KR102566525B1 - Method and apparatus for analyzing traffic situation - Google Patents

Abstract

Detecting an object in a traffic monitoring image using a first machine learning model, acquiring a trajectory of the object by tracking the detected object, and analyzing the trajectory using a second machine learning model, thereby corresponding to the trajectory A method for analyzing a traffic situation is provided through the step of analyzing an event.

Description

Traffic situation analysis method and device {METHOD AND APPARATUS FOR ANALYZING TRAFFIC SITUATION}

The present disclosure relates to a method and apparatus for analyzing traffic conditions based on real-time monitoring images.

Traffic volume can generally be counted as a simple repetitive task by a user who has viewed recorded images by region and by time, which consumes a lot of time and money. In the conventional real-time vehicle detector, since the loop sensor is installed in each direction on each road, a lot of installation cost is required, and the real-time vehicle detector has a disadvantage in that it is difficult to repair.

In addition, conventional traffic measurement devices use a limited rule-based algorithm tailored to a specific situation in a specific area to recognize an abnormal situation, so that the number of abnormal situations that can be recognized is limited and sensitive to environmental changes.

One embodiment provides a method of analyzing traffic conditions from traffic monitoring images using a machine learning model.

According to one embodiment, a traffic situation analysis method is provided. The traffic situation analysis method includes detecting an object in a traffic monitoring image using a first machine learning model, acquiring a trajectory of the object by tracking the detected object, and trajectory using a second machine learning model. By analyzing, an event corresponding to the trajectory is analyzed.

Objects and their trajectories can be accurately recognized from real-time traffic monitoring images using artificial neural networks, and the occurrence of abnormal events according to various movements of objects can be quickly and accurately determined.

1 is a block diagram illustrating a traffic condition analysis device according to an exemplary embodiment.
2 is a flowchart illustrating a traffic condition analysis method according to an exemplary embodiment.
3 is an example of an object detected in a traffic surveillance image of an intersection according to an exemplary embodiment.
4 is a flowchart illustrating a method of detecting an object according to an exemplary embodiment.
5 is a conceptual diagram illustrating an object detection method according to an exemplary embodiment.
6 is a conceptual diagram illustrating a method of redetecting an object according to an exemplary embodiment.
7 is a flowchart illustrating a method for tracking an object according to an exemplary embodiment.
8 is an example illustrating an object tracking result in an image frame according to an exemplary embodiment.
9 shows events analyzed from trajectory image data according to an embodiment.
10 is a block diagram illustrating a traffic condition analysis device according to another embodiment.

Hereinafter, with reference to the accompanying drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily carry out the present invention. However, the present disclosure may be implemented in many different forms and is not limited to the embodiments described herein. In addition, in order to clearly explain the present description in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

1 is a block diagram illustrating a traffic condition analysis device according to an exemplary embodiment.

Referring to FIG. 1 , a traffic condition analysis apparatus 100 according to an embodiment includes a pre-processing unit 110 and a post-processing unit 120. The pre-processing unit 110 includes an object detection unit 111, an object tracking unit 112, and an object database (DB) 113. The post-processing unit 120 includes a trajectory conversion unit 121, an event analysis unit 122, and an event DB 123. The traffic condition analysis device 100 according to an embodiment may be located around a camera that generates monitoring images of roads and traffic around the roads, or may be located at a location geographically distant from the camera to transmit traffic from the camera through a wired or wireless network. You can receive surveillance images.

The pre-processing unit 110 may detect a moving object from a traffic monitoring image captured in real time using a machine learning model and track the detected object. The pre-processing unit 110 may receive a real-time traffic monitoring image and/or a traffic information image captured for collection of traffic information from at least one camera capturing a road and an area adjacent to the road.

The object detection unit 111 may detect an object from a real-time traffic monitoring image using a machine learning model (object detection model). The object detector 111 may store object detection results in the object DB 113 and perform training and/or re-training of the object detection model using the data stored in the object DB 113.

The object tracker 112 may obtain a trajectory of the object by tracking the detected object.

The object DB 113 may store object detection results and feed back object detection results to the object detection model of the object detection unit 111 during object detection model training. The object DB 113 according to an embodiment may include an initial DB and a result DB. The initial DB may include public DB and pre-collected data released through various paths, and pre-training of the object detection model may be performed using the data included in the initial DB.

Pre-learning of the object detection model may be performed before the traffic situation analysis apparatus 100 according to an embodiment is actually deployed. After the traffic situation analysis device 100 is deployed in practice, the object detection model is re-learned based on the traffic monitoring image and/or the traffic information image delivered in real time, and the re-learned object detection model is used in the traffic monitoring image. An object can be detected in Object information of the detected object may be matched with the reliability (0 to 100%) of the detection result and stored in the result DB. Since the object detection model can be re-learned using data stored in the result DB at predetermined intervals, the traffic situation analysis device 100 can efficiently adapt to the environment of the place where it is located, and object detection performance can be improved. there is. In addition, the traffic situation analysis device 100 may be used for changes in the place where it is placed, for example, changes in the road environment such as lane expansion and lane narrowing, changes in the environment around the road due to various construction works, and changes in the types of objects that appear. can respond effectively. Among the data in the result DB, only data having relatively high reliability may be used when retraining the object detection model. Here, the traffic condition analysis apparatus 100 may include an interface through which a user may select data in a result DB. The interface through which the user can select data in the result DB is learning data such as selection, classification, and removal when the user is notified of the occurrence of a specific situation (eg, a situation in which the detection performance of an object is not improved or deteriorated). can be used to manually perform a reset of

The post-processing unit 120 may convert trajectory information of the tracked object into data for analysis, and analyze an event corresponding to the trajectory of the object from the data for analysis using a machine learning model. Converting trajectory information into data for analysis may be performed according to a 2D image classification method, a 3D image classification method, a point cloud analysis method, or the like. For example, the post-processing unit 120 may obtain a 2D trajectory image by projecting object trajectory information onto a 2D plane, and convert the 2D trajectory image into data for analysis through 2D image classification. If a two-dimensional image is used, the analysis result is easy to display visually, but the temporal meaning is lost and the calculation time may be increased. Alternatively, the post-processing unit 120 may obtain a 3D trajectory image by projecting the trajectory information of the object onto a 3D space, and convert the 3D trajectory image into data for analysis through 3D image classification. When a 3D image is used, both spatial and temporal information of the trajectory can be used together, but the computation time may be longer than that of a 2D image. Alternatively, the post-processing unit 120 may generate analysis data through point cloud analysis using pointer information of object trajectory information. If pointer information is used, space-time information can also be used and time and memory resources can be saved.

The trajectory conversion unit 121 may convert the trajectory of the object into analysis data such as pointer information.

The event analyzer 122 may analyze an event corresponding to the trajectory of the object from data representing the trajectory of the object using a machine learning model (event analysis model).

The event DB 123 may perform event analysis model training by storing event analysis results and feeding back the event analysis results to the event analysis model of the event analyzer 122 . The event DB 123 may include an initial event DB and a result event DB. The initial event DB may be used for pre-training of the event analysis model, and the resulting event DB may be used for periodic or non-periodic re-training of the event analysis model. For example, an event analysis model pre-trained through an initial event DB including photo-collected data may not yet be optimized for a place where the traffic situation analysis apparatus 100 is installed. Accordingly, event analysis performance may be improved by retraining the event analysis model using images collected from the place where the traffic condition analysis device 100 is installed and detected events. Events such as invading the center line or running in reverse can be analyzed as abnormal events, but when the event analysis model is retrained through images collected in real time, it is learned that the frequency of invading the center line or running in reverse is abnormally increased, and event analysis accordingly The model can analyze center line violations or reverse driving as road detour events due to construction. When the construction work is finished, the frequency of reverse driving decreases, and accordingly, the retrained event analysis model may correctly analyze reverse driving events again.

Hereinafter, a traffic situation analysis method of a traffic situation analysis device according to an embodiment will be described in detail with reference to FIG. 2 .

2 is a flowchart illustrating a traffic condition analysis method according to an exemplary embodiment.

Referring to FIG. 2 , the object detection unit 111 of the traffic condition analysis device 100 according to an exemplary embodiment detects an object in a traffic monitoring image using an object detection model (S110). The object detection model is a machine learning model, and the object detection unit 111 includes various machine learning methods for object detection (eg, R-CNN (Region-based Convolutional Neural Networks), SSD (Single Shot Detector), YOLO (you An object detection model is created through a machine learning algorithm such as a deep learning method such as only look once). In order to create an object detection model, the object detection unit 111 is used as learning data for object detection model training, generated in real time or pre-recorded, traffic surveillance images (which may include road surveillance images and road surrounding images). ) can be used. The object detection unit 111 may train an object detection model using information about each image frame of the traffic monitoring image, an object in the image frame, and the type and location (2D or 3D coordinates) of the object.

3 is an example showing objects detected in a traffic monitoring image of an intersection according to an embodiment, FIG. 4 is a flowchart illustrating an object detection method according to an embodiment, and FIG. 5 is an object detection method according to an embodiment. , and FIG. 6 is a conceptual diagram illustrating a method of redetecting an object according to an embodiment.

Referring to FIG. 3 , the object detection unit 111 according to an exemplary embodiment detects a plurality of vehicles moving from north to south as objects, and detects a vehicle waiting at a signal in the east as objects. Objects are indicated by dotted boxes. In FIG. 3 , since an object far from the camera (eg, an object detected at the top of an image frame) is relatively small, the object detector 111 may improve object detection performance through perspective correction.

Referring to FIG. 4 , the object detection unit 111 according to an exemplary embodiment may divide an image frame into a plurality of image blocks according to perspective (S111). Referring to FIG. 5 , an image frame may be divided into 15 large and small image blocks. Here, the size of the image block varies according to the distance from the camera, and one image block may correspond to the same area in real space.

The object detector 111 may convert each image block into a uniform image block having the same size (S112) and detect an object within the uniform image block (S113). Referring to FIG. 5 , one object is detected in each of block 1 , block 2 , block 4 , block 5 , block 7 , block 8 , block 9 , and block 15 . The object detection unit 111 may improve object detection performance even in an area far from the camera by detecting an object in each uniform image block converted to the same size. Afterwards, the object detector 111 may restore the equal image block to the same size as the original image block (S114). That is, referring to FIG. 5 , when the processor (ie, the object detection unit) detects the object, the image frame of the traffic surveillance video is divided into a plurality of image blocks relatively large in size compared to the upper end of the image frame of the traffic monitoring image. and divides the image frame into a plurality of image blocks that are smaller in size compared to the lower portion of the image frame of the traffic surveillance video as the upper portion of the image frame of the traffic surveillance video goes, reduces the size of the relatively large image block, and reduces the size of the relatively small image block. By enlarging, it is converted into a uniform image block of the same size, and then an object is detected within the corresponding image block.

The object DB 113 may store object detection results. The object DB 113 according to an embodiment may match and store object information such as the type of detected object, location (coordinates or area) where the object was detected, and detection reliability with object images.

The object detector 110 according to an embodiment may include a detector (eg, yolo) based on an artificial neural network (eg, a deep neural network (DNN), a convolutional neural network (CNN), etc.) , faster-RCNN, etc.) can be applied to the object detection model. For example, the object detector 110 may repeat prior learning of the object detection model until a predetermined detection reliability is secured, and the object image and object information of the object obtained as a result of the prior learning may be stored in the object DB 113. ) and can be used for retraining the object detection model. The object detection unit 111 can improve the speed and recognition rate of detecting objects in real-time traffic monitoring images by using the repeatedly trained object detection model, and adaptively and quickly recognize newly added types of objects. can do.

Referring back to FIG. 2 , the object tracker 112 tracks the object detected in the traffic monitoring image and obtains the trajectory of the object (S120).

7 is a flowchart illustrating an object tracking method according to an exemplary embodiment, and FIG. 8 is an example illustrating an object tracking result in an image frame according to an exemplary embodiment.

Referring to FIG. 7 , the object tracking unit 112 according to an exemplary embodiment may track objects in successive image frames according to a predetermined tracking algorithm (S121). The predetermined tracking algorithm may include Intersection over Union (IOU), Kernelized Correlation Filters (KCF), Multiple Instance Learning (MIL), and Tracking, learning and detection (TLD). The object tracking unit 112 may additionally use handcraft feature information such as color, size, direction, distance, and feature information of deep learning to improve object tracking performance. A box surrounding an object in FIG. 8 represents an area of the object being tracked within the image frame shown in FIG. 8 . In FIG. 8 , points that reach the object and run in the direction opposite to the moving direction of the object are trajectories of the object, and correspond to the center coordinates of the object box tracked in the previous image frame. Numbers above the object boxes represent information such as identification number, reliability, and moving distance, respectively. For example, '370' on the object box of the object being tracked in the center of FIG. 8 represents the identification number of the object, 0.7 in '0.7, 1.0, 919.8' is the reliability of object detection, and 1.0 is the reliability of object tracking. Reliability, '919.8' represents the pixel distance the tracked object has moved. The movement distance may represent the total distance traveled by the specific object after the tracking of the specific object started, or the total distance traveled by the specific object after being initialized by stopping the object. The movement distance may be initialized to 0 when the tracked object does not move for a predetermined amount of time.

The object tracking unit 112 may manage an object tracking list in order to manage objects that are newly discovered in the image frame or disappear from the image frame. Referring to FIG. 7 , the object tracking unit 112 according to an embodiment calculates a similarity of an object being tracked in two temporally adjacent image frames (eg, frame n and frame n+1) ( S122), similar object pairs may be matched according to the calculated similarity (S123).

For example, when there is a newly detected object in image frame n+1, the object tracking unit 112 may calculate a similarity between the newly detected object and the object tracked up to image frame n. The object tracking unit 112 uses information (IOU, color information, direction of movement, amount of scale change, etc.) of each detected object in consecutively input frames to track the object up to frame n and the object detected in frame n+1. The degree of similarity between them can be calculated. Here, the similarity may be expressed as a similarity matrix representing the similarity between the tracked object being tracked up to frame n and the object detected in frame n+1. Equation 1 represents object information of n tracking objects tracked up to frame n and m detection objects detected in frame n+1, and Table 1 shows a similarity matrix between n tracking objects and m detection objects.

The similarity between the tracking object and the detection object in Table 1 can be calculated from the IOU value between the object information of all tracking objects being estimated in frame n and the rectangular area of the object information of all detection objects in frame n+1. Or, the similarity between the tracking object and the detection object vectorizes object information such as IOU, x coordinate, y coordinate, width, height, scale, and color, and Euclidean between vectorized object information It can be calculated from various distance operations such as distance, Manhattan distance, and Hemming distance.

The object tracking unit 112 may update the object tracking list according to the matching result of similar object pairs (S124). For example, two objects determined to be the same object in the n-th image frame and the n+1-th image frame may be matched as a similar object pair. Then, when there is no object identical to or determined to be identical to the object in the n+1 th image frame in the n+2 th image frame, the object tracking unit 112 determines that the object has disappeared from the monitoring area of the camera and returns to the object tracking list. You can delete pairs of similar objects from .

In the similarity matrix of Table 1, since the first detection object (Detect_1) detected at frame n+1 has a similarity of 100 to the first tracking object (Track_1) among the n tracking objects tracked up to frame n, Object information may be updated in the object tracking list as object information of the first tracking object. In Table 1, the mth detection object Detect_m detected in frame n+1 may have a similarity lower than a predetermined criterion (eg, 90) with all detection objects tracked up to frame n, and in this case, the mth detection object Object information of an object may be updated in the object tracking list as a new tracking object.

Here, when m is greater than n (that is, when there are more detection objects than tracking objects), there are objects that do not correspond to n tracking objects among detection objects in frame n+1. In this case, a detection object that does not have a corresponding tracking object continuously for a predetermined number of frames may be updated in the object tracking list as a new tracking object. Or when n is greater than m (ie, when there are more tracking objects than detection objects), all detection objects in frame n+1 correspond to all tracking objects up to frame n, and there may be tracking objects without corresponding detection objects. can In this case, object information of a tracking object that does not have a corresponding detection object continuously for a predetermined number of frames may be deleted from the object tracking list.

Referring back to FIG. 2 , the trajectory conversion unit 121 according to an embodiment may convert the trajectory of the tracked object into analyzeable data for analysis (S130). For example, the trajectory conversion unit 121 may convert the trajectory of an object into trajectory image data by projecting it onto a 2D image, or into 2D or 3D pointer information.

When the data for analysis is trajectory image data generated from projection on a 2D image, the event analyzer 122 may analyze an event occurring in an object from the trajectory image data through image pattern recognition. 9 shows events analyzed from trajectory image data according to an embodiment.

When the data for analysis is 2-dimensional or 3-dimensional pointer information, the event analysis unit 122 can train the relationship between trajectory information and events using a machine learning model, so the complexity of the classification model can be reduced, and memory Event analysis can be performed with fast processing speed while using less.

Referring to FIG. 2 , the event analyzer 122 according to an embodiment may analyze an event corresponding to a trajectory by analyzing data for analysis using a machine learning model (S140). The event analyzer 122 according to an embodiment analyzes data for analysis using an artificial neural network such as a multilayer perceptron (MLP) and a DNN, so that the trajectory (ie, the movement of the object) is straight, changed direction, It can be classified into events such as U-turn, collision, stop, and lane change. The event analyzer 122 may use environment information of the road, such as lane information, to analyze data for analysis through an artificial neural network. The type of event may be predefined, and a motion of an object that does not correspond to the predefined event may be regarded as an abnormal event.

The event DB 123 may store event analysis results of the event analyzer 122 and provide the stored event analysis results as learning data for training a machine learning model of the event analyzer 122 .

If the same trajectory as the motion of an object that does not correspond to a predefined event frequently occurs, the event analyzer 122 may temporarily determine a frequently occurring abnormal event as a normal event. For example, when a road or its vicinity is under construction, a number of movements of an object (eg, crossing a center line or driving backwards) that do not correspond to a predefined event may occur in the vicinity of the construction area. In this case, if the cumulative number of motions of the object classified as an abnormal event because it does not correspond to a predefined event exceeds a predetermined threshold, the event analyzer 122 converts the motion trajectory of the object classified as an abnormal event into a temporary normal event. event can be determined. The event analyzer 122 matches the temporary normal event with the area corresponding to the temporary normal event (ie, a specific area where a trajectory corresponding to the temporary normal event occurs (eg, the vicinity of a construction area)) to obtain an event DB ( 123) can be stored. That is, a temporary normal event may be considered for analysis of a trajectory occurring in a specific area, and through this, a violation of the center line or reverse driving around a construction area may be temporarily determined to be a normal event. When the road construction is finished, the event analysis unit 122 uses a machine learning model to learn that the occurrence frequency of trajectories classified as temporary normal events has significantly decreased, and the temporary normal events are stored in the event list of the event DB 123. can be deleted When the temporary normal event is deleted from the event DB 123, whether a trajectory occurring in a specific area corresponding to the temporary normal event is normal/abnormal can be determined based on the existing event list.

The event analysis result of the event analyzer 122 may be transmitted to a traffic control system (not shown) along with object images and object information.

As described above, the traffic situation analysis device according to an embodiment recognizes an object and its trajectory from a real-time traffic monitoring image using an artificial neural network, and can quickly and accurately determine the occurrence of abnormal events according to various movements of the object. there is. In addition, the traffic control system measures the traffic volume in the traffic surveillance video based on the image and object information of the object detected and tracked in the traffic surveillance video and the analysis result of the event according to the object's movement trajectory, and detects the occurrence of abnormal situations. can recognize In addition, the traffic control system takes measures to improve the flow of traffic, such as changing the signal system, based on the image of the object, object information, event information according to the movement trajectory of the object, and the above information according to other traffic monitoring images in the surrounding area. can

10 is a block diagram illustrating a traffic condition analysis device according to another embodiment.

A traffic condition analysis device according to an embodiment may be implemented in a computer system, for example, a computer readable medium. Referring to FIG. 9 , a computer system 1000 includes a processor 1010, a memory 1030, an input interface device 1050, an output interface device 1060, and a storage device 1040 communicating through a bus 1070. ) may include at least one of Computer system 1000 may also include a communication device 1020 coupled to a network. The processor 1010 may be a central processing unit (CPU) or a semiconductor device that executes instructions stored in the memory 1030 or the storage device 1040 . The memory 1030 and the storage device 1040 may include various types of volatile or non-volatile storage media. For example, memory can include read only memory (ROM) and random access memory (RAM). In an embodiment of the present description, the memory may be located inside or outside the processor, and the memory may be connected to the processor through various known means. Memory is a volatile or non-volatile storage medium in various forms, and may include, for example, read-only memory (ROM) or random access memory (RAM).

Accordingly, an embodiment of the present invention may be implemented as a computer-implemented method or as a non-transitory computer-readable medium in which computer-executable instructions are stored. In one embodiment, when executed by a processor, the computer readable instructions may perform a method according to at least one aspect of the present disclosure.

The communication device 1020 may transmit or receive a wired signal or a wireless signal.

Meanwhile, the embodiments of the present invention are not implemented only through the devices and/or methods described so far, and may be implemented through a program that realizes functions corresponding to the configuration of the embodiments of the present invention or a recording medium in which the program is recorded. And, such implementation can be easily implemented by those skilled in the art from the description of the above-described embodiment. Specifically, the method (eg, network management method, data transmission method, transmission schedule creation method, etc.) according to an embodiment of the present invention is implemented in the form of program commands that can be executed through various computer means, and is stored in a computer readable medium. can be recorded. Computer readable media may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on a computer readable medium may be specially designed and configured for the embodiments of the present invention, or may be known and usable to those skilled in the art in the field of computer software. A computer-readable recording medium may include a hardware device configured to store and execute program instructions. For example, computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, floptical disks and It may be the same magneto-optical media, ROM, RAM, flash memory, or the like. The program instructions may include high-level language codes that can be executed by a computer through an interpreter, as well as machine language codes generated by a compiler.

Although the embodiments have been described in detail above, the scope of rights is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts defined in the following claims also fall within the scope of rights.

Claims (10)

As a traffic situation analysis method,
Detecting an object in a traffic monitoring image using a first machine learning model;
Tracking the detected object to obtain a trajectory of the object; and
Analyzing an event corresponding to the trajectory by analyzing the trajectory using a second machine learning model,
The step of detecting the object is,
The lower part of the image frame of the traffic surveillance video divides the image frame into a plurality of image blocks relatively large in size compared to the upper part, and the upper part of the image frame of the traffic monitoring video has a smaller size than the lower part of the image frame. segmenting into a plurality of image blocks; and
and converting the relatively large image block into a uniform image block having the same size by reducing the size of the relatively large image block and enlarging the size of the relatively small image block, and then detecting an object in the corresponding image block. analysis method. In paragraph 1,
The step of detecting the object is,
restoring each of the plurality of uniform image blocks in which the object is detected into an original image block;
Further comprising a traffic situation analysis method. In paragraph 1,
Obtaining the trajectory of the object,
Calculating a similarity of an object being tracked in two temporally adjacent image frames of the traffic monitoring image;
Matching similar object pairs according to the calculated similarity; and
Updating an object tracking list according to a matching result of the similar object pair
Including, traffic situation analysis method. In paragraph 3,
Calculating the similarity of the objects,
Calculating the degree of similarity based on object information of all objects being tracked in a specific image frame and object information of objects detected in an image frame next to the specific image frame
Including, traffic situation analysis method. In paragraph 4,
Calculating the similarity based on object information of a tracking object being tracked in the specific image frame and object information of a detection object detected in an image frame next to the specific image frame,
vectorizing object information of the tracking object and object information of the detection object, and calculating the similarity through distance calculation between the vectorized object information;
Including, traffic situation analysis method. As a traffic situation analysis device,
It includes a processor, a memory, and a communication unit, wherein the processor executes a program stored in the memory,
Detecting an object using a first machine learning model in the traffic monitoring image received through the communication unit;
Tracking the detected object to obtain a trajectory of the object; and
Analyzing an event corresponding to the trajectory by analyzing the trajectory using a second machine learning model,
When the processor performs the step of detecting the object,
The lower part of the image frame of the traffic surveillance video divides the image frame into a plurality of image blocks relatively large in size compared to the upper part, and the upper part of the image frame of the traffic monitoring video has a smaller size than the lower part of the image frame. segmenting into a plurality of image blocks; and
By reducing the size of the relatively large image block and enlarging the size of the relatively small image block, converting the relatively large image block into a uniform image block having the same size, and then detecting an object in the corresponding image block,
Traffic situation analysis device. In paragraph 6,
When the processor performs the step of detecting the object,
restoring each of the plurality of uniform image blocks in which the object is detected into an original image block;
Further, the traffic situation analysis device. In paragraph 6,
When the processor performs the step of acquiring the trajectory of the object,
Calculating a similarity of an object being tracked in two temporally adjacent image frames of the traffic monitoring image;
Matching similar object pairs according to the calculated similarity; and
Updating an object tracking list according to a matching result of the similar object pair
To perform, traffic situation analysis device. In paragraph 8,
When the processor performs the step of calculating the similarity of the object,
Calculating the degree of similarity based on object information of all objects being tracked in a specific image frame and object information of objects detected in an image frame next to the specific image frame
To perform, traffic situation analysis device. In paragraph 9,
When the processor performs the step of calculating the similarity based on object information of a tracking object being tracked in the specific image frame and object information of a detection object detected in an image frame next to the specific image frame,
vectorizing object information of the tracking object and object information of the detection object, and calculating the similarity through distance calculation between the vectorized object information;
To perform, traffic situation analysis device.

Images