and local storage, video transmission function. The utility model discloses can be applied to all-weather indoor and field independently remote monitoring field. The utility model discloses an operating method can realize the real-time degree of depth of edge end learning object type identification ability and independently based on the dynamic power consumption control ability of incident, and the misstatement that significantly reduces improves the target recognition rate.
Drawings
Fig. 1 is a block diagram of the embodiment 1 of the present invention;
fig. 2 is a block diagram of embodiment 2 of the present invention;
in the figure: the system comprises a 1-thermal infrared detector, a 2-preprocessing unit, a 2.1-AD acquisition module, a 2.2-storage module, a 2.3-image recognition module, a 2.4-communication interface module, a 3-recognition unit, a 3.1-target recognition chip, a 3.2-power management circuit, a 4-battery unit, a 5-wireless transmission unit and a 6-rear-end receiving device.
Detailed Description
the recognition unit 3 includes: a target identification chip 3.1 and a power management circuit 3.2 connected to each other;
the thermal infrared detector 1 is in signal connection with the preprocessing unit 2, the preprocessing unit 2 is in signal connection with the target identification chip 3.1 and the power management circuit 3.2 respectively, the preprocessing unit 2 is in signal connection with the rear-end receiving device 6 through the wireless transmission unit 5, and the battery unit 4 is in signal connection with the thermal infrared detector 1, the preprocessing unit 2 and the power management circuit 3.2.
In the scheme, the thermal infrared detector 1 is used for collecting an original image; the preprocessing unit 2 is used for preprocessing an original image and identifying whether a moving target appears in the image, when the moving target appears, the preprocessing unit 2 sends a wake-up signal to the power management circuit 3.2, and the power management circuit 3.2 is connected with the battery unit 4 and the target identification chip 3.1; the target identification chip 3.1 is used for identifying the specific category and position of the moving target in the image; the wireless transmission unit 5 is used for transmitting the specific category, the position information and the image of the moving object to the rear-end receiving equipment 6; the rear-end receiving device 6 is used for displaying an image on which specific category and position information of the moving object are superimposed.
In this example, the thermal infrared detector 1 has a resolution of 320 × 240, and can realize night vision observation. The preprocessing unit 2 adopts an ARM chip, and the target identification chip 3.1 adopts an NPU chip; the target identification part in the device adopts a double-chip scheme of combining an ARM core chip and a NPU chip special for deep learning reasoning, when a moving object is not detected, the NPU chip is always in a low power consumption state, and the power consumption of the whole machine is less than 1.2W. After the ARM chip detects the moving object, the NPU chip is awakened to identify the object type and the position of the moving object, and the power consumption of the whole machine is 3W. The NPU processing part can provide the deep learning reasoning capability of 2TOPS, can reach the real-time speed of each frame of image of 60ms, and meets the computational resources of online real-time object identification.
Example 2, referring to fig. 2, on the basis of example 1, the pretreatment unit 2 is further defined:
the pretreatment unit 2 specifically includes: the system comprises an AD acquisition module 2.1, a storage module 2.2, an image recognition module 2.3 and a communication interface module 2.4; the thermal infrared detector 1 is in signal connection with the AD acquisition module 2.1, the AD acquisition module 2.1 is in signal connection with the image recognition module 2.3, the image recognition module 2.3 is in signal connection with the storage module 2.2 and the communication interface module 2.4 respectively, the storage module 2.2 is in signal connection with the communication interface module 2.4, and the communication interface module 2.4 is in signal connection with the target recognition chip 3.1 and the power management circuit 3.2 respectively.
A. the battery unit 4 supplies power to the thermal infrared detector 1, the preprocessing unit 2 and the power management circuit 3.2;
B. the thermal infrared detector 1 collects an original image and sends the original image to the preprocessing unit 2;
C. the preprocessing unit 2 preprocesses the original image to obtain an infrared image with obvious detail information, identifies whether a moving object appears in the infrared image with the obvious detail information, and if the moving object appears, the preprocessing unit 2 sends a wake-up signal to the power management circuit 3.2;
D. after receiving the wake-up signal, the power management circuit 3.2 switches on the target identification chip 3.1 and the battery unit 4, the target identification chip 3.1 enters a target identification mode, and calls an infrared image with obvious detail information in the preprocessing unit 2to identify the type and position of a moving object; after identification, the target identification chip 3.1 sends the type and position information of the moving object to the preprocessing unit 2;
E. the preprocessing unit 2 matches and superimposes the received type and position information of the moving object with the infrared image with obvious detail information, and sends the infrared image with obvious detail information superimposed with the type and position information of the moving object to the rear-end receiving equipment 6 through the wireless transmission unit 5;
F. the rear-end receiving device 6 displays an infrared image in which the detailed information identifying the type of the moving object and the position information is clear.
In the above scheme, specifically:
in the step C, the method for preprocessing the original image comprises the following steps: firstly, format conversion is carried out on an original infrared image, non-uniform correction and blind pixel compensation are carried out on the infrared image after format conversion to obtain a corrected image, and then super-resolution image enhancement is carried out on the corrected image, so that the infrared image with obvious detail information is obtained.
In the step C, whether a moving object appears in the infrared image with obvious detail information is identified by combining ViBE and GMM, namely, the moving object in the infrared image with obvious detail information is separated from the background by adopting a background dynamic modeling mode, so that the identification of whether the moving object appears is realized.
And D, identifying the type and the position of the moving object in the infrared image with obvious detail information by adopting a method of combining a monochrome input convolutional neural network and the Hog characteristic, namely identifying the type and the position of the moving object in the infrared image with obvious detail information by adopting a monochrome two-dimensional matrix input depth separable convolutional neural network inference weight and infrared Hog characteristic weighted summation mode.
A. the battery unit 4 supplies power to the thermal infrared detector 1, the preprocessing unit 2 and the power management circuit 3.2;
B. the thermal infrared detector 1 collects an original image and sends the original image to the preprocessing unit 2;
C1. the AD acquisition module 2.1 carries out format conversion on the original image sent by the thermal infrared detector 1;
C2. the image recognition module 2.3 preprocesses the original image after format conversion to obtain an infrared image with obvious detail information;
C3. the image recognition module 2.3 recognizes whether a moving object appears in two or more adjacent frames of images in the infrared image sequence with obvious detail information, if the moving object appears, the image recognition module 2.3 sends the two or more adjacent frames of images containing the moving object to the storage module 2.2 for storage, and simultaneously sends a wake-up signal to the power management circuit 3.2 through the communication interface module 2.4;
D1. after receiving the wake-up signal, the power management circuit 3.2 connects the target identification chip 3.1 and the battery unit 4, and the target identification chip 3.1 enters a target identification mode;
D2. the target identification chip 3.1 calls the infrared image with obvious detail information stored in the storage module 2.2 through the communication interface module 2.4 to identify the type and position of the moving object in the image to obtain the type and position information of the moving object;
D3. the target identification chip 3.1 sends the type and position information of the moving object to the storage module 2.2 for storage through the communication interface module 2.4;
E1. the image identification module 2.3 calls the infrared image with obvious type, position information and detail information of the moving object in the storage module 2.2 for matching, and superposes the type and position information of the moving object and the corresponding infrared image with obvious detail information;
E2. the image identification module 2.3 sends the infrared image with the obvious detail information of the type and the position information of the moving object to the rear-end receiving equipment 6 through the communication interface module 2.4 and the wireless transmission unit 5;
F. the rear-end receiving device 6 displays an infrared image in which the detailed information identifying the type of the moving object and the position information is clear.
Although the invention has been described in detail with respect to the general description and the specific embodiments, it will be apparent to those skilled in the art that modifications and improvements can be made based on the invention. Therefore, such modifications and improvements are intended to be within the scope of the invention as claimed.
Claims (4)
1. A low-power consumption on-line thermal infrared target recognition device comprises: thermal infrared detector (1), preprocessing unit (2), recognition cell (3), battery unit (4), wireless transmission unit (5) and rear end receiving equipment (6), its characterized in that:
the identification unit (3) comprises: a target identification chip (3.1) and a power management circuit (3.2) connected with each other;
thermal infrared detector (1) with signal connection is established to preprocessing unit (2), preprocessing unit (2) with target identification chip (3.1) signal connection is established respectively to power management circuit (3.2), preprocessing unit (2) pass through wireless transmission unit (5) with rear end receiving equipment (6) establish signal connection, battery unit (4) with thermal infrared detector (1) preprocessing unit (2) power management circuit (3.2) establish the connection.
2. The low-power online thermal infrared target recognition device of claim 1, wherein: the pre-processing unit (2) comprises: the device comprises an AD acquisition module (2.1), a storage module (2.2), an image recognition module (2.3) and a communication interface module (2.4); thermal infrared detector (1) with AD collection module (2.1)) establishes signal connection, AD collection module (2.1) with image recognition module (2.3) establishes signal connection, image recognition module (2.3) with storage module (2.2) communication interface module (2.4) establishes signal connection respectively, storage module (2.2) with communication interface module (2.4) establishes signal connection, communication interface module (2.4) with target identification chip (3.1) power management circuit (3.2) establishes signal connection respectively.
3. A low power consumption on-line thermal infrared target recognition device as claimed in claim 1 or 2, wherein: the preprocessing unit (2) adopts an ARM chip, and the target identification chip (3.1) adopts an NPU chip.
4. A low power consumption on-line thermal infrared target recognition device as claimed in claim 1 or 2, wherein: the preprocessing unit (2), the identification unit (3), the battery unit (4) and the wireless transmission unit (5) are arranged in a waterproof shell in a connector assembly mode, and the thermal infrared detector (1) and the waterproof shell are sealed by rubber rings; the battery unit (4) is powered by a battery with 20000 mAh.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201921324178.6U CN209993011U (en) | 2019-08-15 | 2019-08-15 | Low-power-consumption online thermal infrared target recognition device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201921324178.6U CN209993011U (en) | 2019-08-15 | 2019-08-15 | Low-power-consumption online thermal infrared target recognition device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN209993011U true CN209993011U (en) | 2020-01-24 |
Family
ID=69297995
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201921324178.6U Active CN209993011U (en) | 2019-08-15 | 2019-08-15 | Low-power-consumption online thermal infrared target recognition device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN209993011U (en) |
-
2019
- 2019-08-15 CN CN201921324178.6U patent/CN209993011U/en active Active
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109559310B (en) | Power transmission and transformation inspection image quality evaluation method and system based on significance detection | |
| CN110543867B (en) | Crowd density estimation system and method under condition of multiple cameras | |
| CN108109385B (en) | System and method for identifying and judging dangerous behaviors of power transmission line anti-external damage vehicle | |
| CN110415267A (en) | A kind of online thermal infrared target identification device of low-power consumption and working method | |
| CN205726177U (en) | A kind of safety defense monitoring system based on convolutional neural networks chip | |
| CN115761537B (en) | Power transmission line foreign matter intrusion identification method oriented to dynamic feature supplementing mechanism | |
| CN104820995A (en) | Large public place-oriented people stream density monitoring and early warning method | |
| CN112907522A (en) | Intelligent infrared gas leakage monitoring device and monitoring method | |
| CN106851229A (en) | A kind of method and system of the security protection intelligent decision based on image recognition | |
| CN112257527A (en) | Mobile phone detection method based on multi-target fusion and space-time video sequence | |
| CN107147877A (en) | FX night fog day condition all-weather colorful video imaging system and its construction method | |
| CN209993011U (en) | Low-power-consumption online thermal infrared target recognition device | |
| CN112129164B (en) | Intelligent assistant decision-making system architecture of weapon station | |
| CN113963373A (en) | Video image dynamic detection and tracking algorithm based system and method | |
| CN113298177A (en) | Night image coloring method, device, medium, and apparatus | |
| CN117437260A (en) | Anchor-free real-time multi-target tracking method based on joint detection and re-identification | |
| CN113872693B (en) | Method for tightly coupling visible light positioning and depth camera based on key frame selection | |
| CN115620106A (en) | Visible light and infrared multi-modal image fusion target identification method | |
| CN115578664A (en) | Video monitoring-based emergency event judgment method and device | |
| CN115100680A (en) | Pedestrian detection method based on multi-source image fusion | |
| CN112270639B (en) | Image processing method, image processing device and storage medium | |
| CN113724295A (en) | Unmanned aerial vehicle tracking system and method based on computer vision | |
| CN207766415U (en) | A kind of state boundary area monitoring system based on OPENCV human bioequivalences | |
| Sousa et al. | Late-Fusion Multimodal Human Detection Based on RGB and Thermal Images for Robotic Perception | |
| CN107918941B (en) | Visual monitoring system and method for power transmission channel external damage protection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant |