Dynamic phase jumping method and system based on real-time vision requirements

Abstract

The application discloses a dynamic phase skipping method and system based on real-time visual demand, and the method comprises the following steps: S1, acquiring real-time vehicle demand numbers of each preset phase by a body intelligent device arranged at an intersection at the end of each signal period, wherein the real-time vehicle demand numbers are obtained through local visual inference on an intersection video stream; S2, scanning the real-time vehicle demand numbers of each phase in a preset order; S3, giving a green light time to the phase with a demand number greater than zero, and directly skipping the phase with a demand number equal to zero; S4, if the real-time vehicle demand numbers of all phases are all zero, controlling the signal machine to enter a full red state or a yellow flashing state; wherein the green light time is dynamically adjusted according to the real-time vehicle demand number; the vehicle demand numbers of each phase are acquired through real-time visual detection at the intersection, the phase without demand is directly skipped, the phase with demand is dynamically allocated with a green light time, and adaptive phase control with zero or extremely low empty release is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic signal control technology, and in particular to a dynamic phase-jumping method and system based on real-time visual requirements. Background Technology

[0002] Traffic signal phase skipping is an important technical means to improve the efficiency of intersection traffic. Its core idea is: in a certain signal cycle, if a certain phase has no traffic demand (i.e., no vehicles are waiting in that direction), then that phase is skipped, and the time is allocated to other phases with demand, thereby reducing the green light idle time.

[0003] Current phase-jumping technologies mainly rely on the following detection methods: 1. Geomagnetic coil detection: Geomagnetic coils are buried under the lane to detect the presence of vehicles through electromagnetic induction. When there is no vehicle above the coil, it is determined that there is no need for that phase, and a phase jump can be performed. The drawbacks of this method are: the installation of geomagnetic coils requires road excavation, maintenance is difficult, and it can only detect the presence of vehicles, but cannot distinguish vehicle types or detect pedestrians, resulting in a single detection dimension.

[0004] 2. Video Detector + Central Platform: Video detectors are installed at intersections, and the video is uploaded to a central platform. The central platform analyzes the number of vehicles in each direction and then issues a phase-changing command. The drawbacks of this solution are: reliance on network transmission and a central platform, high deployment costs, large response latency, and data security risks.

[0005] 3. Phase skipping based on queue length: For example, Chinese patent CN104794910A discloses a phase skipping method based on queue length, which detects the queue length through geomagnetism or video, and executes phase skipping when the queue length is less than a threshold and the waiting time is less than a threshold. The drawback of this scheme is that: the phase skipping judgment depends on a combined threshold of queue length and waiting time, the logic is complex, and parameter tuning is difficult; moreover, "shortening the phase duration" and "directly skipping the phase" are two different control strategies. The former still gives the phase a certain green light time (which may result in idle discharge), while the latter does not give any green light time (zero idle discharge).

[0006] In addition, existing phase-jumping technologies also have the following common problems: Separation of detection and decision-making: The detector is only responsible for collecting data, while the decision-making is completed by the central platform or signal machine. A communication link is required between the two, which increases the system complexity and the number of failure points.

[0007] Unable to respond in real time: The geomagnetic coil can only detect the presence of a vehicle, but cannot track its dynamics in real time; the network latency of the central platform solution causes a delay in phase-jumping decisions.

[0008] Pedestrian needs are difficult to meet: geomagnetic coils cannot detect pedestrians, and video detectors uploading pedestrian data raises privacy concerns, making it difficult for pedestrian phases to effectively coordinate with phase-jumping mechanisms.

[0009] Therefore, there is an urgent need for a dynamic phase-jumping scheme that integrates detection and decision-making, provides real-time response, takes into account the needs of pedestrians, and does not require external systems. Summary of the Invention

[0010] The present invention addresses the shortcomings of the prior art and aims to provide a dynamic phase-jumping method and system based on real-time visual demand. It obtains the number of vehicles in each phase through real-time visual detection at the intersection, skips phases with no demand, and dynamically allocates green light time to phases with demand, thereby achieving adaptive phase control with zero or extremely low idle time.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic phase-hopping method based on real-time vision requirements includes the following steps: S1. At the end of each signal cycle, the embodied intelligent device deployed at the intersection obtains the real-time vehicle demand for each preset phase. The real-time vehicle demand is obtained through local visual reasoning of the intersection video stream. S2. Scan the real-time vehicle demand for each phase in a preset order; S3. Give green light time to phases with a demand number greater than zero, and skip phases with a demand number equal to zero. S4. If the real-time vehicle demand for all phases is zero, the control signal will enter the all-red state or the yellow flashing state. The green light time is dynamically adjusted based on the real-time demand for vehicles.

[0012] As a preferred embodiment, step S1 includes: S11. Collect intersection video streams through the AI ​​camera module built into the embodied smart device; S12. The built-in neural network processing unit of the embodied intelligent device performs local target detection and tracking inference on the video stream to obtain the bounding box, category and tracking ID of each traffic target; S13. Based on the predefined lane area polygon, count the number of tracking targets falling into the lane area corresponding to each phase to obtain the real-time vehicle demand for each preset phase. Local target detection and tracking inference are completed within the embodied intelligent device, eliminating the need to upload video data to an external system.

[0013] As a preferred option, in step S12, the local object detection adopts a lightweight object detection model deployed on the NPU after INT8 quantization. The lightweight object detection model is trained for quantization perception in the intersection scene.

[0014] As a preferred option, step S3, which grants green light time to phases with a demand number greater than zero, includes: Provide the minimum green light time preset for the phase; If the real-time vehicle demand for a phase is still greater than zero after the minimum green light time has ended, the green light time will be extended by a preset unit of time. The number of extensions shall not exceed the preset number; The total green light time for a phase shall not exceed the preset maximum green light time.

[0015] As a preferred option, an early termination mechanism is also included: If the real-time vehicle demand for a phase drops to zero after the green light time for that phase exceeds the minimum green light time, the green light time for that phase will end early and switch to the yellow light state.

[0016] As a preferred embodiment, in step S3, directly skipping phases with a demand number of zero includes: When a phase with a demand of zero is detected, no green light time is given to that phase, and the signal status of that phase is directly maintained as red. Continue scanning the next phase until a phase with a demand greater than zero is found or the scanning of all phases is completed.

[0017] As a preferred option, a dynamic phase priority adjustment step is also included; Calculate the demand urgency of each phase based on the real-time vehicle demand and the historical average demand for each phase; Reorder the phase scan sequence from highest to lowest urgency. Among them, the urgency of demand is positively correlated with the real-time number of vehicles in demand and with the waiting time since the last release.

[0018] As a preferred option, a pedestrian phase insertion step is also included: Real-time detection of pedestrian waiting numbers in zebra crossing areas using local visual reasoning; When the number of pedestrians waiting exceeds a preset threshold, mark the pedestrian phase demand. Pedestrian phases are inserted into the vehicle phase sequence, and the green light duration for pedestrian phases is dynamically adjusted based on the number of waiting pedestrians.

[0019] As a preferred option, a night mode switching step is also included: Determine whether it is nighttime based on the exposure parameters of the AI ​​camera module or the average brightness of the image; During the nighttime period, if the real-time vehicle demand for all phases is zero, the control signal will enter a flashing yellow state. When the real-time vehicle demand number for any phase is detected to be greater than zero, the yellow flashing state is exited and normal adaptive phase-skipping control is restored.

[0020] As a preferred option, fail-safe procedures are also included: When the visual reasoning function of the embodied smart device malfunctions or the video stream is interrupted, it automatically switches to the preset fixed timing mode. The timing scheme for the fixed timing mode is stored in the local storage unit of the smart device.

[0021] A dynamic phase-hopping system based on real-time vision requirements, used to implement the aforementioned dynamic phase-hopping method based on real-time vision requirements, includes: The visual perception module, deployed at the intersection, is used to collect video streams from the intersection and perform local visual reasoning on the video streams to obtain the real-time vehicle demand number for each preset phase. The phase skipping decision module is used to scan the real-time vehicle demand of each phase in a preset order, give green light time to phases with a demand of greater than zero, and skip phases with a demand of zero directly. The signal execution module is used to issue control commands to the signal controller based on the decision result of the phase jump decision module; The visual perception module, phase-jumping decision module, and signal execution module are integrated into the same embodied intelligent device, which does not need to communicate with an external central platform during operation.

[0022] As a preferred embodiment, the visual perception module includes AI camera module, used to capture video streams at intersections; The neural network processing unit is used for local object detection and tracking inference on the video stream; The lane mapping unit is used to count the number of targets in each lane corresponding to each phase based on a predefined lane area polygon.

[0023] As a preferred option, the phase-jumping decision module includes The demand scanning unit is used to scan the real-time vehicle demand of each phase in a preset order; The green light allocation unit is used to allocate green light time to phases with a demand greater than zero. The green light time includes the minimum green light time, the extendable green light time, and the maximum green light time limit. Skip control unit, used to directly skip phases with zero demand, without allocating green light time; The early termination trigger unit is used to end the green light early when the green light time exceeds the minimum green light time and the demand number drops to zero.

[0024] As a preferred option, it also includes The priority calculation unit is used to calculate the urgency of demand based on the real-time number of vehicles in each phase and the waiting time, and to rearrange the phase scanning order according to the urgency of demand.

[0025] Compared with the prior art, the present invention has significant advantages and beneficial effects, specifically: 1. Zero or very low idling: Instead of shortening the green light time, phases with no demand are skipped directly (without any green light time). This fundamentally eliminates the idling during the green light of that phase, and the idling rate can be reduced to below 5%.

[0026] 2. Real-time visual detection: The vehicle demand in each phase is obtained in real time through local visual inference. It has rich detection dimensions (can distinguish vehicle type and detect pedestrians), fast response speed (millisecond level), and does not require infrastructure such as geomagnetic coils.

[0027] 3. Dynamic green light allocation: The green light duration for phases with demand is dynamically adjusted based on the real-time number of vehicles. The green light duration is extended when there are many vehicles and ended earlier when there are few vehicles, thus achieving a refined allocation of time resources.

[0028] 4. Integrated detection and decision-making: Visual perception, phase-hopping decision-making, and signal execution are integrated into the same device, eliminating the need for external communication links and the risk of network latency and communication failure.

[0029] 5. Pedestrian safety is taken into account: Pedestrian waiting needs are automatically identified through visual detection, and pedestrian phases are intelligently inserted. Pedestrian phases cannot be skipped, thus ensuring pedestrians' right of way.

[0030] 6. Intelligent switching at night: When there are no cars at night, it automatically switches to yellow flashing mode and resumes normal control when there are cars, balancing nighttime traffic efficiency and safety.

[0031] To more clearly illustrate the structural features, technical means, and specific objectives and functions achieved by the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0032] Figure 1 A general control block diagram of a dynamic phase-jumping system based on real-time vision requirements provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the visual perception module structure provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the phase-jumping decision module provided in an embodiment of the present invention; Figure 4 A flowchart of a dynamic phase-jumping method based on real-time vision requirements provided in an embodiment of the present invention; Figure 5 A flowchart of dynamic phase priority adjustment provided for embodiments of the present invention; Figure 6 A flowchart of pedestrian phase insertion provided for embodiments of the present invention; Figure 7 A flowchart of the night mode provided in an embodiment of the present invention.

[0033] Explanation of icon numbers: Visual perception module 10; AI camera module 11; Neural network processing unit 12; Lane mapping unit 13; Phase jump decision module 20; Demand scanning unit 21; Green light allocation unit 22; Skip control unit 23; Early interrupt trigger unit 24; Signal execution module 30; Local physical interface 31; Embodied intelligent device 40; Priority calculation unit 41; Signal 50. Detailed Implementation

[0034] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0035] like Figures 1 to 7 As shown, a dynamic phase-hopping method based on real-time visual requirements includes an early termination mechanism, a dynamic phase priority adjustment step, a pedestrian phase insertion step, a night mode switching step, a fault-safe step, and steps S1 to S4.

[0036] This embodiment also includes an early termination mechanism: If the real-time vehicle demand for a phase drops to zero after the green light time for that phase exceeds the minimum green light time, the green light time for that phase will end early and switch to yellow light mode.

[0037] In this embodiment, a phase priority dynamic adjustment step is also included; The demand urgency for each phase is calculated based on the real-time vehicle demand and the historical average demand for each phase. In this embodiment, the demand urgency for each phase is calculated using the following formula: Urgency(i) =α* Count(i) +β* WaitTime(i) / T_avg, where: Count(i) is the real-time vehicle demand for phase i, WaitTime(i) is the waiting time for phase i since the last release, T_avg is the historical average signal cycle duration, and α and β are weighting coefficients (e.g., α=1.0, β=0.5).

[0038] The phase scanning order is rearranged from highest to lowest urgency; for example, if P3 has the highest urgency, it will be scanned first, followed by P1, P4, and P2. This dynamic priority adjustment mechanism can better respond to sudden traffic demands and reduce waiting time in high-demand directions.

[0039] Among them, the urgency of demand is positively correlated with the real-time number of vehicles in demand and with the waiting time since the last release.

[0040] In this embodiment, a pedestrian phase insertion step is also included: While detecting motor vehicles, the visual perception module 10 also detects the number of pedestrians waiting in the zebra crossing area in real time through local visual reasoning. When the number of pedestrians waiting exceeds a preset threshold (e.g., 1 person), mark the pedestrian phase demand; The phase-jumping decision module 20 inserts a pedestrian phase into the vehicle phase sequence, and the green light duration for the pedestrian phase is dynamically adjusted based on the number of waiting pedestrians. Specifically: After the current vehicle phase ends and before the next vehicle phase begins, if there are pedestrian demands from multiple directions, they are inserted sequentially in descending order of the number of pedestrians waiting.

[0041] The green light duration for pedestrians is dynamically adjusted based on the number of pedestrians waiting. Base duration: 15 seconds; extended by 2 seconds for each additional person; maximum duration: 30 seconds.

[0042] Pedestrian phases have the highest priority and cannot be skipped. Even if a vehicle approaching from the vehicular lane is detected during a pedestrian phase, the system must wait until the pedestrian phase ends before switching to another phase.

[0043] In this embodiment, a night mode switching step is also included: The visual perception module 10 determines whether it is nighttime based on the exposure parameters (such as gain value and shutter speed) or average image brightness of the AI ​​camera module 11. When the gain value is greater than a preset threshold (such as 50dB) or the average image brightness is less than a preset threshold (such as 30 gray levels), it is determined to be nighttime.

[0044] During the nighttime period, if the real-time vehicle demand for all phases is zero, the control signal 50 enters the yellow flashing state (all yellow lights flash at a frequency of 1Hz). When the real-time vehicle demand for any phase is detected to be greater than zero, the yellow flashing mode exits and resumes normal adaptive phase-hopping control. Night mode reduces energy consumption and driver waiting time when there are no vehicles, while the yellow flashing mode reminds the driver to slow down.

[0045] This embodiment also includes a fail-safe step: When the visual reasoning function of the embodied smart device 40 fails or the video stream is interrupted, it automatically switches to the preset fixed timing mode. The timing scheme for the fixed timing mode is stored in the local storage unit of the embodied smart device 40.

[0046] Step S1: At the end of each signal cycle, the embodied intelligent device 40 deployed at the intersection obtains the real-time vehicle demand number for each preset phase. The real-time vehicle demand number is obtained through local visual inference of the intersection video stream.

[0047] In this embodiment, step S1 includes: S11. The AI ​​camera module 11 built into the embodied smart device 40 collects the intersection video stream; the video stream is transmitted inside the device in RTSP protocol, with a resolution of not less than 1920×1080 and a frame rate of not less than 15FPS.

[0048] S12: The neural network processing unit 12 built into the embodied intelligent device 40 performs local object detection and tracking inference on the video stream to obtain the bounding boxes, categories, and tracking IDs of each traffic target. In this embodiment, the local object detection in step S12 uses a lightweight object detection model deployed on the NPU after INT8 quantization. The lightweight object detection model is trained for quantization perception in intersection scenarios. Preferably, the lightweight object detection model is a YOLOv8n model deployed on the NPU after INT8 quantization (such as the NPU of Rockchip RK3588), with a model size of approximately 3MB, supporting the detection of categories such as vehicles, pedestrians, and non-motorized vehicles. The single-frame inference latency is approximately 25 milliseconds. The detection output includes the bounding box coordinates (x1, y1, x2, y2), category label, and confidence score of each target. Subsequently, a simplified DeepSORT algorithm based on IOU matching is used to perform multi-target tracking on the detection results of consecutive frames, assigning a unique tracking ID to each target.

[0049] S13, the lane mapping unit 13 counts the number of tracking targets falling into the lane area corresponding to each phase according to the predefined lane area polygon, and obtains the real-time vehicle demand number for each preset phase. In this embodiment, the predefined lane area polygons are pre-defined using a configuration tool. For a standard intersection, the following phases and their corresponding lanes are typically defined: Phase P1 (North-South Straight): Corresponds to the polygons of the northbound straight lane and the southbound straight lane; Phase P2 (North-South Left Turn): Corresponds to the polygon of the northbound left turn lane and the polygon of the southbound left turn lane; Phase P3 (East-West Straight): Corresponds to the polygons of the eastbound straight lane and the westbound straight lane; Phase P4 (East-West Left Turn): Corresponds to the polygon of the eastbound left turn lane and the polygon of the westbound left turn lane.

[0050] For each tracked target, calculate the coordinates (cx, cy) of its bounding box center point and determine whether the center point falls within the lane polygon corresponding to a certain phase. Count the number of tracked targets falling within the lane polygons of each phase to obtain the real-time vehicle demand for each preset phase. For example, if there are 5 tracked targets within the lane polygon corresponding to P1, then the real-time vehicle demand for P1 is 5.

[0051] The local target detection and tracking inference are completed within the embodied intelligent device 40, without the need to upload video data to an external system.

[0052] Step S2: Scan the real-time vehicle demand for each phase in a preset order. In this embodiment, the preset order is usually determined according to the traffic organization rules of the intersection, for example: P1 -> P2 -> P3 -> P4. The scan starts from the next phase after the current phase (cyclic scan).

[0053] Step S3: Grant green light time to phases with a demand number greater than zero, and skip phases with a demand number equal to zero. In this embodiment, granting green light time to phases with a demand number greater than zero in step S3 includes: A minimum green light time is preset for the phase; preferably, the minimum green light time is 7 seconds to ensure that vehicles that have entered the intersection can pass safely.

[0054] If the real-time vehicle demand for a phase is still greater than zero after the minimum green light time has ended, the green light time is extended by a preset unit of time; preferably, the preset unit of time of green light time is 5 seconds.

[0055] The number of extensions shall not exceed a preset number, and once this number is reached, no further extensions will be made even if there are still vehicles. Preferably, the preset number of extensions is 3.

[0056] The total green light time for a phase shall not exceed the preset maximum green light time. Preferably, the preset maximum green light time is 60 seconds.

[0057] If the real-time vehicle demand drops to zero during the green light period (all vehicles have passed) and the green light time has exceeded the minimum green light time, the early termination mechanism is triggered, ending the green light early and switching to the yellow light state.

[0058] Phases with a demand of zero are skipped directly, including: When a phase with a demand of zero is detected, no green light time is given to that phase, and the signal status of that phase is directly maintained as red. Continue scanning the next phase until a phase with a demand greater than zero is found or the scanning of all phases is completed.

[0059] Step S4: If the real-time vehicle demand for all phases is zero, the control signal 50 enters the all-red state or the yellow flashing state; the green light time is dynamically adjusted according to the real-time vehicle demand.

[0060] After scanning all phases, if the real-time vehicle demand for all phases is zero (i.e., there are no vehicles at the intersection), then the control signal 50 enters a fully red state. During nighttime hours (e.g., 22:00 to 06:00), it can be configured to flash yellow to remind drivers to slow down before proceeding.

[0061] In this embodiment, it can be applied in different scenarios.

[0062] Scene 1: Intersection during off-peak hours.

[0063] For example, at intersections with sparse and random traffic during off-peak hours, the system can detect the number of vehicles in each direction in real time and only give a green light to the direction of oncoming traffic, while skipping the direction with no vehicles.

[0064] Scenario 2: Intersection with asymmetrical traffic flow.

[0065] For example, at intersections with high southbound traffic and low northbound traffic during the morning rush hour, by skipping the northbound phase when no northbound traffic is detected and allocating the time to the southbound phase, southbound traffic capacity can be greatly improved.

[0066] Scenario 3: Low-traffic intersections at night For example, at intersections with extremely low traffic volume after 10 PM, the system automatically switches to flashing yellow mode when there are no vehicles, and immediately resumes normal control when vehicles approach. This ensures efficient traffic flow at night while preventing drivers from waiting at red lights for extended periods.

[0067] A dynamic phase-hopping system based on real-time vision requirements, used to implement the aforementioned dynamic phase-hopping method based on real-time vision requirements, includes: The visual perception module 10 is deployed at the intersection to collect video streams from the intersection and perform local visual reasoning on the video streams to obtain the real-time vehicle demand number for each preset phase. The phase skipping decision module 20 is used to scan the real-time vehicle demand of each phase in a preset order, give green light time to phases with a demand of greater than zero, and skip phases with a demand of zero directly. The signal execution module 30 is used to send control commands to the signal controller 50 through the local physical interface 31 (such as RS485, CAN or GPIO) according to the decision result of the phase jump decision module 20. The visual perception module 10, the phase decision module 20, and the signal execution module 30 are integrated into the same embodied intelligent device 40. The embodied intelligent device 40 does not need to communicate with an external central platform or cloud server during operation, and all data is processed locally on the device.

[0068] In this embodiment, the visual perception module 10 includes AI camera module 11 is used to collect video streams at intersections; Neural network processing unit 12 is used for local object detection and tracking inference of video stream; Lane mapping unit 13 is used to count the number of targets in each lane corresponding to each phase based on the predefined lane area polygon, and output the real-time vehicle demand number for each preset phase.

[0069] In this embodiment, the phase-jumping decision module 20 includes Demand scanning unit 21 is used to scan the real-time vehicle demand of each phase in a preset order; Green light allocation unit 22 is used to allocate green light time to phases with a demand number greater than zero. The green light time includes minimum green light time, extendable green light time and maximum green light time limit. Skip control unit 23, which is used to directly skip phases with a demand of zero, without allocating green light time; The early termination trigger unit 24 is used to end the green light early when the green light time exceeds the minimum green light time and the demand number drops to zero.

[0070] In this embodiment, it also includes Priority calculation unit 41 is used to calculate the urgency of demand based on the real-time number of vehicles in each phase and the waiting time, and to rearrange the phase scanning order according to the urgency of demand.

[0071] In summary, the key design features of this invention are: 1. Zero or extremely low idle discharge: directly skipping phases with no demand (without giving any green light time), rather than shortening the green light time, fundamentally eliminating idle discharge during the green light of that phase, and reducing the idle discharge rate to below 5%.

[0072] 2. Real-time visual detection: The vehicle demand in each phase is obtained in real time through local visual inference. It has rich detection dimensions (can distinguish vehicle type and detect pedestrians), fast response speed (millisecond level), and does not require infrastructure such as geomagnetic coils.

[0073] 3. Dynamic green light allocation: The green light duration for phases with demand is dynamically adjusted based on the real-time number of vehicles. The green light duration is extended when there are many vehicles and ended earlier when there are few vehicles, thus achieving a refined allocation of time resources.

[0074] 4. Integrated detection and decision-making: Visual perception, phase-hopping decision-making, and signal execution are integrated into the same device, eliminating the need for external communication links and the risk of network latency and communication failure.

[0075] 5. Pedestrian safety is taken into account: Pedestrian waiting needs are automatically identified through visual detection, and pedestrian phases are intelligently inserted. Pedestrian phases cannot be skipped, thus ensuring pedestrians' right of way.

[0076] 6. Intelligent switching at night: When there are no cars at night, it automatically switches to yellow flashing mode and resumes normal control when there are cars, balancing nighttime traffic efficiency and safety.

[0077] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A dynamic phase-jumping method based on real-time vision requirements, characterized in that, Includes the following steps: S1. At the end of each signal cycle, the embodied intelligent device deployed at the intersection obtains the real-time vehicle demand for each preset phase. The real-time vehicle demand is obtained through local visual reasoning of the intersection video stream. S2. Scan the real-time vehicle demand for each phase in a preset order; S3. Give green light time to phases with a demand number greater than zero, and skip phases with a demand number equal to zero. S4. If the real-time vehicle demand for all phases is zero, the control signal will enter the all-red state or the yellow flashing state. The green light time is dynamically adjusted based on the real-time demand for vehicles.

2. The dynamic phase-jumping method based on real-time vision requirements according to claim 1, characterized in that: Step S1 includes: S11. Collect intersection video streams through the AI ​​camera module built into the embodied smart device; S12. The built-in neural network processing unit of the embodied intelligent device performs local target detection and tracking inference on the video stream to obtain the bounding box, category and tracking ID of each traffic target; S13. Based on the predefined lane area polygon, count the number of tracking targets falling into the lane area corresponding to each phase to obtain the real-time vehicle demand for each preset phase. Local target detection and tracking inference are completed within the embodied intelligent device, eliminating the need to upload video data to an external system.

3. The dynamic phase-jumping method based on real-time vision requirements according to claim 2, characterized in that: In step S12, the local object detection uses a lightweight object detection model deployed on the NPU after INT8 quantization. The lightweight object detection model is trained for quantization perception in the intersection scene.

4. The dynamic phase-jumping method based on real-time vision requirements according to claim 1, characterized in that: Step S3, which grants green light time to phases with a demand number greater than zero, includes: Provide the minimum green light time preset for the phase; If the real-time vehicle demand for a phase is still greater than zero after the minimum green light time has ended, the green light time will be extended by a preset unit of time. The number of extensions shall not exceed the preset number; The total green light time for a phase shall not exceed the preset maximum green light time.

5. The dynamic phase-hopping method based on real-time vision requirements according to claim 1, characterized in that: In step S3, skipping phases with a demand of zero includes: When a phase with a demand of zero is detected, no green light time is given to that phase, and the signal status of that phase is directly maintained as red. Continue scanning the next phase until a phase with a demand greater than zero is found or the scanning of all phases is completed.

6. The dynamic phase-jumping method based on real-time vision requirements according to claim 1, characterized in that: It also includes a dynamic phase priority adjustment step; Calculate the demand urgency of each phase based on the real-time vehicle demand and the historical average demand for each phase; Reorder the phase scan sequence from highest to lowest urgency. Among them, the urgency of demand is positively correlated with the real-time number of vehicles in demand and with the waiting time since the last release.

7. The dynamic phase-jumping method based on real-time vision requirements according to claim 1, characterized in that: It also includes a pedestrian phase insertion step: Real-time detection of pedestrian waiting numbers in zebra crossing areas using local visual reasoning; When the number of pedestrians waiting exceeds a preset threshold, mark the pedestrian phase demand. Pedestrian phases are inserted into the vehicle phase sequence, and the green light duration for pedestrian phases is dynamically adjusted based on the number of waiting pedestrians.

8. The dynamic phase-hopping method based on real-time vision requirements according to claim 1, characterized in that: It also includes the steps for switching to night mode: Determine whether it is nighttime based on the exposure parameters of the AI ​​camera module or the average brightness of the image; During the nighttime period, if the real-time vehicle demand for all phases is zero, the control signal will enter a flashing yellow state. When the real-time vehicle demand number for any phase is detected to be greater than zero, the yellow flashing state is exited and normal adaptive phase-skipping control is restored.

9. The dynamic phase-jumping method based on real-time vision requirements according to claim 1, characterized in that: It also includes fail-safe procedures: When the visual reasoning function of the embodied smart device malfunctions or the video stream is interrupted, it automatically switches to the preset fixed timing mode. The timing scheme for the fixed timing mode is stored in the local storage unit of the smart device.

10. A dynamic phase-jumping system based on real-time vision requirements, characterized in that: It is used to implement the dynamic phase-jumping method based on real-time vision requirements as described in any one of claims 1 to 9, and includes: The visual perception module, deployed at the intersection, is used to collect video streams from the intersection and perform local visual reasoning on the video streams to obtain the real-time vehicle demand number for each preset phase. The phase skipping decision module is used to scan the real-time vehicle demand of each phase in a preset order, give green light time to phases with a demand of greater than zero, and skip phases with a demand of zero directly. The signal execution module is used to issue control commands to the signal controller based on the decision result of the phase jump decision module; The visual perception module, phase-jumping decision module, and signal execution module are integrated into the same embodied intelligent device, which does not need to communicate with an external central platform during operation.

Images