TWI777620B - Automated traffic steering quantitative survey report output system and method - Google Patents

Abstract

An automated traffic steering quantitative investigation report generating system and method thereof, comprising at least one image capturing device and at least one host. After the image capture device captures an intersection image in a measurement area, it is sent to the host for analysis. The host sets the intersection determination line and advanced determination area in the measurement area, and then trains a deep learning model to analyze the types of vehicles and Position; after the host computer calculates the trajectory information of the vehicle, it uses the judgment line to determine the traveling direction of the vehicle. When the trajectory information is continuously broken, the advanced judgment area is used to correct the vehicle's inbound and outbound direction, and the inbound and outbound direction and trajectory information are used for data. Processing to generate a diversion report. By means of the present invention, a report can be generated in real time, and the accuracy of object tracking can be improved.

Description

Automated traffic steering quantitative survey report output system and method

The invention relates to a traffic state analysis, in particular to an automatic traffic steering quantitative investigation report output system and method thereof.

Road intersections are prone to various unexpected situations due to traffic meeting, vehicle turning, and locomotive waiting to turn. Therefore, government traffic units need to conduct traffic steering volume surveys at major intersections. However, traditional manual surveys only do short-term traffic surveys due to labor limitations, and cannot provide comprehensive data collection, and the classification of objects is limited to a few items, such as large cars, small cars, and locomotives, and cannot provide detailed general classification.

There is another method that uses a camera to capture road images for analysis, but currently a fisheye camera is used. Although it can shoot a 360-degree image angle of view, due to the limitation of the optical imaging of the fisheye lens used by the fisheye camera, the image is limited. The edge area is prone to problems such as deformation and poor imaging quality, and it is difficult to perform image recognition and tracking for complete vehicle crossing behavior analysis. Moreover, fisheye cameras cannot cover the behavior of all objects at large intersections, especially the two-stage left-turn detection of locomotives may exceed the image range and cannot be detected.

In view of this, the present invention proposes an automated traffic steering quantitative survey report output system and method to effectively solve the above-mentioned problems in view of the above-mentioned deficiencies and future needs of the prior art. The specific structure and its implementation will be described in detail. Below:

The main purpose of the present invention is to provide an automatic traffic steering quantitative survey report output system and method, which set up a judgment line and an advanced judgment area at the intersection to solve the problems of the judgment line being blocked by large objects or the failure of object detection.

Another object of the present invention is to provide an automated traffic steering quantitative survey report generation system and method thereof, which, combined with a deep learning model and a trajectory algorithm, can improve the analysis effect and achieve the advantages of high accuracy and high speed.

Yet another object of the present invention is to provide an automated traffic steering quantitative survey report output system and method, which use an ultra-wide-angle camera to capture images at intersections, and can see the entire position of the intersection, which solves the problem that it is difficult to observe complete vehicles crossing the intersection in the past. behavior.

In order to achieve the above object, the present invention provides an automated traffic steering quantitative survey report generation system, comprising: at least one image capture device for capturing images of a plurality of intersections in a measurement area; and at least one host computer for signal connection to the image capture device , receives the intersection image and analyzes it to generate a steering volume report. The host includes: a deep learning model module, which uses a deep learning model data to train a deep learning model, and then inputs the intersection image into the deep learning model. Analyze the type and position of at least one vehicle; a trajectory computing module, connected to the deep learning model module, reads the type and position of the vehicle from the deep learning learning module according to the time sequence, and calculates the trajectory information of the vehicle, including the identification serial number, Type and position; a vehicle steering analysis module, connected to the trajectory calculation module, uses the vehicle's trajectory information and the judgment line to determine the vehicle's travel direction, and when the trajectory information has continuous frame breaks, the advanced judgment area is used to correct the vehicle's entry and exit. To determine the direction of the area, the vehicle turns to the analysis module and Output the vehicle's inbound and outbound direction and trajectory information; and a report output module, which is connected to the vehicle steering analysis module to process data on the vehicle's inbound and outbound direction and trajectory information to generate a steering volume report.

According to an embodiment of the present invention, an initialization module is further included, which is connected to the deep learning model module and the vehicle steering analysis module, and provides the deep learning model data, the at least one determination line and the at least one advanced determination area.

According to an embodiment of the present invention, the initialization module uses the cv2 function to connect the output of the image capture device to read the intersection image.

According to an embodiment of the present invention, the deep learning model data loaded by the initialization module includes parameters of the deep learning model and related algorithms.

According to an embodiment of the present invention, the steps of training the deep learning model by the deep learning model module include: collecting at least one public data set of different types of vehicles; collecting plural demand data of intersections; and using an image labeling tool to mark the demand data. , and use the public data set to define the type of vehicle; input the marked demand data into a deep learning neural network for model training, and output an initial deep learning model; and optimize the model according to the characteristics of different intersections to generate a deep learning model .

According to an embodiment of the present invention, after the trajectory computing module reads the type and position of the vehicle, it further includes the following steps: using a linear velocity model and a Kalman filter to predict the type and position of the vehicle at the nth time point , predict the type and location of the vehicle at the n+1th time point; use a Hungarian algorithm to associate the nth time point and the n+1th time point with the vehicle; Track information, including identification number, category, and location.

According to an embodiment of the present invention, after the vehicle steering analysis module reads the trajectory information of the vehicle transmitted by the trajectory calculation module, it further includes the following steps: judging whether there is any part of the trajectory information of the vehicle; The continuous frame break in any one of them exceeds a preset value. If so, it is determined that the vehicle with the continuous frame break has an abnormal trajectory and has left the measurement area, and the first point information and the end point information of the vehicle entering and leaving the measurement area are generated. No, continue to detect; use the judgment line to combine the starting point information and the ending point information to judge the traveling direction of the vehicle with continuous frame breakage; and use the advanced judgment area to correct the entry and exit directions of the vehicle with continuous frame breakage, and then output the The entry and exit directions and trajectory information of vehicles with continuous frame breaks.

According to the embodiment of the present invention, the report generating module performs different types of data processing on the in-out direction and trajectory information of the vehicle according to different measurement areas.

According to an embodiment of the present invention, the report output module makes customized adjustments to the output format of the output steering amount report according to requirements, including adjusting the passenger car unit (PCU) and category parameters in the output steering amount report.

The present invention further provides a method for producing an automated traffic steering quantitative survey report, comprising the following steps: capturing a plurality of intersection images in a measurement area by using at least one image capture device; A deep learning model module uses a deep learning model data to train a deep learning model, and then inputs the intersection image into the deep learning model to analyze the type and position of at least one vehicle; The deep learning learning module reads the type and position of the vehicle, and calculates the trajectory information of the vehicle, including the identification serial number, type and position; a vehicle steering analysis module in the host uses the trajectory information and judgment line of the vehicle to determine the vehicle's travel. Direction, when there are continuous frame breaks in the trajectory information, the advanced judgment area is used to correct the direction of the vehicle entering and exiting the advanced judgment area, the vehicle steering analysis module and output the vehicle's entry and exit direction and trajectory information; and a report output module in the host. The group performs data processing on the vehicle's inbound and outbound direction and trajectory information, and generates a steering amount report.

10: Automated Traffic Steering Quantitative Survey Report Output System

12: Image capture device

20: Host

202: Initialize the module

204: Deep Learning Model Module

206: Trajectory operation module

208: Vehicle Steering Analysis Module

209: Report output module

28: Large Vehicles

30, 31: Judgment line

34: Advanced Judgment Area

36: Advanced Judgment Area

Fig. 1 is a block diagram of the automatic traffic steering quantitative survey report output system of the present invention.

Fig. 2 is a flow chart of the initialization module in the automatic traffic steering quantitative survey report output system of the present invention.

FIG. 3 is a flowchart of the deep learning model module in the automated traffic steering quantitative survey report output system of the present invention.

FIG. 4 is a training flow chart of the pre-operation of the deep learning model module in the automated traffic steering quantitative survey report output system of the present invention.

FIG. 5 is a flow chart of the trajectory calculation module in the automatic traffic steering quantitative survey report output system of the present invention.

FIG. 6 is a flow chart of the vehicle steering analysis module in the automatic traffic steering quantitative survey report output system of the present invention.

FIG. 7 is a flow chart of the report output module in the automatic traffic diversion quantitative survey report output system of the present invention.

FIG. 8A and FIG. 8B are schematic diagrams of setting the determination line and the advanced determination area in the present invention.

The invention provides an automatic traffic steering quantitative investigation report generating system and method thereof, which are used for real-time and rapid processing of big data intersection images, and outputting high-precision traffic steering quantity reports, so as to achieve the purpose of real-time big data analysis.

Please refer to FIG. 1 , which is a block diagram of an automated traffic steering quantitative survey report output system of the present invention. The automated traffic steering quantification survey report generating system 10 of the present invention includes at least one image capture device 12 and at least one host 20. The host 20 is installed on street lights, traffic signs or other devices at intersections, and can capture images at close distances. Take the device 12 and connect it. In one embodiment, the host 20 is a small body The integrated chassis can perform image processing and data calculation. The image capture device 12 can be a camera or other machine that can continuously capture image frames to capture multiple intersection images in a measurement area of a road intersection. In a preferred embodiment, the image capture device 12 used in the present invention is For an ultra-wide-angle camera, a single ultra-wide-angle camera includes at least two lenses, so the purpose of capturing images at an ultra-wide angle can be achieved. In one embodiment, if it is a large-scale intersection, two image capture devices 12 can be used to shoot oppositely at the same time to ensure that the entire intersection is captured; the host 20 is connected to the image capture device 12 for signal connection to receive and analyze the image of the intersection. , to generate a vector report. The host 20 includes an initialization module 202 , a deep learning model module 204 , a trajectory calculation module 206 , a vehicle steering analysis module 208 and a report generation module 209 . The deep learning model module 204 is connected to the initialization module 202, the trajectory calculation module 206 is connected to the deep learning model module 204, the vehicle steering analysis module 208 is connected to the initialization module 202 and the trajectory calculation module 206, and the report generation module 209 is connected to Vehicle steering analysis module 208 . The operation flow of these modules will be described in detail below with reference to FIG. 2 to FIG. 7 .

Please also refer to FIG. 2 , which is a flowchart of the initialization module 202 in the automated traffic steering quantitative survey report generation system 10 of the present invention. First, in step S10, the initialization module 202 connects the output of the image capture device 12 to read the image of the intersection. In particular, the present invention uses the cv2 function to connect the output of the image capture device 12 to facilitate future image reading The speed is fast and convenient. Next, in step S12, the initialization module 202 loads the default values required by the system, such as a deep learning model data, which includes parameters of the deep learning model and related algorithms. Next, in step S14, at least one determination line is set for the measurement area in the direction of the intersection, so as to facilitate the detection and counting of the incoming and outgoing positions of the traffic flow. In step S16, at least one advanced determination area is set for the measurement area in the direction of the intersection, such as the block behind the zebra crossing, the area divided into four equal parts at the center of the intersection, etc., so as to optimize the determination result of the determination line. fruit. The determination line in step S14 and the advanced determination area in step S16 can also be set externally and then imported into the initialization model 202 .

Then enter the process of the deep learning model module 204, please refer to Figures 3 and 4 at the same time, which are the flow chart of the deep learning model module 204 and the training deep learning model in the automatic traffic steering quantitative survey report output system 10 of the present invention, respectively. The flow chart of the pre-operation of the module 204, and this pre-operation is performed in a back-end server (not shown in the figure), after the pre-operation is completed, the back-end server provides the trained deep learning model To the deep learning model module 204 in the host 20 . First, in step S20, a deep learning model module 204 is trained. Please refer to FIG. 4 for the training method. In step S202 of FIG. 4, at least one public data set of different types of vehicles is collected; then, in step S204, plural demand data of intersections are further collected. In step S206, an image labeling tool (such as Labelimage) is used to label the required data, and a public data set is used to define the types of vehicles, such as trucks, small trucks, passenger cars, buses, etc.; The required data is input into a deep learning neural network for model training, and an initial deep learning model is output after training; finally, in step S209, customized model optimization is performed according to the characteristics of different intersections, such as three-way intersections, parking lots, and snow intersections. In special cases such as traffic jams, there will be special characteristics, which are different from general intersection characteristics. Therefore, the customized model can improve the accuracy or increase the calculation speed, and generate the deep learning model required by the automatic traffic steering quantitative investigation report output system 10. And deploy the optimized deep learning model to the host 20 .

When the pre-operation step S20 of the deep learning model is completed, please go back to step S22 in FIG. 3, where the deep learning model module 204 reads the intersection image, and then inputs the intersection image into the deep learning model in step S24 for analysis. The type and location of at least one vehicle are as described in step S26. In particular, step S26 also adopts edge calculation, which calculates the data of the intersection image into statistical data, not only It can provide real-time analysis data, and it does not need to use a lot of network bandwidth to transmit high-quality images back to the back-end server for calculation and analysis, which greatly reduces the amount of data transmission.

FIG. 5 is a flow chart of the trajectory calculation module 206 in the automated traffic steering quantitative survey report generation system 10 of the present invention. In step S30, the trajectory calculation module 206 reads the type and position of the vehicle from the deep learning module 204 according to the time sequence. Next, in step S32, a model including a linear velocity model and a Kalman filter is used to predict the type and position of the vehicle at the nth time point, and the type and position of the vehicle at the n+1th time point are predicted. In step S34, a kind of algorithm about multi-object tracking (Multiple Object Tracking, MOT) is used to associate and link the vehicles at the nth time point and the n+1th time point, the preferred embodiment is the Hungarian algorithm, Although there are other algorithms with better accuracy than the Hungarian algorithm, since the system of the present invention is installed in a compact and compact mainframe 20, the computing power is relatively limited compared to a mainframe. Therefore, using the Hungarian algorithm can achieve a balance between the tracking efficiency and accuracy, and is a better choice in the multi-target tracking algorithm. Finally, in step S36, the track information of the linked vehicles is sorted to obtain information including the identification number, category and location of the vehicle, wherein the intersection image obtained at each time point is a frame, and the information of a vehicle will be obtained. Identification number, continuous track will have continuous identification number. If there is an obstacle blocking the trajectory, such as a large car blocking the target vehicle, a broken frame will occur.

Next, please refer to FIG. 6 , which is a flowchart of the vehicle steering analysis module 208 in the automated traffic steering quantitative survey report generating system 10 of the present invention. First, in step S40, the vehicle steering analysis module 208 reads the trajectory information of the vehicle transmitted by the trajectory calculation module 206, and then in step S41, it is determined whether any of the trajectory information of the vehicle has a continuous frame break for more than a predetermined time. Set the value, such as whether there is any vehicle whose identification serial number is interrupted by 30 consecutive numbers, which is equal to the number of consecutive interrupted frames reaching 30 frames. like If yes, then it is determined that the vehicle with continuous frame breaks is an abnormal track and has left the measurement area, and the first point information and an end point information of the vehicle entering and leaving the measurement area are generated, and then go to step S42; if not, go back to step S40 to continue detecting . In step S42, the determination line is used to combine the starting point information and the end point information to determine the direction of travel of the vehicle with continuous frame breakage; then, in step S44, the advanced judgment area that the vehicle with the continuous frame breakage has passed through is analyzed, and then in step S46, using The advanced decision area algorithm corrects the in and out directions of vehicles with continuous frame breaks. Finally, in step S48, the entry and exit direction and track information of the vehicle with continuous frame breakage can be output.

Next, please refer to FIG. 7 , which is a flow chart of the report generating module 209 in the automatic traffic steering quantitative survey report generating system 10 of the present invention. In step S50, the report generating module 209 receives the information on the in-out direction and trajectory of the vehicle from the vehicle steering analysis module 208, and then in step S52, according to different measurement areas, different types of data processing are performed on the in-out direction and trajectory information of the vehicle. For example, a large intersection requires two image capture devices 12 to be photographed oppositely, and one of the image capture devices 12 needs to be mirrored so that the directions of the images are consistent. Next, step S54 customizes the adjustment of the output format of the output steering amount report according to the user's needs, including adjusting the passenger car unit (PCU) and category parameters in the output steering amount report. The final step S56 outputs the steering amount report.

8A and 8B are schematic diagrams of setting a determination line and an auxiliary setting of an advanced determination area in the present invention, respectively. Generally speaking, the upper, right, lower, and left intersections of the intersection are represented by A, B, C, and D, respectively, as shown in Figure 8A. Appropriate sensing line segments are set at the position of the vehicle entering and exiting the intersection as the judgment line, for example, the judgment line 30 in the D direction and the judgment line 31 in the C direction. Although the judgment lines in the A and B directions are not marked in the figure, there may still be a judgment line at the two intersections in the A and B directions. The judging method of the judgment line is that as long as the vehicle passes through the two judgment lines, the system 10 can judge according to the position of the line segment. Cut off the direction of the vehicle entering and exiting the intersection. For example, if the vehicle passes through a determination line 30, if it passes another determination line, it can be determined that the vehicle is not going straight, and it is further determined whether the vehicle is turning right or left. For example, if the vehicle passes the determination line 30 and then passes the determination line 31, it can be determined that the vehicle is turning from the D direction to the right direction C. This method is quite intuitive, but when facing a target object with a broken track, for example, when the target vehicle is blocked by a large vehicle 28 or the object detection fails, it is necessary to find the closest determination line according to the distance. This method will cause the position of the target object to shift due to the problem of the 2D picture angle, which will cause errors in the distance, which will affect the judgment.

Therefore, in Fig. 8B, advanced judgment areas 34 and 36 are additionally provided to assist the prediction result based on the judgment of the judgment line. When the trajectory of the target vehicle is blocked by a large object (such as the large vehicle 28 shown in FIG. 8A ) and the trajectory is interrupted, in the face of the trajectory of the broken track, the present invention uses a special corresponding matrix to achieve human-like wisdom To accurately determine the target vehicle that has broken the track. As shown in FIG. 8B , it is assumed that the target vehicle moves from D to B direction, and the starting point is at point S and the end point is at point E because it is blocked by the large vehicle 28 . If only the determination line is used to determine the steering of the vehicle, the starting point will be closer to the determination line of the A azimuth due to the angle problem, and the starting point will be determined as A, but in fact the starting point of the target vehicle is D. If the advanced judgment areas 34 and 36 are added to assist judgment, the computer can learn in advance what the target vehicle's trajectory is when point S is in the advanced judgment area 34 and point E is in the advanced judgment area 36, and indirectly assist the computer to judge the result. Increase the accuracy and speed of the auxiliary judgment in the advanced judgment area. The efficacy analysis of simply using the judgment line and adding the advanced judgment area to assist judgment is shown in Table 1:

From the experimental results, it can be seen that the judgment performance of using the advanced judgment area assistance is better, and only trucks and pickup trucks (pickup-truck) have a decrease in the accuracy rate. However, because trucks and small trucks account for very little in the training set and test set (0.63%, 2.06%), insufficient training samples lead to detection errors. When only the judgment line is used, the result of the judgment error is unexpectedly mapped to the correct direction, which is not due to the high accuracy of the judgment line. The performance of the rest of the vehicles has improved well, and the locomotives with a high proportion (39.60%) have an accuracy improvement of nearly 20%. Due to the small size of the locomotive, the trajectory algorithm will fail to track due to being blocked by larger vehicles (such as buses, trucks, passenger cars, etc.) on the screen, and the trajectory line will be broken. cause errors in judgment. Using the Advanced Judgment Area as an aid will allow the computer to learn in advance how to judge the track break, so as to solve this problem. Through the performance evaluation, it can be concluded that the advanced judgment area has a great function in assisting the computer in judging the steering amount. Although the improvement of taxis in Table 1 is limited, only 2.89%, but it is limited by the accuracy of the detection performance of the system itself and needs to be adjusted. If the number of samples in the system increases, the deep learning model will undergo more training, which will improve the detection performance, so that the impact of the advanced judgment area on the accuracy will gradually become more obvious.

The present invention also has a de-identification module (not shown in the figure). The purpose of de-identification is to hide personal information in the intersection image, such as faces, license plates, etc. The system of the present invention will use a bounding box to frame the target object, and mosaic the image inside the bounding box. Such fuzzy processing to achieve the function of de-identification. The results produced by the system of the present invention are only for the types of the target objects, such as vehicle types, pedestrians, bicycles, etc., to do relevant data analysis, and the final output steering amount report will not leave data related to personal information.

The present invention also has a weather analysis factor module (not shown in the figure). The weather analysis system will be carried out in two phases. The first stage will focus on distinguishing between sunny days and rainy days. The drop object is put into the deep learning model for training and model building, and the object detection of raindrops is carried out on the intersection image. If the detection exceeds a certain threshold, the weather is judged to be rainy. In one embodiment, the deep learning model may perform raindrop detection at intervals of N minutes (N=30) to reduce the detection time and reduce the impact on the original steering system. The second stage is for the cloudy part, which needs to capture the picture in the sky, which is not suitable for all intersections, and the cloudy day has less impact on the traffic, so the priority is placed in the later stage. In one embodiment, the method for discriminating cloudy days uses automatic detection of the sky part in the intersection image, performs RGB analysis on the sky block or directly inputs it into the deep learning model to determine whether the weather is cloudy or not.

To sum up, the present invention provides an automatic traffic steering quantitative survey report production system and method, which adopts an ultra-wide-angle camera, an external real-time analysis AI host, and is equipped with analysis tools such as a deep learning model and a trajectory algorithm to provide a fully automatic Output traffic diversion report, and has the function of de-identification and weather analysis factor. Therefore, compared with the prior art, whether it is manual investigation or analysis using the judgment line, the present invention further combines the setting of the advanced judgment area, which can make the analysis effect better, and achieve the advantages of high accuracy and fast speed.

Only the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Therefore, all equivalent changes or modifications made according to the features and spirits described in the scope of the application of the present invention shall be included in the scope of the application for patent of the present invention.

10: Automated Traffic Steering Quantitative Survey Report Output System

12: Image capture device

20: Host

202: Initialize the module

204: Deep Learning Model Module

206: Trajectory operation module

208: Vehicle Steering Analysis Module

209: Report output module

Claims (18)

An automated traffic steering quantification investigation report output system, comprising: at least one image capture device for capturing images of a plurality of intersections in a measurement area; and at least one host for signal connection to the at least one image capture device to receive the intersections The image is analyzed to generate a steering report. The at least one host includes: a deep learning model module, receiving a deep learning model data to train a deep learning model, and then inputting the intersection images into the depth In the learning model, the type and position of at least one vehicle are analyzed; a trajectory calculation module is connected to the deep learning model module, reads the type and position of the at least one vehicle from the deep learning learning module according to the time sequence, and calculates The trajectory information of the at least one vehicle includes the identification number, type and location of the vehicle; a vehicle steering analysis module is connected to the trajectory computing module, and uses the trajectory information of the vehicle and at least one determination line to determine the travel of the at least one vehicle direction, and when the trajectory information has continuous frame breaks, at least one advanced determination area is used to correct the direction of the at least one vehicle entering and exiting the at least one advanced determination area, the vehicle turns to the analysis module and outputs the at least one vehicle and a report output module, which is connected to the vehicle steering analysis module, and performs data processing on the inbound and outbound direction and trajectory information of the at least one vehicle to generate the steering amount report. The automated traffic steering quantitative investigation report generating system as described in claim 1, further comprising an initialization module, connecting the deep learning model module and the vehicle steering analysis module, and providing the deep learning model data, the at least one judgment line and the at least one advanced determination area. The automatic traffic steering quantification survey report generating system according to claim 2, wherein the initialization module uses the cv2 function to connect the output of the at least one image capturing device to read the intersection images. The automatic traffic steering quantitative survey report generating system according to claim 2, wherein the deep learning model data loaded by the initialization module includes parameters of the deep learning model and related algorithms. The automatic traffic steering quantitative survey report output system as claimed in claim 1, wherein the step of training the deep learning model by the deep learning model module comprises: collecting at least one public data set of different types of vehicles; collecting plural numbers of the intersections demand data; use an image labeling tool to mark the demand data, and use the at least one public data set to define the type of vehicle; input the marked demand data into a deep learning neural network for model training, Output an initial deep learning model; and optimize the model according to the characteristics of different intersections to generate the deep learning model. The automated traffic steering quantification investigation report generating system as claimed in claim 1, wherein after the trajectory computing module reads the type and position of the at least one vehicle, the following steps are further included: using a linear velocity model and a Kalman filter predicting the type and position of the at least one vehicle at the nth time point, and predicting the type and position of the at least one vehicle at the n+1th time point; Using a Hungarian algorithm to associate and link the at least one vehicle at the n th time point and the n+1 th time point; and calculate the track information of the at least one vehicle after the association link. The automated traffic steering quantification investigation report generating system according to claim 1, wherein after the vehicle steering analysis module reads the trajectory information of the at least one vehicle transmitted by the trajectory computing module, it further comprises the following steps: judging whether Any one of the trajectory information of the at least one vehicle has a continuous frame break exceeding a predetermined value, if so, it is determined that the vehicle with the continuous frame break is an abnormal track and has left the measurement area, and the vehicle enters and exits The starting point information and the ending point information of the measurement area, if not, continue to detect; use the at least one determination line to combine the starting point information and the ending point information to determine the traveling direction of the vehicle with continuous frame breaks; and use the The at least one advanced judging area corrects the entry and exit directions of the vehicle with continuous frame breakage, and then outputs the entry and exit direction and track information of the vehicle with continuous frame breakage. The automatic traffic steering quantitative survey report output system as claimed in claim 1, wherein the report output module performs different types of data processing on the inbound and outbound direction and trajectory information of the at least one vehicle according to different measurement areas. The automatic traffic steering quantitative survey report output system according to claim 1, wherein the report output module makes customized adjustments to the output format of the output steering volume report according to requirements, including adjusting the small value in the output steering volume report. Passenger car equivalent (passenger car unit, PCU), category parameters. A method for producing an automated traffic steering quantitative investigation report, comprising the following steps: capturing a plurality of intersection images in a measurement area by using at least one image capture device; the intersection images are transmitted to a host for analysis; one of the hosts The deep learning model module uses a deep learning model data to train a deep learning model, and then inputs the intersection images into the deep learning model to analyze the type and position of at least one vehicle; a trajectory calculation module in the host Read the type and position of the at least one vehicle from the deep learning learning module according to the time series, and calculate the trajectory information of the at least one vehicle, including the identification number, type and position; a vehicle steering analysis module in the host uses The trajectory information of the vehicle and the at least one determination line determine the traveling direction of the at least one vehicle, and when the trajectory information has continuous frame breaks, the at least one advanced determination area is used to correct the at least one vehicle entering and exiting the at least one advanced determination area. direction, the vehicle steering analysis module and output the inbound and outbound direction and trajectory information of the at least one vehicle; and a report output module in the host performs data processing on the inbound and outbound direction and trajectory information of the at least one vehicle to generate the steering volume report. The method for producing an automated traffic steering quantification survey report as claimed in claim 10, further comprising using an initialization module to provide the deep learning model data to the deep learning model module, and providing the at least one determination line of the intersection and the At least one advanced determination area is assigned to the vehicle steering analysis module. The method for producing an automated traffic steering quantitative survey report as claimed in claim 11, wherein the initialization module uses the cv2 function to connect the output of the at least one image capture device to read the intersection images. The method for producing an automated traffic steering quantitative survey report according to claim 11, wherein the deep learning model data loaded by the initialization module includes parameters of the deep learning model and related algorithms. The method for producing an automated traffic steering quantification survey report according to claim 10, wherein the step of training the deep learning model by the deep learning model module comprises: collecting at least one public data set of different types of vehicles; collecting the plural numbers of the intersection. demand data; use an image labeling tool to mark the demand data, and use the at least one public data set to define the type of vehicle; input the marked demand data into a deep learning neural network for model training, Output an initial deep learning model; and optimize the model according to the characteristics of different intersections to generate the deep learning model. The method for generating an automated traffic steering quantification survey report as claimed in claim 10, wherein after the trajectory computing module reads the type and position of the at least one vehicle, it further comprises the following steps: using a linear velocity model and a Kalman filter predicting the type and position of the at least one vehicle at the nth time point, and predicting the type and position of the at least one vehicle at the n+1th time point; Using a Hungarian algorithm to associate and link the at least one vehicle at the n th time point and the n+1 th time point; and calculate the track information of the at least one vehicle after the association link. The method for generating an automated traffic steering quantification survey report according to claim 10, wherein after the vehicle steering analysis module reads the trajectory information of the at least one vehicle transmitted by the trajectory calculation module, it further comprises the following steps: judging whether Any one of the trajectory information of the at least one vehicle has a continuous frame break exceeding a predetermined value, if so, it is determined that the vehicle with the continuous frame break is an abnormal track and has left the measurement area, and the vehicle enters and exits The starting point information and the ending point information of the measurement area, if not, continue to detect; use the at least one determination line to combine the starting point information and the ending point information to determine the traveling direction of the vehicle with continuous frame breaks; and use the The at least one advanced judging area corrects the entry and exit directions of the vehicle with continuous frame breakage, and then outputs the entry and exit direction and track information of the vehicle with continuous frame breakage. The method for generating an automated traffic steering quantitative survey report according to claim 10, wherein the report generating module performs different types of data processing on the inbound and outbound direction and trajectory information of the at least one vehicle according to different measurement areas. The method for producing an automated traffic steering quantitative survey report according to claim 10, wherein the report output module makes customized adjustments to the output format of the output steering quantity report according to requirements, including adjusting the small value in the output steering quantity report. Passenger car equivalent (passenger car unit, PCU), category parameters.

Images