Intelligent road traffic signal lamp supporting real-time traffic flow perception

Abstract

The application discloses an intelligent road traffic signal lamp supporting real-time traffic flow perception, comprising a multi-modal perception unit, an edge computing control unit, a signal lamp execution unit and a cloud collaborative unit. Compared with the prior art, the application has the following advantages: 1. The perception accuracy and reliability are greatly improved. The triple perception fusion of millimeter wave radar, binocular vision and geomagnetic coil is adopted, which covers three dimensions of side, front and road surface in space and realizes millisecond-level synchronization in time. The three technical means are complementary to each other: the radar is not affected by light and weather, the vision provides fine classification and pedestrian detection, and the geomagnetic provides accurate through time verification. After fusion, the vehicle flow detection accuracy can reach more than 95%, and it still works stably in bad environment, solving the problems of many blind areas and poor robustness of single perception means; forward timing, from passive response to active prediction; combination of single-point intelligence and regional cooperation; multi-level degradation mechanism, strong operation stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control technology, specifically to an intelligent road traffic signal light that supports real-time traffic flow perception. Background Technology

[0002] Traffic lights are the core control facilities at urban road intersections, and their timing directly affects the intersection's traffic efficiency and vehicle delay time. Traditional traffic lights mostly adopt fixed-time-period timing schemes, which preset fixed cycles and phase durations for different periods such as morning peak, off-peak, and evening peak based on historical statistical data. This cannot respond to the dynamic fluctuations of real-time traffic flow, often resulting in a resource mismatch phenomenon of "one side congested, the other side idle."

[0003] Existing adaptive traffic signal systems are mainly divided into two categories: one is a single-point sensing control based on geomagnetic coils and microwave radar, which can only detect whether a vehicle has arrived, but cannot obtain complete fine parameters such as queue length and vehicle type composition, and the timing adjustment granularity is relatively coarse; the other is a visual detection scheme based on monocular video, which can identify vehicles and count their number, but the detection accuracy drops significantly in adverse environments such as night, rain, fog, strong light, and backlight, and there are blind spots, making it difficult to cover the traffic flow status of long road sections upstream of the intersection.

[0004] In addition, existing solutions generally have three shortcomings: First, the sensing methods are singular, lacking multi-source redundancy and fusion mechanisms, resulting in insufficient reliability and environmental adaptability; second, the control logic is mostly "reactive," adjusting timing only based on the current queuing status, unable to predict traffic flow trends, and only responding passively after congestion occurs; third, the coordination capability between intersections is weak, mostly single-point independent control, making it difficult to form regional green wave linkages, which restricts the overall improvement of traffic efficiency on main roads. Summary of the Invention

[0005] To address the shortcomings mentioned in the background art, this invention provides an intelligent road traffic signal light that supports real-time traffic flow perception.

[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: an intelligent road traffic signal light supporting real-time traffic flow perception, comprising a multimodal perception unit, an edge computing control unit, a traffic signal execution unit, and a cloud-based coordination unit; the multimodal perception unit consists of a millimeter-wave radar module, a binocular vision module, and a geomagnetic detection module, which synchronously collect traffic flow data from lanes in all directions at the intersection from lateral spatial dimensions, frontal visual dimensions, and road surface contact dimensions, respectively; the edge computing control unit is deployed locally at the intersection, receives the three-channel data from the multimodal perception unit, performs spatiotemporal fusion processing, outputs real-time traffic flow status parameters and short-term traffic flow prediction results, and dynamically generates a traffic signal timing scheme accordingly; the cloud-based coordination unit communicates with the edge computing control units at multiple intersections to achieve regional-level traffic flow collaborative scheduling and dynamic optimization of green wave zones.

[0007] Furthermore, the edge computing control unit has a built-in multi-source data spatiotemporal registration module. It uses timestamp alignment and spatial coordinate mapping algorithms to uniformly map the point cloud data of millimeter-wave radar, the target detection box data of binocular vision, and the through pulse data of geomagnetic coils to the same lane coordinate system. Through weighted fusion, it obtains four core parameters for each lane: vehicle queue length, number of vehicles, average vehicle speed, and vehicle type classification.

[0008] Furthermore, the edge computing control unit also has a built-in short-term traffic flow prediction module, which uses a lightweight GRU time-series prediction model to output traffic flow prediction values ​​for each direction for the next 30 seconds to 3 minutes based on the fused traffic flow data from the past 5 minutes. The timing scheme is generated by taking the current real-time traffic flow data and the predicted traffic flow data as inputs, and using a fuzzy control algorithm to calculate the green light duration and phase switching sequence for each phase.

[0009] Furthermore, the binocular vision module integrates a lightweight YOLO target detection model, which can simultaneously identify three types of targets: motor vehicles, non-motor vehicles, and pedestrians, and separately count the number of targets in the queuing area before the stop line and the upstream detection area of ​​the intersection entrance lane; the millimeter-wave radar module operates in the 77GHz frequency band, with a detection range of not less than 150 meters, and can supplement visual detection blind spots in rainy, foggy, and low-light environments at night.

[0010] Furthermore, the cloud-based collaborative unit constructs a regional traffic spatiotemporal map, collects real-time and predicted traffic flow data from all intersections within its jurisdiction, and adjusts the phase difference and cycle duration of each intersection in real time through a dynamic programming algorithm to generate a dynamic green wave. When the predicted traffic flow in a certain direction exceeds a threshold, it automatically links with upstream and downstream intersections to extend the green light window for that direction.

[0011] Furthermore, the edge computing control unit is equipped with a fault self-diagnosis and degraded operation module. When any sensing module malfunctions, it automatically switches to the other two sensing data channels to maintain adaptive control. When all sensing modules fail, it automatically degrades to a multi-period fixed timing mode and sends a fault alarm to the cloud.

[0012] Furthermore, the geomagnetic detection module uses wireless geomagnetic nodes, which are buried in front of the stop line of each lane and 50 meters and 100 meters upstream. It detects vehicle passing events by changing the magnetic induction intensity and uploads the data to the edge computing control unit in a low-power wireless manner.

[0013] Furthermore, the fuzzy control algorithm uses queue length, predicted traffic increment, and pedestrian crossing requests as input variables to output the adjustment amount for extending or shortening the green light. The green light duration for each phase is set with an upper and lower limit to ensure minimum right-of-way in each direction.

[0014] The advantages of this intelligent road traffic light supporting real-time traffic flow perception compared to existing technologies are: 1. Significantly improved perception accuracy and reliability. It employs a triple perception fusion of millimeter-wave radar, binocular vision, and geomagnetic coils, spatially covering three dimensions: lateral, frontal, and road surface, and achieving millisecond-level synchronization. The three technologies complement each other: radar is unaffected by lighting and weather, vision provides fine classification and pedestrian detection, and geomagnetic coils provide accurate passage time verification. After fusion, the traffic flow detection accuracy can reach over 95%, maintaining stable operation even in harsh environments, solving the problems of numerous blind spots and poor robustness associated with single perception methods.

[0015] 2. Proactive timing, shifting from passive response to proactive prediction. By introducing a lightweight short-term traffic flow prediction model at the edge, traffic flow trends can be predicted up to 3 minutes in advance. Timing decisions take into account both the current state and future trends, allowing for the extension of green light duration before peak traffic arrives, thus avoiding queue congestion. Compared to traditional sensor-based control, this can reduce the average delay at intersections by 15% to 25%.

[0016] 3. Combining single-point intelligence with regional collaboration. The edge computing unit ensures millisecond-level response at a single intersection without relying on the network; the cloud collaboration unit enables multi-intersection linkage and dynamically generates green wave bands, which not only ensures the reliability of single points but also improves the overall traffic efficiency of the region, forming a two-level control architecture of "edge autonomous decision-making + cloud global optimization".

[0017] 4. Multi-level degradation mechanism ensures strong operational stability. It features three operating modes: full-sensing adaptive mode, dual-sensing degradation mode, and fixed-time backup mode. The failure of any single device will not affect the overall system operation, significantly improving system availability and operational fault tolerance. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of an intelligent road traffic signal light that supports real-time traffic flow perception.

[0019] As shown in the figure: 1. Millimeter-wave radar module; 2. Binocular vision module; 3. Geomagnetic detection module; 4. Edge computing control unit; 5. Cloud collaboration unit; 6. Traffic light execution unit. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings.

[0021] Combined with appendix Figure 1 A smart road traffic signal that supports real-time traffic flow perception includes a multimodal perception unit, an edge computing control unit 4, a signal light execution unit 6, and a cloud collaboration unit 5.

[0022] The multimodal sensing unit consists of a millimeter-wave radar module 1, a binocular vision module 2, and a geomagnetic detection module 3. The millimeter-wave radar module 1 uses the 77GHz frequency band and is installed on the crossbar at the intersection, providing lateral coverage of all approach lanes with a detection range of at least 150 meters, outputting point cloud data of vehicle position, speed, and angle. The binocular vision module 2 is installed on the top of the traffic light pole, facing the stop line of the approach lane. It integrates a lightweight target detection model and can simultaneously identify motor vehicles, non-motor vehicles, and pedestrians, outputting target category, location box, and quantity statistics. The geomagnetic detection module 3 consists of multiple wireless geomagnetic nodes, buried in the center of the lane before the stop line and 50 meters and 100 meters upstream, respectively. It detects vehicle passage events through changes in magnetic induction and uploads data wirelessly in a low-power manner. All three sensing devices are synchronized to ensure data timestamp synchronization.

[0023] The edge computing control unit 4 is deployed within the traffic signal cabinet at the intersection and includes a multi-source data spatiotemporal registration module, a short-term traffic flow prediction module, a fuzzy control timing module, and a fault self-diagnosis module. The spatiotemporal registration module maps radar point clouds, visual detection boxes, and geomagnetic pulses to the same lane coordinate system, and outputs four core parameters for each lane through a weighted fusion algorithm: queue length, total number of vehicles, average vehicle speed, and vehicle type classification. The short-term traffic flow prediction module, based on fused time-series data from the past 5 minutes, runs a lightweight GRU neural network model and outputs predicted traffic flow values ​​for each direction for the next 30 seconds to 3 minutes. The fuzzy control timing module takes real-time queue length, predicted traffic volume increments, and pedestrian crossing requests as inputs, and outputs the green light duration adjustment for each phase and the phase switching sequence, while setting upper and lower limits for green light duration to ensure fairness in passage.

[0024] The traffic light execution unit 6 receives instructions from the edge computing control unit 4 and drives the switching of traffic light colors in each direction, retaining safe transition phases such as flashing yellow and all red, and the switching process complies with national standard timing requirements.

[0025] The cloud-based collaborative unit 5 connects to the edge computing control unit 4 at all intersections within the jurisdiction via 5G / fiber optics to construct a regional traffic spatiotemporal map. It collects real-time traffic flow data and prediction data from all intersections and uses dynamic programming algorithms to calculate the optimal phase difference and cycle duration of each intersection in real time, with the optimization goal of minimizing the total regional delay and maximizing the traffic volume. This forms a dynamic green wave. When a surge in traffic flow is detected in a certain main direction, the unit automatically links with upstream and downstream intersections to extend the passage window in that direction, achieving regional-level collaborative scheduling.

[0026] The fault self-diagnosis module continuously monitors the data validity and output stability of the three sensing devices. When any module's data is abnormal or interrupted, it automatically removes the data from that channel and uses the fusion results of the other two channels to maintain adaptive control. When two or more sensing channels fail, it automatically degrades to a multi-period fixed timing mode and pushes fault alarm information to the cloud.

[0027] The corresponding control method includes the following steps: S1. Three-way sensing devices simultaneously collect traffic flow data from all directions and stamp them with a unified timestamp; S2. At the edge, spatiotemporal registration and multi-source fusion are performed to output refined traffic flow parameters; S3. Run a lightweight prediction model based on historical time-series data to generate short-term traffic flow predictions; S4. Combine real-time status with prediction results to calculate dynamic timing scheme using fuzzy control algorithm; S5: Locally execute timing and upload data to the cloud. The cloud performs regional collaborative optimization and then provides feedback to correct the parameters.

[0028] In a specific implementation of this invention, Example 1: Single-point adaptive control example at a crossroads This embodiment is applied to the intersection of a main road and a secondary road in the city. The east-west direction is the main road with six lanes in both directions; the north-south direction is the secondary road with four lanes in both directions.

[0029] A 77GHz millimeter-wave radar module 1 is installed above each of the east and west entrance lanes of the intersection, mounted laterally, with a beam covering all lanes of the entrance lanes and a detection range of 150 meters. It can detect vehicle position, speed, and direction of travel. A binocular vision module 2 is installed at the top of each traffic light pole at each entrance lane, with the lens facing the stop line. The field of view covers an area 80 meters in front of the stop line. It has a built-in lightweight YOLOv8n detection model and can identify four types of targets: cars, trucks, non-motorized vehicles, and pedestrians. A wireless geomagnetic node 3 is buried 50 meters and 100 meters upstream of the stop line in each lane, for a total of 20 geomagnetic nodes, used for accurate detection of vehicle passage time and speed estimation.

[0030] All sensing devices achieve microsecond-level time synchronization via the PTP precision time protocol, and the data is uniformly fed into the edge computing control unit 4. The edge computing control unit adopts an industrial-grade edge gateway, equipped with a quad-core ARM processor and an NPU acceleration module with 8 TOPS computing power, and is deployed in the traffic signal cabinet at the intersection.

[0031] The work process is as follows: Data acquisition and synchronization: The millimeter-wave radar outputs a point cloud target list at a frequency of 20Hz, the binocular vision outputs the detection results at a frequency of 15fps, and the geomagnetic node uploads pulse data in an event-triggered manner. All three are accompanied by high-precision timestamps.

[0032] Spatiotemporal registration and fusion: The spatiotemporal registration module first converts the radar polar coordinate data into geodetic plane coordinates, and maps them to the same lane coordinate system as the pixel coordinates of the visual detection box through calibration parameters; then, it performs sliding window alignment based on the timestamp, and correlates and matches the three data sources within the same time window. The fusion algorithm adopts the weighted DS evidence theory to fuse and decide the vehicle count and queue length of the three detection results, and outputs the real-time vehicle count, queue length, average vehicle speed, and vehicle type ratio for each lane.

[0033] Short-term traffic flow prediction: The short-term traffic flow prediction module slides through the fused data from the past 5 minutes, constructs time-series samples with a 30-second step size, inputs them into a lightweight GRU model, and predicts the traffic flow in each direction at four time points: 30 seconds, 1 minute, 2 minutes, and 3 minutes. After pruning and quantization, the model runs on the NPU, with a single inference time of less than 20 milliseconds.

[0034] Dynamic timing calculation: The fuzzy control timing module uses queue length, predicted traffic increase, and pedestrian crossing requests as input variables to establish a three-input, single-output fuzzy inference system. The input variables are divided into three fuzzy levels: "small," "medium," and "large," with the output being the green light duration adjustment. Each phase has a minimum green light duration of 10 seconds and a maximum green light duration of 60 seconds, ensuring basic right-of-way in all directions. When the queue length in a certain direction exceeds the threshold and the predicted traffic flow continues to increase, the green light duration for that phase is automatically extended; when the lane is cleared and no new traffic flow is predicted, the current phase ends early and the transition to the next phase begins.

[0035] Signal execution: The timing scheme is sent to the traffic light execution unit through the IO interface to drive the switching of red, yellow and green traffic lights. A 3-second yellow light and a 2-second all-red clearing period are inserted between phase switching, which complies with the requirements of GB14886 standard.

[0036] In this embodiment, the system can dynamically adjust the duration of each phase according to the real-time traffic flow. Compared with the traditional fixed timing, the average vehicle delay at the intersection is reduced by about 22%, and the traffic capacity is increased by about 18%.

[0037] Example 2: Implementation of Green Wave Coordination on Main Roads with Bus Priority This embodiment is applied to five consecutive intersections on an east-west main road in the city, implementing regional collaborative green wave control and public transport priority strategies.

[0038] Each intersection is equipped with a multimodal sensing unit and an edge computing control unit as described in Example 1, and each intersection is connected to the regional traffic control cloud via fiber optic cable. In addition, dedicated geomagnetic detection nodes and bus RFID identification devices are added to the bus lanes on the main roads to detect bus arrivals.

[0039] The cloud-based collaborative unit constructs a regional traffic spatiotemporal map encompassing five intersections, receiving real-time traffic flow data, prediction data, and public transport detection information uploaded from each intersection. Its operational mechanism is as follows: Basic green wave dynamic generation: Based on the real-time cycle and predicted traffic flow at each intersection, the cloud uses a dynamic programming algorithm to solve for the optimal phase difference between intersections, aiming to minimize the number of stops and maximize the travel speed of vehicles on the main road, thus generating a dynamic green wave band. The green wave speed is adjusted in real time according to the average vehicle speed on the main road, unlike traditional fixed-speed green waves. When a traffic peak is predicted at a certain intersection, the cloud fine-tunes the phase difference between upstream and downstream intersections in advance, so that the green wave bandwidth adapts to the increased traffic flow.

[0040] Bus priority linkage control: When a bus is detected approaching at an intersection during a red light phase, the edge computing control unit first determines whether bus priority can be achieved locally by compressing other phase times. If the compression exceeds a threshold, a bus priority request is sent to the cloud. The cloud coordinates the phase difference between upstream and downstream adjacent intersections, allocating a bus priority green light window for the intersection without disrupting the overall green wave, thus balancing bus priority and the green wave band.

[0041] Anomaly event linkage: When an intersection detects an anomaly such as a traffic accident or vehicle congestion through multimodal sensing, it immediately reports it to the cloud. The cloud automatically adjusts the timing strategies of surrounding intersections, appropriately shortening the green light duration at intersections upstream of the incident to control inflow, and extending the green light duration at intersections in diversion directions to accelerate evacuation, forming a regional coordinated emergency control system.

[0042] This embodiment realizes dynamic green wave coordination at continuous intersections on main roads, reducing the average travel time of vehicles in the main direction by about 25% and improving the on-time rate of buses by about 30%.

[0043] Example 3: Degraded Operation Implementation in Harsh Environments This embodiment describes the system's adaptive degradation operation mechanism under extreme weather and equipment failure scenarios.

[0044] The system's self-diagnosis module continuously monitors the health status of the three sensing devices, with monitoring indicators including data output frequency, valid data range, reasonableness of target quantity, and historical consistency. Each device is assigned a health score, and a score below a set threshold is considered abnormal.

[0045] Scenario 1: In a nighttime downpour, the binocular vision module experiences a decrease in detection accuracy due to road surface reflections and raindrop obstruction, causing the health score to drop below the threshold. At this time, the fault self-diagnosis module automatically masks the visual detection data and switches to a dual-source fusion mode using millimeter-wave radar and a geomagnetic coil. The radar is unaffected by rain and lighting conditions, providing long-distance traffic flow detection; the geomagnetic coil provides accurate vehicle counts at stop lines, and the fusion of these two technologies still yields accurate queue length and traffic flow data. The system maintains full adaptive timing functionality, with only a slight decrease in the detection accuracy for non-motorized vehicles and pedestrians. In this case, pedestrian crossings are switched to a button-triggered mode.

[0046] Scenario 2: Road construction damages the geomagnetic coil circuit in a certain direction, causing data interruption at the geomagnetic nodes. The system automatically switches to a dual-source fusion mode of millimeter-wave radar + binocular vision. The fine classification capability of vision complements the long-range detection capability of radar, and the system's adaptive function remains unaffected.

[0047] Scenario 3: Communication fiber optic cable interruption, cloud connection lost. Each intersection automatically switches to a completely independent single-point adaptive mode, all calculations are completed at the local edge, timing function operates normally, only regional green wave coordination is temporarily disabled. After communication is restored, it automatically reconnects to the cloud and synchronizes data.

[0048] Scenario 4: Two of the three sensing devices fail simultaneously. The system automatically downgrades to a multi-period fixed timing mode, calling up the four built-in fixed timing schemes for morning peak, noon, evening peak, and night to run according to the time periods. At the same time, it pushes a level-two fault alarm to the cloud and the operation and maintenance platform to notify personnel for emergency repairs.

[0049] Through the above four-level degradation mechanism, the system can maintain basic traffic control functions under various local failures and harsh environments, with availability reaching over 99.9%, which is significantly better than the single perception solution.

[0050] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent road traffic signal light supporting real-time traffic flow perception, characterized in that, It includes a multimodal perception unit, an edge computing control unit (4), a traffic light execution unit (6), and a cloud collaboration unit (5). The multimodal perception unit consists of a millimeter-wave radar module (1), a binocular vision module (2), and a geomagnetic detection module (3). The three modules synchronously collect traffic flow data of lanes in all directions of the intersection from the lateral spatial dimension, the frontal visual dimension, and the road surface contact dimension, respectively. The edge computing control unit (4) is deployed locally at the intersection. It receives the three-way data from the multimodal perception unit and performs spatiotemporal fusion processing. It outputs real-time traffic flow status parameters and short-term traffic flow prediction results, and dynamically generates traffic light timing schemes accordingly. The cloud collaboration unit (5) communicates with the edge computing control units (4) of multiple intersections to realize regional traffic flow collaborative scheduling and dynamic optimization of green wave.

2. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 1, characterized in that, The edge computing control unit (4) has a built-in multi-source data spatiotemporal registration module. It uses timestamp alignment and spatial coordinate mapping algorithm to uniformly map the millimeter-wave radar point cloud data, binocular vision target detection box data and geomagnetic coil through pulse data to the same lane coordinate system. Through weighted fusion, it obtains four core parameters for each lane: vehicle queue length, number of vehicles, average vehicle speed and vehicle type classification.

3. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 2, characterized in that, The edge computing control unit (4) also has a built-in short-term traffic flow prediction module. Based on the fused traffic flow data in the past 5 minutes, it uses a lightweight GRU time-series prediction model to output the traffic flow prediction values ​​for each direction in the next 30 seconds to 3 minutes. The timing scheme is generated by taking the current real-time traffic flow data and the predicted traffic flow data as inputs and using a fuzzy control algorithm to calculate the green light duration and phase switching sequence for each phase.

4. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 1, characterized in that, The binocular vision module (2) integrates a lightweight YOLO target detection model, which can simultaneously identify three types of targets: motor vehicles, non-motor vehicles and pedestrians, and count the number of targets in the queuing area in front of the stop line and the upstream detection area of ​​the intersection entrance lane respectively; the millimeter-wave radar module (1) operates in the 77GHz band, with a detection distance of not less than 150 meters, and can supplement the visual detection blind area in rain, fog and low light conditions at night.

5. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 1, characterized in that, The cloud-based collaborative unit (5) constructs a regional traffic spatiotemporal map, collects real-time and predicted traffic flow data of all intersections within its jurisdiction, and adjusts the phase difference and cycle duration of each intersection in real time through a dynamic programming algorithm to generate a dynamic green wave. When the predicted traffic flow in a certain direction exceeds the threshold, it automatically links the upstream and downstream intersections to extend the green light passage window in that direction.

6. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 1, characterized in that, The edge computing control unit (4) is equipped with a fault self-diagnosis and degraded operation module. When any sensing module malfunctions, it automatically switches to the other two sensing data to maintain adaptive control. When all sensing modules fail, the system automatically downgrades to a multi-period fixed timing mode and sends a fault alarm to the cloud.

7. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 1, characterized in that, The geomagnetic detection module (3) uses a wireless geomagnetic node, which is buried in front of the stop line of each lane and 50 meters and 100 meters upstream. It detects vehicle passing events by changing the magnetic induction intensity and uploads the data to the edge computing control unit (4) in a low-power wireless manner.

8. The intelligent road traffic signal light supporting real-time traffic flow perception according to claim 3, characterized in that, The fuzzy control algorithm uses queue length, predicted traffic increment, and pedestrian crossing requests as input variables, and outputs the adjustment amount for extending or shortening the green light. The green light duration for each phase is set with an upper and lower limit value to ensure minimum right-of-way in each direction.

Images