1 Introduction
The detection of traffic violations belongs to the high-level content of intelligent monitoring and is a manifestation of event detection, behavior understanding and description. Relatively speaking, most of the previous studies have focused on the underlying visual problems such as motion detection and tracking, and there has been little research in this area. In recent years, with the rapid development of economy and road traffic, the number of motor vehicles has continued to rise, and traffic violations have also increased simultaneously. This raises higher requirements for road traffic management. Research on this topic has also become hot.
In traditional applications, there are many sensor-specific targeted monitoring and handling of various types of traffic violations. This method is somewhat expensive, such as radar and laser; some need to destroy the road surface, and periodically replace equipment, such as the ground coil. Dealing with a variety of illegal types at the same time is often a handy move, which makes it increasingly less adaptable to current trends in traffic management applications.
Based on the image processing algorithm and the principle of pattern recognition, this paper analyzes the vehicle traffic behavior based on the vehicle trajectory, greatly improves the road security department's monitoring of unattended intersection violations, and greatly reduces the project implementation cost.
2. Algorithm flow and trajectory acquisition 2.1 Algorithm flow framework
This paper firstly obtains the target initial information through the vehicle video detection algorithm, uses the mean-shift algorithm and the Kalman filter algorithm to achieve the target's video tracking, and then statistically analyzes the trajectory, obtains the trajectory characteristics of the target's actual driving direction, and uses the on-site rules (signal lights State, lane attributes, guidance rules, etc.) are assisted and make a judgment of illegal behavior.
Figure 1 Flow of traffic violation detection algorithm
2.2 Video Detection
The video triggering is based on the result of the detection of the motion of the vehicle after the smart image recognition, and automatic back-shooting of all vehicles passing through the monitored lane. The vehicle motion detection result integrates the motion frame difference information, vehicle characteristics (vehicle license plate, vehicle tail structure characteristics, AdaBoost training characteristics, etc.) and the tracking state of the vehicle, and finally forms the judgment of the arrival and departure states of the vehicle, with good real-time performance and stability.
2.3 Vehicle Video Tracking
Video tracking is an effective way for us to obtain real-time location information after the target vehicle leaves the video trigger area. The mean shift algorithm (Mean-shift) [2] is a non-parametric probability density estimation algorithm, which can converge to the local maximum of the probability density function quickly and iteratively, so it has a very high application value in the field of target real-time tracking.
The Mean-shift algorithm itself is to find the best matching candidate target by continuously iterating the original region of the target. When the target motion speed is large, redundant iteration or local convergence is easy to occur. For this case, Kalman filtering is used. 3] Model each target's movement process, reduce the iteration range through Kalman's prediction, reduce the amount of calculation, and reduce the influence of noise at the same time to obtain the optimal estimate of the target.
2.4 Trajectory Acquisition and Example
We save the target vehicle from the field of view to the trajectory that leaves the field of view, and use it as the basis for follow-up analysis of traffic violations. According to the "Road Traffic Safety Illegal Behavior Image Forensics Technical Specification (2009)", the key frame traffic The status of the signal lamp is saved, and the site parameters such as the position of the stop line, the position of the lane line, the lane guidance rule, and the road prohibition flag are set.
Figure 2 site trace
3. Trajectory Characteristics of Vehicles and Traffic Illegal Classification 3.1 Traffic Illegal Classification
According to the status of vehicles and China's road traffic safety rules [1], we can divide some major traffic violations as follows:
Table 1 Illegal type Classification Position Illegal Compaction lane line, (double) yellow line and other vertical lines Change lanes, red light, pressure stop line, etc. Lateral line speed violation limit Area speed continues to be 0 (illegal parking)
Vehicle speed is greater than or less than the specified limit value violation of the violation of the route signal does not follow the guide sign illegal illegal occupation of non-motorized lanes driving violations U-turn, turning retrograde vehicles illegal large trucks banned, yellow card vehicles banned motor vehicles use dedicated lanes where position violation refers to The positional relationship between the vehicle and the lane prohibition marking at a certain time violates the relevant provisions of the Road Traffic Safety Law. The violation of the route indicates that the target vehicle's trajectory during a certain period of time is inconsistent with the setting rules of the current road section. Vehicle violation is defined as inconsistency between the type of vehicle and the permitted vehicle type of the lane. When comparing various types of traffic violations, it was discovered that some violations of laws and regulations involve a certain inclusion relationship. For example, if a real lane change lane is preceded by a laneway line, a breach of a traffic light must be preceded by a red stop line, a clear distinction must be made between such lane violations. The situation, the target vehicle trajectory will be its obvious characteristics.
The detection of vehicle violations may not be based on vehicle trajectories, so this article only proposes not to proceed.
3.2 Position violation trajectory analysis
Location violation means that the positional relationship between the vehicle and the lane prohibition marking at a certain time violates the relevant provisions of the Road Traffic Safety Law. This includes three key points, one is the target location, the second is the prohibition of marking positions, and the third is traffic regulations.
Taking the "pressure line" as an example, the traffic law stipulates that the yellow and white solid lines are prohibited marking lines. When the position of the vehicle body overlaps with the lane boundary line, it can be considered as a pressure line. At this point, it is not necessary to pursue its trajectory.
However, the traffic regulations also define that the lanes for the rolling of solid rolling lines for vehicles should be “solid lane change lanesâ€. Its difference from "pressing line" can be understood as follows. Please see the figure below:
Figure 3 Road intersection line, solid line lane change, machine occupancy scenario simulation
As can be seen from the figure, when the target vehicle lane attribution changes, the horizontal distance between the trajectory point and the same lane line has a U-shaped transition on the horizontal axis, and at this time, it can be identified as a "solid line change lane. ". When a vehicle is driven by a motorized lane (LANE3) through a non-motorized vehicle lane, it is defined as “the vehicle is occupying a non-motorized vehicle lane†that is illegal and referred to as “machine-ownedâ€. Others do "press (yellow, white) line" treatment. In other words, once the vehicle changes lanes in violation of the solid line, it will inevitably press the line, and vice versa.
In the same situation, there is also the phenomenon that the former is preceded by the “red light pressure stop line†and the “violation signal lightâ€. Such situations need to be combined with other attributes of the target to achieve differentiation, such as target speed, lanes, etc.
3.3 Speed ​​violations and their trajectory characteristics
From Table 1, we know that speed violation means that the speed attribute of the target vehicle violates the current road traffic regulations. The speed attribute here mainly refers to whether the speed exists and its specific value.
The two-dimensional image processing is mainly the operation of image pixels. However, due to the loss of field depth information, accurate calculation of the target vehicle speed involves perspective transformation or calibration of the site. Both of these requirements are difficult to meet in practice. Therefore, versatility is limited. This article defines the speed as the ratio of the track adjacent point coordinates to the frame interval including (horizontal speed, vertical speed, direction) in units of (pixels/frames). Since the origin of the image is located in its upper left corner, we set Velocity<0 as the positive direction, based on which a qualitative analysis of speed violations is performed.
Below, we analyze the speed characteristics of illegal parking.
Figure 4 Simulation of illegal parking at road intersections
When the speed of the vehicle in the prohibited area is 0 and the duration exceeds the legal time, it can be considered as a violation. Here, the "prohibited area" generally includes a zebra crossing area, a crossing yellow line area, and the like.
3.4 Analysis of route violation and its trajectory
As described in 3.1 above, the route violation refers to the fact that the trajectory of the target vehicle does not conform to the established traffic rules of the current road section. We still use the “road intersection†as an example to analyze three kinds of trajectories, going straight, turning left, and turning right. The schematic diagram is as follows:
Fig. 5 Scenarios of Vehicle Routes at Road Intersections
It is easy to see from Fig. 5 that there is a big difference in the displacement in the horizontal direction for the straight, left-turn, and right-turn trajectories. We show the following change diagrams corresponding to the left-turning, right-turning, and straight-through vehicle trajectories as a comparison and at the same time given the change process.
Figure 6 Horizontal and vertical displacements of a straight trajectory
From figures 6, 7, and 8, we see that the straight trajectory is relatively smooth or slowly changing. The turning trajectory has a more obvious trend of change, in which the left turn track increases in a negative direction as a general trend, and the right turn track takes the opposite direction.
From a statistical point of view, the straight trajectory has more trajectory stability and convergence than the turning trajectory. In the turning trajectory, the right-turn trajectory is characterized by its "small turn", and its vertical displacement changes also appear more discrete. In the actual judgment, we use the law of positive and negative excursion direction, with the negative deviation as the left turn, the positive deviation as the right turn, and with reference to the characteristics of its vertical change, effectively distinguish the left-turn track and the right turn Track. Based on this, combined with lane guidance rules and signal lights real-time status, the relevant violations were accurately distinguished and captured.
4, statistical data results
Based on this method, the traffic flow from the south to the north section of a road intersection in a city in Jiangsu Province was tested from 12 to 15 points. The test results are as follows:
Table 4 Traffic Illegal Behavior Test Results Statistics Illegal Classification Illegal Types Illegal Capture Rate Illegal Capture Effective Position Illegal Compaction Line, (Double) Yellow Line 99.60% 98.73%
Change of real line 90.38% 95.19%
When the red light pressure stop line and other horizontal lines 92.59% 96.32%
Speed ​​violation limit Regional speed continues to be 0 (illegal parking) 90.02% 89.27%
Vehicle speed is greater than or less than the specified limit value does not test the line violation violation signal lamp 97.70% 95.17%
Do not follow the guide sign 95.75% 96.08%
Illegal occupation of non-motorized traffic 98.54% 93.85%
Turning illegally 98.81% 96.83%
Retrograde 96.44% 100%
Vehicles illegally banned by large-size vehicles, (Yellow Label Vehicles banned from the line) It is necessary to explain that the use of dedicated lanes for motor vehicles is not tested. Because it is impossible to effectively count the number of actual violations of road sections or intersections, the field environment is forcibly set or modified during testing. The method, for example, the test violation of the signal that is set to test the time period for the continuous red light, test the red light when the pressure stop line, then the actual line will be read down. In this way, a total of about 2,766 cars were tested at the scene.
5, summary
Based on the vehicle trajectory and combined with relevant traffic safety regulations, this paper forms a method that can intelligently capture and classify a variety of common traffic violations. In the test results also obtained a more ideal results, effectively solve the traditional program of bulky, expensive, single-function shortcomings, and has good practicality and scalability.
The detection of traffic violations belongs to the high-level content of intelligent monitoring and is a manifestation of event detection, behavior understanding and description. Relatively speaking, most of the previous studies have focused on the underlying visual problems such as motion detection and tracking, and there has been little research in this area. In recent years, with the rapid development of economy and road traffic, the number of motor vehicles has continued to rise, and traffic violations have also increased simultaneously. This raises higher requirements for road traffic management. Research on this topic has also become hot.
In traditional applications, there are many sensor-specific targeted monitoring and handling of various types of traffic violations. This method is somewhat expensive, such as radar and laser; some need to destroy the road surface, and periodically replace equipment, such as the ground coil. Dealing with a variety of illegal types at the same time is often a handy move, which makes it increasingly less adaptable to current trends in traffic management applications.
Based on the image processing algorithm and the principle of pattern recognition, this paper analyzes the vehicle traffic behavior based on the vehicle trajectory, greatly improves the road security department's monitoring of unattended intersection violations, and greatly reduces the project implementation cost.
2. Algorithm flow and trajectory acquisition 2.1 Algorithm flow framework
This paper firstly obtains the target initial information through the vehicle video detection algorithm, uses the mean-shift algorithm and the Kalman filter algorithm to achieve the target's video tracking, and then statistically analyzes the trajectory, obtains the trajectory characteristics of the target's actual driving direction, and uses the on-site rules (signal lights State, lane attributes, guidance rules, etc.) are assisted and make a judgment of illegal behavior.
Figure 1 Flow of traffic violation detection algorithm
2.2 Video Detection
The video triggering is based on the result of the detection of the motion of the vehicle after the smart image recognition, and automatic back-shooting of all vehicles passing through the monitored lane. The vehicle motion detection result integrates the motion frame difference information, vehicle characteristics (vehicle license plate, vehicle tail structure characteristics, AdaBoost training characteristics, etc.) and the tracking state of the vehicle, and finally forms the judgment of the arrival and departure states of the vehicle, with good real-time performance and stability.
2.3 Vehicle Video Tracking
Video tracking is an effective way for us to obtain real-time location information after the target vehicle leaves the video trigger area. The mean shift algorithm (Mean-shift) [2] is a non-parametric probability density estimation algorithm, which can converge to the local maximum of the probability density function quickly and iteratively, so it has a very high application value in the field of target real-time tracking.
The Mean-shift algorithm itself is to find the best matching candidate target by continuously iterating the original region of the target. When the target motion speed is large, redundant iteration or local convergence is easy to occur. For this case, Kalman filtering is used. 3] Model each target's movement process, reduce the iteration range through Kalman's prediction, reduce the amount of calculation, and reduce the influence of noise at the same time to obtain the optimal estimate of the target.
2.4 Trajectory Acquisition and Example
We save the target vehicle from the field of view to the trajectory that leaves the field of view, and use it as the basis for follow-up analysis of traffic violations. According to the "Road Traffic Safety Illegal Behavior Image Forensics Technical Specification (2009)", the key frame traffic The status of the signal lamp is saved, and the site parameters such as the position of the stop line, the position of the lane line, the lane guidance rule, and the road prohibition flag are set.
Figure 2 site trace
3. Trajectory Characteristics of Vehicles and Traffic Illegal Classification 3.1 Traffic Illegal Classification
According to the status of vehicles and China's road traffic safety rules [1], we can divide some major traffic violations as follows:
Table 1 Illegal type Classification Position Illegal Compaction lane line, (double) yellow line and other vertical lines Change lanes, red light, pressure stop line, etc. Lateral line speed violation limit Area speed continues to be 0 (illegal parking)
Vehicle speed is greater than or less than the specified limit value violation of the violation of the route signal does not follow the guide sign illegal illegal occupation of non-motorized lanes driving violations U-turn, turning retrograde vehicles illegal large trucks banned, yellow card vehicles banned motor vehicles use dedicated lanes where position violation refers to The positional relationship between the vehicle and the lane prohibition marking at a certain time violates the relevant provisions of the Road Traffic Safety Law. The violation of the route indicates that the target vehicle's trajectory during a certain period of time is inconsistent with the setting rules of the current road section. Vehicle violation is defined as inconsistency between the type of vehicle and the permitted vehicle type of the lane. When comparing various types of traffic violations, it was discovered that some violations of laws and regulations involve a certain inclusion relationship. For example, if a real lane change lane is preceded by a laneway line, a breach of a traffic light must be preceded by a red stop line, a clear distinction must be made between such lane violations. The situation, the target vehicle trajectory will be its obvious characteristics.
The detection of vehicle violations may not be based on vehicle trajectories, so this article only proposes not to proceed.
3.2 Position violation trajectory analysis
Location violation means that the positional relationship between the vehicle and the lane prohibition marking at a certain time violates the relevant provisions of the Road Traffic Safety Law. This includes three key points, one is the target location, the second is the prohibition of marking positions, and the third is traffic regulations.
Taking the "pressure line" as an example, the traffic law stipulates that the yellow and white solid lines are prohibited marking lines. When the position of the vehicle body overlaps with the lane boundary line, it can be considered as a pressure line. At this point, it is not necessary to pursue its trajectory.
However, the traffic regulations also define that the lanes for the rolling of solid rolling lines for vehicles should be “solid lane change lanesâ€. Its difference from "pressing line" can be understood as follows. Please see the figure below:
Figure 3 Road intersection line, solid line lane change, machine occupancy scenario simulation
As can be seen from the figure, when the target vehicle lane attribution changes, the horizontal distance between the trajectory point and the same lane line has a U-shaped transition on the horizontal axis, and at this time, it can be identified as a "solid line change lane. ". When a vehicle is driven by a motorized lane (LANE3) through a non-motorized vehicle lane, it is defined as “the vehicle is occupying a non-motorized vehicle lane†that is illegal and referred to as “machine-ownedâ€. Others do "press (yellow, white) line" treatment. In other words, once the vehicle changes lanes in violation of the solid line, it will inevitably press the line, and vice versa.
In the same situation, there is also the phenomenon that the former is preceded by the “red light pressure stop line†and the “violation signal lightâ€. Such situations need to be combined with other attributes of the target to achieve differentiation, such as target speed, lanes, etc.
3.3 Speed ​​violations and their trajectory characteristics
From Table 1, we know that speed violation means that the speed attribute of the target vehicle violates the current road traffic regulations. The speed attribute here mainly refers to whether the speed exists and its specific value.
The two-dimensional image processing is mainly the operation of image pixels. However, due to the loss of field depth information, accurate calculation of the target vehicle speed involves perspective transformation or calibration of the site. Both of these requirements are difficult to meet in practice. Therefore, versatility is limited. This article defines the speed as the ratio of the track adjacent point coordinates to the frame interval including (horizontal speed, vertical speed, direction) in units of (pixels/frames). Since the origin of the image is located in its upper left corner, we set Velocity<0 as the positive direction, based on which a qualitative analysis of speed violations is performed.
Below, we analyze the speed characteristics of illegal parking.
Figure 4 Simulation of illegal parking at road intersections
When the speed of the vehicle in the prohibited area is 0 and the duration exceeds the legal time, it can be considered as a violation. Here, the "prohibited area" generally includes a zebra crossing area, a crossing yellow line area, and the like.
3.4 Analysis of route violation and its trajectory
As described in 3.1 above, the route violation refers to the fact that the trajectory of the target vehicle does not conform to the established traffic rules of the current road section. We still use the “road intersection†as an example to analyze three kinds of trajectories, going straight, turning left, and turning right. The schematic diagram is as follows:
Fig. 5 Scenarios of Vehicle Routes at Road Intersections
It is easy to see from Fig. 5 that there is a big difference in the displacement in the horizontal direction for the straight, left-turn, and right-turn trajectories. We show the following change diagrams corresponding to the left-turning, right-turning, and straight-through vehicle trajectories as a comparison and at the same time given the change process.
Figure 6 Horizontal and vertical displacements of a straight trajectory
From figures 6, 7, and 8, we see that the straight trajectory is relatively smooth or slowly changing. The turning trajectory has a more obvious trend of change, in which the left turn track increases in a negative direction as a general trend, and the right turn track takes the opposite direction.
From a statistical point of view, the straight trajectory has more trajectory stability and convergence than the turning trajectory. In the turning trajectory, the right-turn trajectory is characterized by its "small turn", and its vertical displacement changes also appear more discrete. In the actual judgment, we use the law of positive and negative excursion direction, with the negative deviation as the left turn, the positive deviation as the right turn, and with reference to the characteristics of its vertical change, effectively distinguish the left-turn track and the right turn Track. Based on this, combined with lane guidance rules and signal lights real-time status, the relevant violations were accurately distinguished and captured.
4, statistical data results
Based on this method, the traffic flow from the south to the north section of a road intersection in a city in Jiangsu Province was tested from 12 to 15 points. The test results are as follows:
Table 4 Traffic Illegal Behavior Test Results Statistics Illegal Classification Illegal Types Illegal Capture Rate Illegal Capture Effective Position Illegal Compaction Line, (Double) Yellow Line 99.60% 98.73%
Change of real line 90.38% 95.19%
When the red light pressure stop line and other horizontal lines 92.59% 96.32%
Speed ​​violation limit Regional speed continues to be 0 (illegal parking) 90.02% 89.27%
Vehicle speed is greater than or less than the specified limit value does not test the line violation violation signal lamp 97.70% 95.17%
Do not follow the guide sign 95.75% 96.08%
Illegal occupation of non-motorized traffic 98.54% 93.85%
Turning illegally 98.81% 96.83%
Retrograde 96.44% 100%
Vehicles illegally banned by large-size vehicles, (Yellow Label Vehicles banned from the line) It is necessary to explain that the use of dedicated lanes for motor vehicles is not tested. Because it is impossible to effectively count the number of actual violations of road sections or intersections, the field environment is forcibly set or modified during testing. The method, for example, the test violation of the signal that is set to test the time period for the continuous red light, test the red light when the pressure stop line, then the actual line will be read down. In this way, a total of about 2,766 cars were tested at the scene.
5, summary
Based on the vehicle trajectory and combined with relevant traffic safety regulations, this paper forms a method that can intelligently capture and classify a variety of common traffic violations. In the test results also obtained a more ideal results, effectively solve the traditional program of bulky, expensive, single-function shortcomings, and has good practicality and scalability.
1 Can stop anywhere when sliding.
2 Strong cord keep high using age at least 5 years.
3 Simple assembly directly onto window frames using screws.
|
Material |
Alloy aluminum |
|
Color |
By RAL color code or by customer`s request |
|
Max.width |
1100 |
|
Max.height |
1600 |
|
Mounting condition |
Should be installed inside of a window with outside opening way |
|
Close way |
Cord control |
|
Installation way |
Recess Fit |
Pleated Screen Window,Plisse Screen Window,Pleated Insect Screen For Roof Window,Aluminum Plisse Screen Window
Hebei Charlotte Enterprise Co., Ltd. , http://www.cnscreening.com