CN112686082B - In-vehicle people counting system based on visible light polarization - Google Patents
In-vehicle people counting system based on visible light polarization Download PDFInfo
- Publication number
- CN112686082B CN112686082B CN201910990791.XA CN201910990791A CN112686082B CN 112686082 B CN112686082 B CN 112686082B CN 201910990791 A CN201910990791 A CN 201910990791A CN 112686082 B CN112686082 B CN 112686082B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- polarization
- visible light
- subsystem
- people counting
- 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
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an in-vehicle people counting system based on visible light polarization, which comprises a data acquisition and information processing subsystem, a main control and people counting subsystem and an illumination subsystem. The data acquisition and information processing subsystem comprises a polarization module and an information processing module, and is used for image acquisition and polarization information fusion vehicle-passing membrane processing; the illumination subsystem supplements light for the environment in the vehicle; the master control and people counting system is used for detecting passengers and counting people in the vehicle. The invention has high statistical accuracy and good detection effect on the number of people in the vehicle with the film.
Description
Technical Field
The invention belongs to the technical field of intelligent security and protection, and relates to a system for counting the number of passengers in a film-attached vehicle.
Background
With the gradual development of economy and the increasing progress of science and technology, the living standard of people is continuously improved, automobiles are used as important transportation means for people to travel, with the national development trend, the centralized construction of some enterprise parks also increases the difficulty of park access management, and important bayonets or areas need to be checked for personnel conditions in vehicles to ensure safety. Taking Shanghai city chemical industry area as an example, the area occupied by the park is 36.1 square kilometers, which relates to 153 enterprises, more than 5 ten thousands people, more than ten roads, more than ten vehicles in the park at peak of going to and from work, and the park cannot be checked one by one. Therefore, the technical means for verifying the number of personnel in the vehicle without stopping is urgent, and the safety and the efficient personnel verification management are performed on the entering and exiting of the park license-holding vehicle. The method has the same significance for clearly detecting the personnel in the passing vehicles in some important bayonets or special areas, and can prompt or warn whether the passing of the important bayonets and the special areas is possible by detecting and analyzing the quantity, the gesture, the behavior and the like of the personnel in the passing vehicles.
The detection and identification of the personnel in the running vehicle are influenced by the film sticking of the vehicle, the movement of the vehicle and the external natural environment, so that the difficulty of detection and identification is greatly increased. The automobile film is a multilayer polyester composite film material with the functions of improving the optical performance, the safety performance and the like of automobile glass, and is a generic name of a solar film, a sunshade film, an explosion-proof film and the like. The heat insulating film for automobiles has been attracting attention because of its functions such as heat insulation and ultraviolet light insulation. The visible light and infrared reflection energy is increased, the transmissivity of solar energy is effectively reduced, and the purposes of energy conservation and heat insulation are achieved. However, in order to ensure the safety of driving a driver, the national mandatory standards prescribe that the visible light transmittance of the front windshield glass of an automobile and the visual area of the driver should be not less than 70% and the visible light transmittance of the side windshield glass of an automobile should be not less than 60%. Therefore, blocking the radiation of the middle infrared band of sunlight becomes a main means for reducing the temperature in the vehicle to achieve the heat insulation effect. In summary, the wavelength range of the window glass that is mainly transmitted is 480 to 760nm, and accordingly, if the person in the vehicle is to be clearly detected, the camera should select the wavelength range for imaging.
At present, a main mode of detecting personnel in a vehicle is to add a sensor in the vehicle for sensing, but the correctness of the detection cannot be normally evaluated for further confirmation of external personnel, the external detection is affected by a glass film and the movement of the vehicle, the application of infrared, millimeter wave, ultrasonic wave and the like is often limited, and the detection probability of the influence of the glass film, strong light and stray light of a common visible light camera is not ideal.
Disclosure of Invention
The invention aims to solve the technical problem of providing a statistics system for accurately counting the number of people in a vehicle with a film.
In order to solve the technical problems, the invention provides an in-vehicle people counting system based on visible light polarization, which adopts the following technical scheme:
the statistics system comprises a data acquisition and information processing subsystem, a main control and people counting subsystem and an illumination subsystem.
The data acquisition and information processing subsystem comprises a polarization module and an information processing module;
the information processing module judges whether the system is started or not and sends a synchronous signal to the polarization module and the illumination subsystem; carrying out polarization information fusion car passing membrane processing on the received image data, and sending the processed image to a main control and people counting system in real time;
the polarization module acquires visible light images after receiving the synchronous signals;
the illumination subsystem receives the synchronous signal and supplements light for the environment in the vehicle;
the master control and people counting system is used for detecting passengers and counting people in the vehicle.
Further, the method comprises the steps of,
the polarization module collects images by adopting a polarization camera.
The method for processing the polarization information fusion car-penetrating film comprises the following steps: and decomposing the image acquired by the polarization camera into images in four different vibration directions of 0 degree, 45 degrees, 90 degrees and 135 degrees according to the positions of all pixel points of a 2 x 2 matrix, wherein the size of the images is 1/4 of that of the original image, extracting the images in the 90 degrees direction, and transmitting the images to a main control and people counting subsystem.
The main control and people counting system adopts a deep learning technology based on big data processing to detect and count the people in the car, and the method comprises the following steps:
firstly, a large amount of calibrated data is input into a deep network, and a face and position model of a person in the vehicle is obtained after network training. And then, when the algorithm is operated in real time, loading the model obtained by training in advance and the image acquired in real time into an algorithm network, judging the content of the current image according to the model by the network, and outputting the detected positions and the detected number of people.
The algorithm network adopts a Cascade-RCNN series detection framework and an HRnet network as main structures, meanwhile, FPN and an HRnet are adopted for feature extraction, and a ROIAlign method is adopted for feature map mapping and division.
Compared with the prior art, the invention has the beneficial effects that:
the invention adopts the visible light polarization detector and the polarization film-penetrating information processing technology to realize detection of the mobile film-sticking vehicle, and compared with the technologies of light intensity imaging, spectrum imaging and the like, the polarization imaging can obtain information of other dimensions, and the polarization device has the function of well removing background noise (strong light, stray light and the like), can increase the difference between a target signal and a background clutter signal, and enhances the target profile in imaging.
The invention realizes the detection and statistics of the personnel in the vehicle based on the deep learning technology of big data, and ensures the accuracy of the detection effect.
The invention has the characteristics of low power consumption, less time consumption, high detection definition of personnel in a film-sticking vehicle, high statistical probability of the personnel number and the like, and can meet the requirements of functionality, real-time performance and adaptability in industrial application.
Drawings
The drawings in the specification of the invention are 3 in total.
FIG. 1 is a schematic diagram of an in-vehicle people counting system according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an installation layout of an in-vehicle human detection system according to an embodiment of the present invention;
fig. 3 is a front and back driver side polarization film processing image according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described in detail below with reference to the accompanying drawings and examples.
In an embodiment, the statistical system comprises a data acquisition and information processing subsystem, a master control and people counting subsystem and an illumination subsystem, and the detailed reference is made to fig. 1.
The system arrangement method refers to the installation layout schematic diagram of the human detection system in the vehicle in detail in fig. 2. And erecting data acquisition and information processing subsystems at two sides of the bayonet, namely in front of the front and back drivers. The data acquisition and information processing subsystem comprises a polarization module and an information processing module, and is used for image acquisition, polarization information fusion and vehicle-passing membrane processing, and real-time sending of processed images to the main control and people counting subsystem.
The polarization module collects images by adopting a polarization camera.
The lighting subsystem is erected at the vehicle line collision position and is used for supplementing light to the environment in the vehicle.
The main control and people counting system is arranged on one side of the front or back driver seat and is used for detecting and counting the number of passengers in the vehicle.
In the embodiment, the vehicle lifting device further comprises a bayonet comprehensive control system, and the bayonet comprehensive control system judges whether the vehicle lifting device releases the vehicle or not.
The working procedure of the embodiment is as follows:
and (3) installing the hardware system in place, waiting for passing vehicles in a standby state.
And (2) after the front and back driver seat data acquisition and information processing system receives the traffic vehicle collision signal, the system starts to work. The starting work mainly comprises the following contents:
(1) an information processing module in the data acquisition and information processing subsystem sends a synchronous signal to illumination systems on two sides of a front and back driver's seat, and illumination is started;
(2) and an information processing module in the data acquisition and processing subsystem sends a synchronous signal to the polarization camera, and starts the polarization camera of the front and back driving position to shoot and acquire images.
And (3) carrying out polarization information fusion vehicle-penetrating film processing after the data acquisition and information processing subsystem information processing module receives the image data, and sending the processed image to a main control and people counting subsystem in real time, wherein the front and back driver position polarization vehicle-penetrating film processing image is shown in fig. 3, fig. 3 (a) is a driver position image, and fig. 3 (b) is a back driver position image.
And (4) the main control and people counting system performs in-vehicle passenger detection and people counting on the film-penetrating images on the two sides of the front passenger seat.
(1) The main control and people counting subsystem inquires data signals in real time and transmits the data signals to the information processing subsystem, inquires the data transmission signals, and starts data acquisition and data receiving of the information processing subsystem on two sides of the front and back drivers;
(2) after the data is received, the main control and people counting system detects and counts the passengers in the vehicle by aligning the film-penetrating images on two sides of the front passenger seat.
And (5) after the master control and people counting subsystem finishes the statistics of the number of people in the vehicle, the number of people and the evidence obtaining image are transmitted to the comprehensive control system of the bayonet, and the comprehensive control system of the bayonet judges whether the lifting vehicle is released or not.
The polarization information fusion car-penetrating film treatment, in the embodiment, adopts the method that: and decomposing the image acquired by the polarization camera into images in four different vibration directions of 0 degree, 45 degrees, 90 degrees and 135 degrees according to the positions of all pixel points of a 2 x 2 matrix, wherein the size of the images is 1/4 of that of the original image, extracting the images in the 90 degrees direction, and transmitting the images to a main control and people counting subsystem.
In a preferred embodiment, the master control and demographics system uses deep learning techniques based on big data processing to detect and demographics occupants in a vehicle.
Firstly, a large amount of calibrated data is input into a deep network, and a face and position model of a person in the vehicle is obtained after network training. And then, when the algorithm is operated in real time, loading the model obtained by training in advance and the image acquired in real time into an algorithm network, judging the content of the current image according to the model by the network, and outputting the detected positions and the detected number of people.
In some embodiments, the algorithm network adopts a Cascade-RCNN series detection framework and a HRnet network as main structures, and meanwhile, the FPN and ROIAlign methods are adopted to improve reliability and detection effect, and finally, personnel position detection and personnel counting are realized.
Cascade-RCNN is selected as a main body detection frame: the test shows that the effect of the trained detector is very limited in general single-threshold training, and in order to ensure high quality of results and not to reduce training samples, a Cascade Stage method is adopted, and the output of one Stage is used for training the next Stage. Different thresholds are set, and the higher the threshold is, the better the effect of the candidate frame with higher network alignment accuracy is. Through threshold setting, the trained network has an optimization effect on the input suggested area.
HRnet is selected as a feature extraction network: HRnet is able to maintain a high resolution representation throughout the process, as compared to most existing methods that pass input through a network, and then increase resolution (such as by dilation convolution, etc.). Starting from the first stage with high resolution subnetworks, the high resolution to low resolution subnetworks are formed into progressively more stages and the multi-resolution subnetworks are connected in parallel. Through multi-scale fusion for multiple times in the whole process, each high-resolution representation repeatedly receives information from other parallel representations, so that rich high-resolution representations are obtained.
FPN is selected to be matched with HRnet for feature extraction: the method is mainly used for solving the defect of the RCNN algorithm in processing multi-scale change, and the FPN method structure constructs a hierarchical structure with lateral connection from top to bottom to construct high-level semantic features of each scale. The FPN may be used as a generic feature extractor, in conjunction with a detection algorithm.
And selecting the ROIALign method to map and divide the feature map: and when the feature map mapping and the feature map dividing are carried out, rounding operation is not carried out, but non-integer pixel points can be sampled in a bilinear interpolation mode finally, so that the regression position of the feature map is not affected by rounding.
Claims (8)
1. A vehicle interior people counting system based on visible light polarization is characterized in that,
the statistics system comprises a data acquisition and information processing subsystem, a main control and people counting subsystem and an illumination subsystem;
the data acquisition and information processing subsystem comprises a polarization module and an information processing module,
the information processing module judges whether the system is started or not and sends a synchronous signal to the polarization module and the illumination subsystem; the method for processing the polarization information fusion car passing film comprises the steps of: decomposing an image acquired by a polarization camera into four images with different vibration directions of 0 degree, 45 degrees, 90 degrees and 135 degrees according to the positions of all pixel points of a 2 x 2 matrix, extracting the image with the size of 1/4 of the original image, and transmitting the image with the direction of 90 degrees to a main control and people counting subsystem;
the polarization module acquires visible light images after receiving the synchronous signals;
after receiving the synchronous signal, the illumination subsystem supplements light for the environment in the vehicle;
the master control and people counting subsystem is used for detecting and counting the number of passengers in the vehicle, and a deep learning technology based on big data processing is adopted for detecting and counting the number of passengers in the vehicle.
2. The in-vehicle occupant count system based on visible light polarization of claim 1, wherein said polarization module employs a polarization camera to capture images.
3. The in-vehicle people counting system based on visible light polarization according to claim 1, wherein a large amount of calibrated data is firstly input into a depth network, a face and position model of in-vehicle people is obtained after network training, then the model obtained through pre-training and the images acquired in real time are loaded into an algorithm network when an algorithm is operated in real time, and the network judges the current image content according to the model and outputs the detected positions and the detected numbers of the people.
4. The in-vehicle people counting system based on visible light polarization according to claim 3, wherein the algorithm network adopts a Cascade-RCNN series detection frame and an HRnet network as a main structure, meanwhile adopts FPN and the HRnet to perform feature extraction, and adopts a ROIAlign method to perform feature map mapping and division.
5. The visible light polarization based in-vehicle occupant count system of claim 4, wherein cascades-RCNN are used as the subject detection framework: and training the next Stage by using the output of one Stage by adopting a Cascade Stage method, setting different thresholds, and optimizing an input suggested area through a trained network.
6. The in-vehicle people counting system based on visible light polarization according to claim 4, wherein HRnet is adopted as the feature extraction network: starting from the first stage with high-resolution subnetworks, gradually forming more stages from high-resolution subnetworks to low-resolution subnetworks, and connecting multi-resolution subnetworks in parallel, wherein each high-resolution characterization repeatedly receives information from other parallel representations through multi-scale fusion in the whole process, so that rich high-resolution characterization is obtained.
7. The in-vehicle people counting system based on visible light polarization according to claim 4, wherein the feature extraction is performed by using FPN in combination with HRnet: the FPN method structure is used for constructing a top-down hierarchical structure with lateral connection to construct high-level semantic features of various scales.
8. The in-vehicle people counting system based on visible light polarization according to claim 4, wherein the ROIAlign method is adopted for mapping and dividing the feature map: and when the feature map mapping and the feature map dividing are carried out, a bilinear interpolation mode is adopted to realize sampling of non-integer pixel points.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910990791.XA CN112686082B (en) | 2019-10-18 | 2019-10-18 | In-vehicle people counting system based on visible light polarization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910990791.XA CN112686082B (en) | 2019-10-18 | 2019-10-18 | In-vehicle people counting system based on visible light polarization |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112686082A CN112686082A (en) | 2021-04-20 |
CN112686082B true CN112686082B (en) | 2023-07-14 |
Family
ID=75444722
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910990791.XA Active CN112686082B (en) | 2019-10-18 | 2019-10-18 | In-vehicle people counting system based on visible light polarization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112686082B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001095247A1 (en) * | 2000-06-08 | 2001-12-13 | Honeywell International Inc. | Near-infrared disguise detection |
CN107525571A (en) * | 2017-09-05 | 2017-12-29 | 东南大学 | A kind of automobile intelligent overload detection system |
WO2018165027A1 (en) * | 2017-03-06 | 2018-09-13 | Polaris Sensor Technologies, Inc. | Polarization-based detection and mapping method and system |
CN108960107A (en) * | 2018-06-25 | 2018-12-07 | 安徽百诚慧通科技有限公司 | A kind of overcrowding recognition methods of small mini van and device |
CN110111565A (en) * | 2019-04-18 | 2019-08-09 | 中国电子科技网络信息安全有限公司 | A kind of people's vehicle flowrate System and method for flowed down based on real-time video |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013029451A (en) * | 2011-07-29 | 2013-02-07 | Ricoh Co Ltd | Deposit detection device and deposit detection method |
US9395437B2 (en) * | 2013-06-06 | 2016-07-19 | The United States Of America, As Represented By The Secretary Of The Army | Moving multi-polarization multi-transmitter/receiver ground penetrating radar system and signal processing for buried target detection |
-
2019
- 2019-10-18 CN CN201910990791.XA patent/CN112686082B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001095247A1 (en) * | 2000-06-08 | 2001-12-13 | Honeywell International Inc. | Near-infrared disguise detection |
WO2018165027A1 (en) * | 2017-03-06 | 2018-09-13 | Polaris Sensor Technologies, Inc. | Polarization-based detection and mapping method and system |
CN107525571A (en) * | 2017-09-05 | 2017-12-29 | 东南大学 | A kind of automobile intelligent overload detection system |
CN108960107A (en) * | 2018-06-25 | 2018-12-07 | 安徽百诚慧通科技有限公司 | A kind of overcrowding recognition methods of small mini van and device |
CN110111565A (en) * | 2019-04-18 | 2019-08-09 | 中国电子科技网络信息安全有限公司 | A kind of people's vehicle flowrate System and method for flowed down based on real-time video |
Non-Patent Citations (2)
Title |
---|
Farzan Erlik Nowruzi等.In-Vehicle Occupancy Detection with Convolutional Networks on Thermal Images.《2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops》.2019,941-948. * |
黄 雯等.一种基于多分类器融合的人数统计算法.《南开大学学报(自然科学版)》.2019,第52卷(第1期),15-20. * |
Also Published As
Publication number | Publication date |
---|---|
CN112686082A (en) | 2021-04-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR101924647B1 (en) | Determining a number of objects in an ir image | |
CN104442924B (en) | All-weather high speed railway vehicle-mounted obstacle detection system and method | |
CN107884345B (en) | Bidirectional remote sensing detection system and detection method for tail gas of motor vehicle | |
CN103465857A (en) | Mobile-phone-based active safety early-warning method for automobile | |
CN107985189B (en) | Early warning method for lane changing depth of driver in high-speed driving environment | |
CN206114513U (en) | Vapour firewood integration motor vehicle exhaust remote sensing monitoring system | |
CN104392629A (en) | Method and device for detecting car distance | |
CN102902983B (en) | A kind of taxi identification method based on support vector machine | |
CN110222596B (en) | Driver behavior analysis anti-cheating method based on vision | |
JP7280874B2 (en) | Solid-state image sensor, imaging device, and control method for solid-state image sensor | |
CN112381014A (en) | Illegal parking vehicle detection and management method and system based on urban road | |
CN106926794A (en) | Vehicle monitoring system and method thereof | |
CN112686082B (en) | In-vehicle people counting system based on visible light polarization | |
CN113673527B (en) | License plate recognition method and system | |
CN207809236U (en) | A kind of cartborne ultrasound anticollision gear with night vision target detection function | |
CN103559793B (en) | A kind of car internal sunshade board detecting method and device | |
CN103076641B (en) | A kind of safety detecting system and detection method | |
CN103559501A (en) | Vehicle sun visor detecting method and device based on image analysis | |
Hao et al. | A near-infrared imaging method for capturing the interior of a vehicle through windshield | |
CN201837782U (en) | Visual auxiliary device for virtual image projection imaging of car window | |
CN111489560A (en) | Detection method and control method for detecting front vehicle starting based on shallow convolutional neural network | |
CN107097722A (en) | Multi-screen ganged display methods and vehicle-mounted central control system for vehicle | |
CN113343903B (en) | License plate recognition method and system in natural scene | |
CN201936452U (en) | Automatic license plate recognition system | |
CN204124043U (en) | A kind of pedestrian's comprhensive deterministic system based on vehicle-mounted vision |
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 |