CN111145551A - Intersection traffic planning system based on CNN detection follows chapter rate - Google Patents
Intersection traffic planning system based on CNN detection follows chapter rate Download PDFInfo
- Publication number
- CN111145551A CN111145551A CN202010003795.7A CN202010003795A CN111145551A CN 111145551 A CN111145551 A CN 111145551A CN 202010003795 A CN202010003795 A CN 202010003795A CN 111145551 A CN111145551 A CN 111145551A
- Authority
- CN
- China
- Prior art keywords
- intersection
- module
- vehicles
- vehicle
- violation
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/042—Detecting movement of traffic to be counted or controlled using inductive or magnetic detectors
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a crossing traffic planning system based on CNN detection, which analyzes the deficiency and defect of violation detection in the current intelligent traffic system, adopts a target detection method based on CNN to calculate the violation rate of the crossing in real time and realize crossing intelligent police force scheduling, thereby completing the traffic planning of the crossing, not only improving the precision and speed of crossing violation detection, but also simultaneously detecting the violation of vehicles and pedestrians, realizing crossing intelligent scheduling of police force by calculating the violation rate and greatly saving human resources. The method is beneficial to promoting the intelligent traffic system to be further developed by combining the current popular deep learning algorithm, and provides a key technical scheme for the intelligent traffic system.
Description
Technical Field
The invention relates to the technical field of traffic planning, in particular to a road junction traffic planning system based on CNN detection and chapter-following rate.
Background
In recent years, with the rapid development of economy, people use transportation means more and more frequently, and how to perform effective and accurate traffic management is more and more important. The continuous progress of science and technology provides a new idea for solving various traffic problems by using an intelligent traffic management system, and the intelligent traffic management system is developed from a concept stage, a theory stage and a test stage to a large-scale implementation stage, and is obviously different from the traditional traffic transportation system, such as the advantages of high informatization degree, high overall global property, strong system openness, real-time dynamism and the like. The development of intelligent traffic has a close and inseparable relationship with the development of the internet of things, and along with the continuous development of the technology of the internet of things, an intelligent traffic system is more and more perfect. The intelligent transportation system is a system with strong technical performance, and compared with a common technical system, the overall requirement in the construction process of the intelligent transportation system is stricter. This integrity is mainly reflected in:
(1) cross-industry: the construction of an intelligent traffic system relates to the field of numerous industries, and is a complex large-scale system project which is widely participated in by the society, thereby causing the coordination problem among complex industries. (2) The technical subject field is complex: the intelligent traffic system integrates achievements in various scientific fields such as traffic engineering, information engineering, control engineering, communication technology, big data, cloud computing and the like, and technical personnel in various fields need to cooperate together. (3) Government, enterprises, scientific research units and colleges participate together, and proper role positioning and task sharing are important preconditions for effective development of the system.
The intelligent crossing traffic management system mainly comprises the steps of managing cameras at all crossings in batches, storing real-time crossing monitoring videos, automatically judging vehicle violation behaviors by using the cameras, automatically identifying license plates of violation vehicles, storing violation evidences and maintaining and managing violation data in a background, automatically identifying faces of pedestrians when the pedestrians cross a road, and matching the faces of the pedestrians in combination with identity information of the pedestrians stored in the background so as to determine specific identity information of the pedestrians. The intelligent traffic management system relates to a plurality of fields, comprises big data, image acquisition, information processing and online communication, and needs a plurality of industries to coordinate and jointly develop.
For the violation judgment of the vehicle at the intersection, an entity coil mode is adopted at first, the induction coil is buried under the lane line of the intersection, when the vehicle passes through the intersection under the condition of red light, the induction coil can judge that the vehicle violates the regulations, and the camera takes a picture for storage. The technology of the induction coil is mature at present, and the violation condition of the vehicle can be accurately judged. However, the induction coil is buried underground, which causes a certain problem, the road surface needs to be damaged, the traffic is affected, the adaptability is poor, if the lane of the current intersection needs to be changed, the position of the induction coil needs to be rearranged, the secondary damage can be carried out on the road surface, the service life is affected, meanwhile, the influence of the environment is great, the induction coil is very easy to damage under the environment with extreme ground surface air pressure and air temperature, and the workload of repeated construction is increased. At present, the method for judging vehicle violation by using the virtual coil is widely used, the position of the vehicle in the image is identified by adopting an optical flow method after the image is collected by a camera and preprocessed, various violation judging areas are freely defined on a rear-end platform to adapt to different requirements of each intersection, various vehicle information such as the vehicle, the vehicle distance, the vehicle speed and the like can be identified by adopting the virtual coil method, multiple lanes can be judged at the same time, and the intellectualization of a certain degree is realized.
Most modern traffic monitoring all adopt the camera of fixed position to monitor the vehicle, generally adopt the light stream method, utilize infrared ray light to carry out illumination compensation, however the light stream method is comparatively serious by the illumination influence to there is the drawback that can't discern the video vehicle number, this patent adopts the vehicle detection method violating the regulations based on degree of depth learning convolution neural network, through training convolution neural network, generate the target detection model, combine openCV to judge the behavior violating the regulations simultaneously to vehicle and pedestrian, and pedestrian and vehicle number in the recognizable image. Real-time obeying rate is generated through the number of pedestrians and vehicles in the image, and therefore police force dispatching is achieved.
The convolutional Neural network cnn (convolutional Neural network) is a feed-forward Neural network, and its artificial neurons can respond to a part of surrounding units in the coverage range, and has excellent performance for large-scale image processing. The principle of using the convolutional neural network to realize target detection in the target detection module is end-to-end detection, and considering that the system needs to be built on a mobile embedded device, a mobilent _ SSD target detection model is used, and the principle of SSD target detection is adopted.
A Single Shot MultiBox Detector, abbreviated SSD. The whole network adopts a one stage idea so as to improve the detection speed. And Faster is merged into the network; and the anchors idea in the R-CNN is adopted, and the characteristic layering extraction is performed, and the frame regression and classification operation is sequentially calculated, so that the training and detection tasks of various scale targets can be adapted. The advent of SSDs has enabled one to see the feasibility of real-time high-precision target detection.
The Mobileent _ SSD is realized by using a Mobileent neural network, the Mobileent is designed based on a streamline structure and is a neural network suitable for embedded equipment, a lightweight deep neural network is constructed by using depth separable convolutions (depthwise separable convolutions), two simple global hyper-parameters are introduced, and the speed and the accuracy are compromised. These hyper-parameters allow the model builder to select the correct size model for their application depending on the constraints of the problem. In addition, extensive experiments show that MobileNets have extremely strong performance on ImageNet classification problems and other popular models. The use of the target detection model has the following three advantages:
(1) pedestrian and vehicle capable of detecting intersection simultaneously
(2) Deep neural network capable of being operated quickly on embedded device
(3) The method has the advantages of good detection precision and speed in the daytime and at night and under different weather conditions of different intersection types, and wide applicability.
According to the invention, the accuracy and efficiency of acquiring image information by the camera are improved by adopting the deep convolutional neural network, the number of pedestrians and vehicles in the image can be identified, intelligent police dispatch is realized through the intersection obeying rate information, a large amount of human resources are saved, rich intersection information is displayed through the human-computer interaction interface with complete functions, and the intelligent management of intersection traffic is further improved.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides a road junction traffic planning system based on CNN detection, which can calculate the intersection compliance rate in real time and realize intelligent police force dispatching of the road junction, thereby completing traffic planning of the road junction, not only improving the accuracy and speed of intersection violation detection, but also detecting violation of vehicles and pedestrians, realizing intelligent dispatching of the intersection police force by calculating the compliance rate and greatly saving human resources.
The technical scheme is as follows: in order to achieve the purpose, the invention adopts the technical scheme that:
a road junction traffic planning system based on CNN detection follows the chapter rate, including goal detection module, orbit tracking module, judging module of breaking rules and regulations, follow chapter rate detection module and man-machine interaction module;
the target detection module is arranged on a camera of the intersection, and adopts a Mobilene _ SSD target detection model based on a convolutional neural network CNN, wherein the target detection model is realized by using a MobileNet convolutional neural network and is used for detecting each vehicle and pedestrian at the intersection in a video and acquiring the position coordinate information, confidence coefficient, label and flow of each vehicle and pedestrian;
the method comprises the following specific steps:
step 1.1, inputting a video frame, performing image preprocessing on the video frame, and zooming the image into 224 × 224;
step 1.2, inputting the scaling result into a trained Mobilene _ SSD target detection model, and obtaining a Bounding Box; the Bounding Box is a multidimensional matrix array and comprises a target type of a detected target, coordinate information and confidence;
step 1.3, filtering a detection result with the confidence coefficient lower than 0.7, and obtaining coordinate information of vehicles and pedestrians through the types and coordinate labels of objects in the Bounding Box;
the track tracking module receives and processes position coordinate information acquired by the target detection module in real time, calculates coordinates of a central point, and classifies vehicles and pedestrians through labels; calculating the Euclidean distance of each vehicle between two adjacent frames, setting a proper threshold value, and acquiring the motion track of each unit in the image;
the violation judging module adopts a virtual coil method as a judging principle, extracts and judges the traffic light state through an HSV color channel, and judges the motion track of the vehicle or the pedestrian according to the motion track of the vehicle or the pedestrian obtained by the track tracking module; in the image, a plurality of virtual coils are arranged according to different road conditions; when the motion track passes through a specific virtual coil, determining violation of regulation, and capturing and storing the picture;
the detection module for the rule-violating rate counts current rule-violating information, determines the police force required by the intersection to be detected by setting a threshold value, gives a police force scheduling scheme by the rule-violating rate, is used for maintaining the traffic state of the intersection and guides a traffic department to arrange the police force to a specified intersection to correspondingly process the rule-violating condition;
the man-machine interaction module is a man-machine interaction interface developed based on Qt and comprises a background liquid crystal display screen and an intersection display screen; in a background liquid crystal display screen, displaying an image acquired by a camera in a video display area in real time, displaying violation information and violation types obtained in a violation judging module in right-side violation data, and displaying the chapter compliance rate information, traffic flow and pedestrian flow information obtained in a chapter compliance rate detecting module in right-side statistical data; and displaying screenshots of the illegal vehicles and pedestrians and the current intersection compliance rate information on the intersection display screen.
Further, the specific working steps of the target detection module are as follows:
2.1, acquiring image information of various vehicles and pedestrians at the road junction according to different time periods, different road junctions and different weather conditions to manufacture a training set;
2.2, inputting the training interface into a Mobilene _ SSD detection model, and training to obtain higher position coordinate precision;
and 2.3, transmitting the identified target information to a track tracking module for calculating the moving track of the target unit.
Further, the specific method for calculating the euclidean distance between units by the trajectory tracking module is as follows:
according to target detection information obtained by a target detection module, firstly calculating the coordinates of the center point of an object, and classifying the object according to different label values; the coordinates of the center point of the object represent the position of the object in the video and are also used as the basis for tracking the track; and then calculating the Euclidean distance of each vehicle between two adjacent frames as follows:
n, n-1 represents adjacent frames, a and b represent vehicles in the front and rear video frames respectively, and x and y represent coordinates set by the vehicles in the video frames by taking the upper left corner as an origin;
representing the Euclidean distance of all vehicles between the front video frame and the rear video frame;
setting a threshold value for the maximum Euclidean distance of the same vehicle between two frames, if the calculated distance is less than or equal to the threshold value, the vehicles between adjacent frames are the same vehicle, if the calculated distance is greater than the threshold value, judging that the vehicles are not the same vehicle, and adjusting the threshold value to obtain the running track of each object in the image.
Furthermore, the violation judging module defines a traffic light detection area in the video frame as an area defined by a traffic light red frame, and performs image denoising processing on the selected area to obtain a smoother traffic light image; converting the color space of the traffic light image into an HSV model, and carrying out histogram equalization processing on the V channel to eliminate noise influence caused by illumination change; finally, threshold segmentation is carried out according to H channel ranges of the red, green and yellow colors, and a characteristic response graph which accords with the three color ranges is obtained; and respectively counting the number of the response pixels, and selecting the color corresponding to the maximum value as the state of the current traffic light.
Further, the method for defining the virtual coil by the violation judging module is as follows:
(1) for the left-hand driving of vehicles at the intersection, 2 virtual coils are respectively drawn at the intersection and in the left front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, the module judges that the vehicles violate the regulations;
(2) for right-going of vehicles at the intersection, respectively marking 2 virtual coils at the intersection and at the right front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, judging that the vehicles violate the regulations by the module;
(3) for crossing vehicle straight going, respectively drawing 2 virtual coils at the crossing and right in front of the crossing, when the vehicle continuously passes through the 2 virtual coils, judging that the vehicle breaks rules and regulations;
(4) dividing the yellow line and the lane line of the road into 1 virtual coil respectively, and judging that the vehicle breaks rules and regulations by the module when the vehicle track passes through any one of the virtual coils;
(5) for the sidewalk, 1 virtual coil is respectively arranged at two ends of the sidewalk, and when the pedestrian continuously passes through 2 virtual coils, the module judges that the pedestrian breaks rules and regulations.
Further, the rule following rate detection module calculates the rule following rate of the number of violations and the total traffic flow or the pedestrian flow obtained by the rule violation determination module; the calculation of the compliance rate is as follows:
wherein n is the number of passing vehicles or total number within a fixed time, m is the number of vehicles violating the regulations, and after the rule compliance rate is calculated, the rule compliance rate can be updated in real time by the system and is refreshed every 24 hours; and presetting a threshold value of the compliance rate, and determining the police force required to be dispatched at the current intersection according to the corresponding threshold value so as to realize intelligent intersection traffic planning.
Has the advantages that: the system provided by the invention has the advantages that:
(1) the CNN-based target detection method is used for replacing the traditional method, so that the accuracy and speed of the system in the aspect of violation judgment are improved;
(2) police dispatching is completed by automatically detecting the obedience rate of each road, and the intellectualization of the system is further realized;
(3) the modular design is applied, the upgrading is simple, and excessive codes are not needed for changing hardware.
Drawings
FIG. 1 is a flow chart of the system of the present invention;
FIG. 2 is a schematic view of the camera mounting of the system of the present invention;
FIG. 3 is a schematic diagram of vehicle violation virtual coil partitioning in accordance with the present invention;
FIG. 4 is a schematic diagram of pedestrian violation virtual coil partitioning in the present invention;
FIG. 5 is a schematic view of a back-end LCD panel according to the present invention;
FIG. 6 is a schematic view of a junction display screen according to the present invention;
FIG. 7 is a flow chart of object detection in the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings.
The invention provides a road junction traffic planning system based on CNN detection of an obedience rate, which comprises a target detection module, a track tracking module, a violation judgment module, a obedience rate detection module and a human-computer interaction module.
The camera is adopted as intersection image acquisition equipment, a liquid crystal display screen is used for respectively displaying a human-computer interaction interface and intersection violation information, a development platform with an RK3399 framework of an Arm SoC is used as a main control board, as shown in figure 1, the development platform is embedded into the camera, and the camera is arranged at a position adjacent to an intersection traffic light.
The target detection module is arranged on a camera of the intersection, a Mobilenet _ SSD target detection model based on a convolutional neural network CNN is adopted, the target detection model is realized by using a Mobilenet convolutional neural network, a lightweight deep neural network is constructed by adopting deep separable convolution, and compromise processing is performed on speed and precision, so that the target detection module is used for detecting each vehicle and pedestrian at the intersection in a video, and acquiring position coordinate information, confidence, labels and flow of each vehicle and pedestrian is shown in fig. 7. The specific working steps are as follows:
step 1.1, inputting a video frame, performing image preprocessing on the video frame, and zooming the image into 224 × 224;
step 1.2, inputting the scaling result into a trained Mobilene _ SSD target detection model, and obtaining a Bounding Box; the Bounding Box is a multidimensional matrix array and comprises a target type of a detected target, coordinate information and confidence;
step 1.3, filtering a detection result with the confidence coefficient lower than 0.7, and obtaining coordinate information of vehicles and pedestrians through the types and coordinate labels of objects in the Bounding Box; the track tracking module receives and processes the position coordinate information acquired by the target detection module in real time, calculates the coordinates of the central point, and classifies the vehicles and the pedestrians through the labels.
And calculating the Euclidean distance of each vehicle between two adjacent frames, setting a proper threshold value, and acquiring the motion track of each unit in the image. The specific method for calculating the Euclidean distance between units is as follows:
according to target detection information obtained by a target detection module, firstly calculating the coordinates of the center point of an object, and classifying the object according to different label values; the coordinates of the center point of the object represent the position of the object in the video and are also used as the basis for tracking the track; and then calculating the Euclidean distance of each vehicle between two adjacent frames as follows:
n, n-1 represents adjacent frames, a and b represent vehicles in the front and rear video frames respectively, and x and y represent coordinates set by the vehicles in the video frames by taking the upper left corner as an origin;
representing the Euclidean distance of all vehicles between the front video frame and the rear video frame;
setting a threshold value for the maximum Euclidean distance of the same vehicle between two frames, if the calculated distance is less than or equal to the threshold value, the vehicles between adjacent frames are the same vehicle, if the calculated distance is greater than the threshold value, judging that the vehicles are not the same vehicle, and adjusting the threshold value to obtain the running track of each object in the image.
The violation judging module adopts a virtual coil method as a judging principle, extracts and judges the traffic light state through an HSV color channel, and obtains the motion track of the vehicle or the pedestrian according to the track tracking module; in the image, a plurality of virtual coils are arranged according to different road conditions; when the motion track passes through a specific virtual coil, the violation is judged, and the screenshot is saved.
The violation judging module is used for defining a traffic light detection area in the video frame as an area defined by a traffic light red frame, and performing image denoising processing on the selected area to obtain a smooth traffic light image; converting the color space of the traffic light image into an HSV model, and carrying out histogram equalization processing on the V channel to eliminate noise influence caused by illumination change; finally, threshold segmentation is carried out according to H channel ranges of the red, green and yellow colors, and a characteristic response graph which accords with the three color ranges is obtained; and respectively counting the number of the response pixels, and selecting the color corresponding to the maximum value as the state of the current traffic light.
The method for defining the virtual coil is shown in figures 2-3:
(1) for the left-hand driving of vehicles at the intersection, 2 virtual coils are respectively drawn at the intersection and in the left front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, the module judges that the vehicles violate the regulations;
(2) for right-going of vehicles at the intersection, respectively marking 2 virtual coils at the intersection and at the right front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, judging that the vehicles violate the regulations by the module;
(3) for crossing vehicle straight going, respectively drawing 2 virtual coils at the crossing and right in front of the crossing, when the vehicle continuously passes through the 2 virtual coils, judging that the vehicle breaks rules and regulations;
(4) dividing the yellow line and the lane line of the road into 1 virtual coil respectively, and judging that the vehicle breaks rules and regulations by the module when the vehicle track passes through any one of the virtual coils;
(5) for the sidewalk, 1 virtual coil is respectively arranged at two ends of the sidewalk, and when the pedestrian continuously passes through 2 virtual coils, the module judges that the pedestrian breaks rules and regulations.
The rule-violating rate detection module counts current rule-violating information, police force required by the intersection to be detected is determined by setting a threshold value, a police force scheduling scheme is given through the rule-violating rate and used for maintaining the traffic state of the intersection, and a traffic department is guided to arrange the police force to a specified intersection to perform corresponding processing on the rule-violating condition. Specifically, the violation number and the total traffic flow or the pedestrian flow obtained by the violation judging module are subjected to the rule following rate calculation; the calculation of the compliance rate is as follows:
wherein n is the number of passing vehicles or total number within a fixed time, m is the number of vehicles violating the regulations, and after the rule compliance rate is calculated, the rule compliance rate can be updated in real time by the system and is refreshed every 24 hours; the preset obeying chapter rate threshold is shown in table 1, and the police force required to be dispatched at the current intersection is determined according to the corresponding threshold, so that the intelligent intersection traffic planning is realized.
TABLE 1
| Rate of observing the chapter | Following the situation of chapter rate | Number of police dispatch persons |
| >95％ | Is excellent in | 0 |
| <95％ | Good effect | 2 |
| <90％ | Is poor | 4 |
| <85％ | Extreme difference | 8 |
The human-computer interaction module uses a Qt development tool to design two human-computer interaction interfaces with complete functions. QT is a cross-platform C + + graphical user interface library produced by TrollTech corporation in Norway, and currently includes Qt Creator, QtEmbedded, Qt Designer rapid development tool, Qt Linguist internationalization tool and other parts, and Qt supports all Linux/Unix systems and Windows platforms. The system comprises a background liquid crystal display screen and a crossing display screen. The back-end liquid crystal display screen is shown in figure 4, the right part of the interface displays various violation information, the left part of the interface displays real-time intersection information collected by a camera, various violation information can be inquired through the upper button, and the man-machine interaction interface is displayed on the display screen of the back-end of the system. The intersection display screen is shown in fig. 5 and is used for displaying information of people and vehicles violating the traffic regulations at present on a screen of an actual intersection and displaying the information of the compliance rate of the current intersection in real time, so that a traffic police can know the violation conditions of the intersection in the jurisdiction more conveniently, and the effect of warning other people is achieved by displaying the violation vehicles or pedestrians, and the traffic conditions of the intersection can be effectively improved.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.
Claims (6)
1. A crossing traffic planning system based on CNN detects and follows chapter rate, its characterized in that: the system comprises a target detection module, a track tracking module, a violation judging module, a chapter following rate detection module and a human-computer interaction module;
the target detection module is arranged on a camera of the intersection, and adopts a Mobilene _ SSD target detection model based on a convolutional neural network CNN, wherein the target detection model is realized by using a MobileNet convolutional neural network and is used for detecting each vehicle and pedestrian at the intersection in a video and acquiring the position coordinate information, confidence coefficient, label and flow of each vehicle and pedestrian;
the method comprises the following specific steps:
step 1.1, inputting a video frame, performing image preprocessing on the video frame, and zooming the image into 224 × 224;
step 1.2, inputting the scaling result into a trained Mobilene _ SSD target detection model, and obtaining a Bounding Box; the Bounding Box is a multidimensional matrix array and comprises a target type of a detected target, coordinate information and confidence;
step 1.3, filtering a detection result with the confidence coefficient lower than 0.7, and obtaining coordinate information of vehicles and pedestrians through the types and coordinate labels of objects in the Bounding Box;
the track tracking module receives and processes position coordinate information acquired by the target detection module in real time, calculates coordinates of a central point, and classifies vehicles and pedestrians through labels; calculating the Euclidean distance of each vehicle between two adjacent frames, setting a proper threshold value, and acquiring the motion track of each unit in the image;
the violation judging module adopts a virtual coil method as a judging principle, extracts and judges the traffic light state through an HSV color channel, and judges the motion track of the vehicle or the pedestrian according to the motion track of the vehicle or the pedestrian obtained by the track tracking module; in the image, a plurality of virtual coils are arranged according to different road conditions; when the motion track passes through a specific virtual coil, determining violation of regulation, and capturing and storing the picture;
the detection module for the rule-violating rate counts current rule-violating information, determines the police force required by the intersection to be detected by setting a threshold value, gives a police force scheduling scheme by the rule-violating rate, is used for maintaining the traffic state of the intersection and guides a traffic department to arrange the police force to a specified intersection to correspondingly process the rule-violating condition;
the man-machine interaction module is a man-machine interaction interface developed based on Qt and comprises a background liquid crystal display screen and an intersection display screen; in a background liquid crystal display screen, displaying an image acquired by a camera in a video display area in real time, displaying violation information and violation types obtained in a violation judging module in right-side violation data, and displaying the chapter compliance rate information, traffic flow and pedestrian flow information obtained in a chapter compliance rate detecting module in right-side statistical data; and displaying screenshots of the illegal vehicles and pedestrians and the current intersection compliance rate information on the intersection display screen.
2. The intersection traffic planning system based on CNN detection and chapter-following rate as claimed in claim 1, wherein: the specific working steps of the target detection module are as follows:
2.1, acquiring image information of various vehicles and pedestrians at the road junction according to different time periods, different road junctions and different weather conditions to manufacture a training set;
2.2, inputting the training interface into a Mobilene _ SSD detection model, and training to obtain higher position coordinate precision;
and 2.3, transmitting the identified target information to a track tracking module for calculating the moving track of the target unit.
3. The intersection traffic planning system based on CNN detection and chapter-following rate as claimed in claim 1, wherein: the specific method for calculating the Euclidean distance between units by the trajectory tracking module is as follows:
according to target detection information obtained by a target detection module, firstly calculating the coordinates of the center point of an object, and classifying the object according to different label values; the coordinates of the center point of the object represent the position of the object in the video and are also used as the basis for tracking the track; and then calculating the Euclidean distance of each vehicle between two adjacent frames as follows:
n, n-1 represents adjacent frames, a and b represent vehicles in the front and rear video frames respectively, and x and y represent coordinates set by the vehicles in the video frames by taking the upper left corner as an origin;
representing the Euclidean distance of all vehicles between the front video frame and the rear video frame;
setting a threshold value for the maximum Euclidean distance of the same vehicle between two frames, if the calculated distance is less than or equal to the threshold value, the vehicles between adjacent frames are the same vehicle, if the calculated distance is greater than the threshold value, judging that the vehicles are not the same vehicle, and adjusting the threshold value to obtain the running track of each object in the image.
4. The intersection traffic planning system based on CNN detection and chapter-following rate as claimed in claim 1, wherein: the violation judging module is used for defining a traffic light detection area in the video frame as an area defined by a traffic light red frame, and performing image denoising processing on the selected area to obtain a smooth traffic light image; converting the color space of the traffic light image into an HSV model, and carrying out histogram equalization processing on the V channel to eliminate noise influence caused by illumination change; finally, threshold segmentation is carried out according to H channel ranges of the red, green and yellow colors, and a characteristic response graph which accords with the three color ranges is obtained; and respectively counting the number of the response pixels, and selecting the color corresponding to the maximum value as the state of the current traffic light.
5. The intersection traffic planning system based on CNN detection and chapter-following rate as claimed in claim 1, wherein: the method for defining the virtual coil by the violation judging module comprises the following steps:
(1) for the left-hand driving of vehicles at the intersection, 2 virtual coils are respectively drawn at the intersection and in the left front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, the module judges that the vehicles violate the regulations;
(2) for right-going of vehicles at the intersection, respectively marking 2 virtual coils at the intersection and at the right front of the intersection, and when the vehicles continuously pass through the 2 virtual coils, judging that the vehicles violate the regulations by the module;
(3) for crossing vehicle straight going, respectively drawing 2 virtual coils at the crossing and right in front of the crossing, when the vehicle continuously passes through the 2 virtual coils, judging that the vehicle breaks rules and regulations;
(4) dividing the yellow line and the lane line of the road into 1 virtual coil respectively, and judging that the vehicle breaks rules and regulations by the module when the vehicle track passes through any one of the virtual coils;
(5) for the sidewalk, 1 virtual coil is respectively arranged at two ends of the sidewalk, and when the pedestrian continuously passes through 2 virtual coils, the module judges that the pedestrian breaks rules and regulations.
6. The intersection traffic planning system based on CNN detection and chapter-following rate as claimed in claim 1, wherein: the rule following rate detection module calculates rule following rates of the number of violations and the total traffic flow or the pedestrian flow obtained by the rule violation determination module; the calculation of the compliance rate is as follows:
wherein n is the number of passing vehicles or total number within a fixed time, m is the number of vehicles violating the regulations, and after the rule compliance rate is calculated, the rule compliance rate can be updated in real time by the system and is refreshed every 24 hours; and presetting a threshold value of the compliance rate, and determining the police force required to be dispatched at the current intersection according to the corresponding threshold value so as to realize intelligent intersection traffic planning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010003795.7A CN111145551A (en) | 2020-01-03 | 2020-01-03 | Intersection traffic planning system based on CNN detection follows chapter rate |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010003795.7A CN111145551A (en) | 2020-01-03 | 2020-01-03 | Intersection traffic planning system based on CNN detection follows chapter rate |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111145551A true CN111145551A (en) | 2020-05-12 |
Family
ID=70523437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010003795.7A Pending CN111145551A (en) | 2020-01-03 | 2020-01-03 | Intersection traffic planning system based on CNN detection follows chapter rate |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111145551A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112699827A (en) * | 2021-01-05 | 2021-04-23 | 长威信息科技发展股份有限公司 | Traffic police affair handling method and system based on block chain |
| CN114783181A (en) * | 2022-04-13 | 2022-07-22 | 江苏集萃清联智控科技有限公司 | Traffic flow statistical method and device based on roadside perception |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070030171A1 (en) * | 2005-08-08 | 2007-02-08 | University Of Washington | Mobile system and method for traffic detector simulations |
| CN102496281A (en) * | 2011-12-16 | 2012-06-13 | 湖南工业大学 | Vehicle red-light violation detection method based on combination of tracking and virtual loop |
| CN102768799A (en) * | 2011-12-21 | 2012-11-07 | 湖南工业大学 | Method for detecting red light running of vehicle at night |
| CN102810250A (en) * | 2012-07-31 | 2012-12-05 | 长安大学 | Video based multi-vehicle traffic information detection method |
| CN102903239A (en) * | 2012-09-27 | 2013-01-30 | 安科智慧城市技术（中国）有限公司 | Method and system for detecting illegal left-and-right steering of vehicle at traffic intersection |
| CN103354029A (en) * | 2013-07-26 | 2013-10-16 | 安徽三联交通应用技术股份有限公司 | Multi-functional intersection traffic information collection method |
| CN108875588A (en) * | 2018-05-25 | 2018-11-23 | 武汉大学 | Across camera pedestrian detection tracking based on deep learning |
| CN109522854A (en) * | 2018-11-22 | 2019-03-26 | 广州众聚智能科技有限公司 | A kind of pedestrian traffic statistical method based on deep learning and multiple target tracking |
-
2020
- 2020-01-03 CN CN202010003795.7A patent/CN111145551A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070030171A1 (en) * | 2005-08-08 | 2007-02-08 | University Of Washington | Mobile system and method for traffic detector simulations |
| CN102496281A (en) * | 2011-12-16 | 2012-06-13 | 湖南工业大学 | Vehicle red-light violation detection method based on combination of tracking and virtual loop |
| CN102768799A (en) * | 2011-12-21 | 2012-11-07 | 湖南工业大学 | Method for detecting red light running of vehicle at night |
| CN102810250A (en) * | 2012-07-31 | 2012-12-05 | 长安大学 | Video based multi-vehicle traffic information detection method |
| CN102903239A (en) * | 2012-09-27 | 2013-01-30 | 安科智慧城市技术（中国）有限公司 | Method and system for detecting illegal left-and-right steering of vehicle at traffic intersection |
| CN103354029A (en) * | 2013-07-26 | 2013-10-16 | 安徽三联交通应用技术股份有限公司 | Multi-functional intersection traffic information collection method |
| CN108875588A (en) * | 2018-05-25 | 2018-11-23 | 武汉大学 | Across camera pedestrian detection tracking based on deep learning |
| CN109522854A (en) * | 2018-11-22 | 2019-03-26 | 广州众聚智能科技有限公司 | A kind of pedestrian traffic statistical method based on deep learning and multiple target tracking |
Non-Patent Citations (2)
| Title |
|---|
| 刘永祥 等: "电子警察闯红灯前端抓拍系统的研究与设计", 《计算机技术与发展》 * |
| 杨洁琼: "嵌入式交通违章抓拍系统的研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112699827A (en) * | 2021-01-05 | 2021-04-23 | 长威信息科技发展股份有限公司 | Traffic police affair handling method and system based on block chain |
| CN112699827B (en) * | 2021-01-05 | 2023-07-25 | 长威信息科技发展股份有限公司 | Traffic police treatment method and system based on blockchain |
| CN114783181A (en) * | 2022-04-13 | 2022-07-22 | 江苏集萃清联智控科技有限公司 | Traffic flow statistical method and device based on roadside perception |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Distribution line pole detection and counting based on YOLO using UAV inspection line video | |
| CN106875424B (en) | A kind of urban environment driving vehicle Activity recognition method based on machine vision | |
| Al-qaness et al. | An improved YOLO-based road traffic monitoring system | |
| CN112750150B (en) | Vehicle flow statistical method based on vehicle detection and multi-target tracking | |
| CN114970321A (en) | Scene flow digital twinning method and system based on dynamic trajectory flow | |
| CN109740420A (en) | Vehicle illegal recognition methods and Related product | |
| CN103577875A (en) | CAD (computer-aided design) people counting method based on FAST (features from accelerated segment test) | |
| CN112241969A (en) | Target detection tracking method and device based on traffic monitoring video and storage medium | |
| CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
| CN113592905B (en) | Vehicle driving track prediction method based on monocular camera | |
| CN111598069B (en) | Highway vehicle lane change area analysis method based on deep learning | |
| CN109272482A (en) | A kind of urban road crossing vehicle queue detection system based on sequence image | |
| CN113450573A (en) | Traffic monitoring method and traffic monitoring system based on unmanned aerial vehicle image recognition | |
| CN111145551A (en) | Intersection traffic planning system based on CNN detection follows chapter rate | |
| CN114299456A (en) | Intersection pedestrian crossing risk assessment method based on real-time track detection | |
| CN115170611A (en) | Complex intersection vehicle driving track analysis method, system and application | |
| CN113903008A (en) | Ramp exit vehicle violation identification method based on deep learning and trajectory tracking | |
| CN116993681A (en) | Substation inspection defect detection method and system | |
| CN113239725B (en) | Pedestrian waiting for crossing and crossing direction recognition method and system | |
| CN104331708B (en) | A kind of zebra crossing automatic detection analysis method and system | |
| CN113111876A (en) | Method and system for obtaining evidence of traffic violation | |
| Rouf et al. | Real-time vehicle detection, tracking and counting system based on YOLOv7 | |
| CN114820931B (en) | Virtual reality-based CIM (common information model) visual real-time imaging method for smart city | |
| CN114495421B (en) | Intelligent open type road construction operation monitoring and early warning method and system | |
| Wang et al. | Vehicle key information detection algorithm based on improved SSD |
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 | ||
| CB02 | Change of applicant information |
Address after: 210003 Gulou District, Jiangsu, Nanjing new model road, No. 66 Applicant after: NANJING University OF POSTS AND TELECOMMUNICATIONS Address before: Yuen Road Qixia District of Nanjing City, Jiangsu Province, No. 9 210003 Applicant before: NANJING University OF POSTS AND TELECOMMUNICATIONS |
|
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200512 |
|
| RJ01 | Rejection of invention patent application after publication |