Complex scene-oriented automatic driving active separation induction method and system

Abstract

The invention discloses an automatic driving active departure induction method and system for a complex scene, and the method comprises the steps: arranging a roadside active departure induction device in a complex scene region of a target road, and collecting the operation state information of a vehicle through a vehicle-mounted interaction device; and the cloud evaluation platform performs risk evaluation based on the information, determines a comprehensive risk level, and generates an active induction separation strategy according to the comprehensive risk level. The roadside device executes multi-mode separation prompting, and the vehicle-mounted device synchronously outputs in-vehicle prompting information and monitors the separation operation state in real time. And the cloud evaluation platform judges whether to terminate induction or adjust a strategy according to vehicle separation execution feedback, so that safe and orderly separation of the automatic driving vehicle in a complex environment is realized. According to the invention, through vehicle-road cloud cooperation, multi-modal induction and dynamic risk assessment, the safety and reliability of the automatic driving system in a complex scene are improved, and a feasible active separation solution is provided for the automatic driving vehicle.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving safety and traffic guidance, and particularly relates to a method and system for automatic driving active disengagement in complex scenarios. BACKGROUND

[0002] With the rapid development of automatic driving technology, especially the gradual implementation of L3 and higher level automatic driving systems, the autonomous driving capability of vehicles in structured roads has made significant progress. However, in unstructured or dynamically changing complex scenarios, the automatic driving system still faces many challenges, especially in the case of insufficient environmental perception accuracy and limited decision logic adaptability. These problems may cause the system to trigger disengagement mechanism, forcing the driver to take over in time. Although the current automatic driving technology can identify and handle these problems to some extent, in complex environments, disengagement operation still has many hidden dangers, mainly in the form of delayed disengagement, chaotic takeover operation and other safety risks.

[0003] In the prior art, Chinese patent CN112885116B discloses a highway rain and fog scene vehicle-road cooperative guidance system, which includes: an intelligent roadside system including a roadside information acquisition subsystem, a roadside communication subsystem, and a traffic information publishing subsystem; an intelligent vehicle system including a vehicle information acquisition subsystem, a vehicle communication subsystem, and a vehicle warning and control subsystem; a vehicle-road cooperative system for communication data transmission between the intelligent roadside system and the intelligent vehicle system.

[0004] However, this method has the following limitations: 1. The current technology generally lacks a clear definition and grading mechanism for "complex scenarios". Complex scenarios usually include sharp turns, tunnels, traffic accident areas, extreme weather conditions, etc. These scenarios have high uncertainty and dynamic variability. Lack of accurate definition and grading of these complex scenarios leads to delayed disengagement guidance trigger timing, and fails to respond to sudden situations in time. The system often starts to take action only after identifying the risk, failing to provide early warning and prevention. 2. The existing vehicle system usually relies on its own sensors for environmental perception, but lacks effective cooperation with roadside devices and cloud platforms. The perception range of the vehicle system is limited in many complex scenarios such as tunnels and sharp turns, which makes the system unable to respond in time in these scenarios. Due to the lack of vehicle-road cooperation, the vehicle may miss the best disengagement timing, causing delayed or failed disengagement operation. 3. The existing automatic driving guidance scheme usually relies on voice prompts, lacks multi-modal guidance methods and dynamic strategy adjustment. In complex scenarios, a single voice prompt cannot meet the driver's demand for rapid information acquisition in emergency situations. Moreover, the current guidance strategy often lacks real-time adjustment mechanism, and cannot flexibly optimize the disengagement scheme according to the scenario changes and risk assessment results.

[0005] Therefore, the automatic driving field urgently needs to build a new solution that can conduct comprehensive and dynamic risk assessment for complex scenarios. SUMMARY

[0006] The purpose of the present application is to overcome the defects of the prior art and provide an automatic driving active separation induction method and system for complex scenarios.

[0007] The purpose of the present application can be achieved by the following technical solutions: The present application provides an automatic driving active separation induction method for complex scenarios, comprising the following steps: Deploying a roadside active separation induction device in a complex scenario area of a target road according to the scene type; The cloud evaluation platform receives external environment information collected by the roadside active separation induction device and vehicle operation state information collected by the vehicle-mounted interactive device in real time; Based on the external environment information and the operation state information, the cloud evaluation platform conducts risk assessment and determines the comprehensive risk level; The cloud evaluation platform generates a corresponding active separation induction strategy according to the comprehensive risk level and sends it to the roadside active separation induction device and the vehicle-mounted interactive device respectively; The roadside active separation induction device executes a multi-modal separation induction prompt for the autonomous vehicle, and the vehicle-mounted interactive device synchronously outputs the in-vehicle separation guidance information; The vehicle-mounted interactive device monitors the separation execution state of the autonomous vehicle and feeds back to the cloud evaluation platform, and the cloud evaluation platform controls whether to terminate the induction or adjust the induction strategy according to the feedback result.

[0008] Further, the roadside active separation induction device is deployed in a complex scenario area of a target road according to the scene type, specifically including: For road morphology type complex scenarios, at least two roadside active separation induction devices are arranged at different distances in front of the scene starting point; the road morphology type complex scenarios include sharp bends, continuous slopes, long downhills, and tunnel exits; For traffic event type complex scenarios, multiple roadside active separation induction devices are arranged at multiple distance positions in front of the event influence range; the traffic event type complex scenarios include construction areas, traffic accident areas, and traffic congestion areas; For environmental meteorological type complex scenarios, roadside active separation induction devices are arranged at a preset interval along the road section; the environmental meteorological type complex scenarios include heavy fog, heavy rain, and icy road surfaces; For special operation type complex scenarios, multiple roadside active separation induction devices are arranged in different directions within a preset range outside the target operation area; the special operation type complex scenarios include emergency rescue areas and temporary traffic control areas.

[0009] Further, the roadside active decoupling induction device comprises a main support structure, a core perception module, a multi-modal induction module, and a data transmission and power supply unit. The main support structure is used to carry the functional units of the device and provide outdoor deployment capability, including an adjustable vertical pole and a base adapted to different road conditions. The core perception module comprises a positioning device for collecting position and road conditions, a radar device for collecting traffic flow and obstacle information, a camera device for collecting image information, an environmental sensing device for detecting weather and road conditions, and a vibration sensor and infrared thermal imaging device for identifying accidents and work activities. The multi-modal induction module comprises a display unit for displaying graphical and textual prompts, an audio-visual prompt unit for outputting sound and light signals, and a projection unit for presenting graphical guidance on the road surface. The data transmission and power supply unit is used to realize communication between the roadside active decoupling induction device and the cloud evaluation platform and the vehicle-mounted interactive device, and to provide the energy required for continuous operation of the device.

[0010] Further, the vehicle-mounted interactive device comprises a vehicle-road cooperation communication module, a risk warning and interaction unit, a decoupling state feedback module, and an emergency takeover assistance module. The vehicle-road cooperation communication module is used to realize data interaction between the autonomous vehicle and the roadside active decoupling induction device and the cloud evaluation platform, and to receive external environmental information and active decoupling induction strategies. The risk warning and interaction unit is used to output visual, audio or tactile prompts to the driver, display scene risk level, decoupling operation timing and driving path information, and provide a human-machine interaction interface to assist the driver in understanding the induction strategy. The decoupling state feedback module is used to collect the decoupling execution state of the vehicle and feedback to the cloud evaluation platform and the roadside active decoupling induction device, indicating whether the vehicle has completed the decoupling operation according to the induction strategy. The emergency takeover assistance module is used to trigger an emergency takeover prompt when the driver does not respond to the decoupling warning within the specified time limit, control the vehicle to slow down and signal warning, and send identification information to the roadside active decoupling induction device that emergency assistance is needed.

[0011] Further, the external environmental information includes traffic flow information, weather information, road condition information, and accident and work information collected by the core perception module of the roadside active decoupling induction device. The traffic flow information is collected by the radar device, including vehicle speed, vehicle distance and vehicle type. The meteorological information is collected through environmental sensing devices, including visibility and road surface friction conditions; The road surface condition information is collected by camera devices and environmental sensors, including road smoothness and obstacle location; The accident and operation information is acquired in real time through vibration sensors and infrared thermal imaging devices. When the vehicle vibration reaches a preset threshold or the vehicle dwell time exceeds a preset threshold, emergency data collection is triggered to increase the collection frequency.

[0012] Furthermore, the autonomous vehicle operating status information includes autonomous driving system status information, vehicle driving data, and MPR data collected in real time through the in-vehicle interactive device; The autonomous driving system status information includes the working status of each sensor and the decision response delay; The vehicle driving data includes vehicle speed, lane position, and steering angle; The MPR data is used to reflect the minimum identifiable capability of an autonomous driving system for environmental perception and decision-making response, including system perception latency, decision calculation latency, and execution response latency.

[0013] Furthermore, the risk assessment and determination of the comprehensive risk level by the cloud-based assessment platform based on the external environment information and operational status information specifically includes: Step A1: Road segment division and simulation modeling: Divide the entire road into several analysis segments, including construction sections, ramp sections, straight sections or curved sections; build a traffic simulation model for each road segment on the simulation platform, and input information such as road geometry, traffic flow, vehicle type ratio, speed limit, and warning zone length; consider the mixed traffic environment of manually driven vehicles and autonomous vehicles in the simulation. Step A2: Setting Influencing Factors: In the simulation of each road segment, consider the following key factors: Disengagement threshold of autonomous driving system: that is, the time threshold at which the system triggers a takeover request when encountering a complex scenario, and is set to three levels: short, medium and long. Driver takeover style: divided into aggressive, normal, and conservative; aggressive drivers maintain high attention during takeover with minimal performance degradation; conservative drivers experience significant performance degradation during takeover; normal drivers fall somewhere in between. Traffic flow levels are categorized into low, medium, and high flow. Large vehicle ratio: This refers to the proportion of large vehicles such as trucks and buses in traffic flow, and is set at three levels: low, medium, and high. Warning zone length: The length of the warning zone in front of the construction area is set to three levels: short, medium, and long. Speed ​​limits in construction zones: set at three levels: low speed, medium speed, and near-normal speed limit; Autonomous vehicle penetration rate: This refers to the proportion of autonomous vehicles in traffic flow, ranging from fully manual driving to fully autonomous driving.

[0014] Step A3: Conflict Detection and Equivalent Traffic Conflict Count. Vehicle trajectories are extracted in the simulation. Using a full-type traffic conflict detection method, single-vehicle conflicts, two-vehicle conflicts, and multi-vehicle conflicts are identified. The conflict count is converted into an equivalent traffic conflict count. Taking into account the probability and severity of accidents, different types of conflicts are unified into a quantitative index. The single-vehicle conflict includes vehicles deviating from the road or colliding with fixed objects. Step A4: Statistical analysis of the entire road: Calculate the equivalent number of traffic conflicts for each road segment; summarize the results of all road segments and calculate the overall mean, variance, and standard deviation; Step A5: Risk Index and Risk Level, defining the risk index. for: According to the risk index Risk level classification: Low risk: Risk index Less than or equal to 1.5 times the standard deviation; Medium risk: Risk index Between 1.5 and 2.5 standard deviations; High risk: Risk index Greater than 2.5 times the standard deviation; Step A6: Real-time application: Input real-time road information into the simulation model or regression equation; quickly calculate the predicted equivalent number of traffic conflicts for the road segment; dynamically update the risk level based on the risk index to provide decision support for traffic management departments; the real-time road information includes traffic flow, proportion of large vehicles, speed limit, length of construction zone, and proportion of autonomous vehicles.

[0015] Furthermore, the generation of a corresponding proactive disengagement strategy based on the comprehensive risk level specifically includes: When the cloud-based assessment platform determines the overall risk level to be high, it generates an emergency evacuation strategy, including: The setting of the departure trigger timing, the planning of the target lane departure, the safe speed range of the departure process, and the application of multimodal guidance methods; the application of multimodal guidance methods includes roadside projection, audio and visual prompts, in-vehicle voice and steering wheel vibration; For medium-risk situations, a warning exit strategy is generated, including advance reminders for drivers to slow down, lane keeping optimization, and multimodal guidance prompts; For low-risk situations, a monitoring strategy is generated, which only provides information prompts and real-time status monitoring, without any active intervention.

[0016] Furthermore, the in-vehicle interactive device monitors the disengagement status of the autonomous vehicle and feeds it back to the cloud evaluation platform. The cloud evaluation platform controls whether to terminate the guidance or adjust the guidance strategy based on the feedback results, specifically including: The in-vehicle interactive device collects real-time data on the disengagement operation of autonomous vehicles, including whether the vehicle has completed lane changes, whether the vehicle speed during the disengagement process is within a safe range, and whether there are potential conflict risks. The disengagement operation data is then synchronized to the cloud evaluation platform and the roadside active disengagement guidance device. When the cloud-based assessment platform receives feedback that the vehicle has successfully entered the target lane and that its speed is stable with no risk of conflict, it issues a stop guidance command. The roadside active detachment guidance device turns off the audio-visual prompts and road surface projection, and the in-vehicle interactive device stops issuing warnings and displays: Detachment complete, current scenario is safe. When the cloud-based assessment platform receives feedback signals that a vehicle has not changed lanes according to the planned route or that its speed exceeds the safe range, it reassesses the overall risk level and adjusts the guidance strategy, including extending the departure time and changing the target lane, and sends the updated guidance strategy to the roadside active departure guidance device and the vehicle interaction device. When the cloud-based assessment platform receives an emergency assistance signal from the in-vehicle interactive device, it coordinates with surrounding roadside devices to expand the warning range and pushes emergency event information to surrounding autonomous vehicles, prompting them to take evasive action.

[0017] Another aspect of the present invention provides a system for any of the above-described methods for active detachment guidance in complex scenarios for autonomous driving, comprising: The roadside active disengagement guidance device is used to deploy and execute multimodal disengagement guidance in complex scene areas of the target road. It includes a main support structure, a core perception module, a multimodal guidance module, and a data transmission and power supply unit, which are used to collect external environmental information and provide disengagement guidance to autonomous vehicles. The in-vehicle interactive device is installed inside the autonomous vehicle to receive guidance strategies from the roadside active detachment guidance device and the cloud evaluation platform in real time, provide detachment prompts to the driver or autonomous driving system, and collect vehicle operating status information and detachment execution status feedback to the cloud evaluation platform. The cloud-based assessment platform receives external environmental information collected by the roadside active detachment guidance device and vehicle operating status information collected by the in-vehicle interactive device. It assesses the risks of autonomous vehicles and the external environment, generates a comprehensive risk level, and generates an active detachment guidance strategy based on the risk level. The strategy is then distributed to the roadside device and the in-vehicle interactive device. Simultaneously, it receives vehicle detachment feedback and adjusts the guidance strategy based on the feedback results.

[0018] Compared with the prior art, the present invention has the following advantages: (1) Existing technologies lack a clear definition and classification mechanism for complex scenarios, resulting in the inability to trigger detachment guidance in a timely manner in complex environments, increasing the accident risk of vehicles in high-risk scenarios. This invention deploys roadside active detachment guidance devices in complex scenario areas of the target road according to different scenario types (such as road morphology, traffic incidents, environmental weather, special operations, etc.), and dynamically evaluates and classifies these complex scenarios to ensure timely and accurate triggering of detachment guidance strategies under different risk levels. Through this method, the system's adaptability in complex environments is effectively improved, ensuring that autonomous vehicles can perform detachment operations in a timely and accurate manner under dangerous conditions.

[0019] (2) Existing autonomous vehicles mainly rely on single onboard sensors for environmental perception, lacking coordination with external facilities. This is especially true in blind spot scenarios such as tunnels or sharp bends, where they are prone to missing the optimal time to disengage. This invention constructs a complete vehicle-road-cloud collaborative network through the collaborative work of onboard interactive devices, roadside active disengagement guidance devices, and cloud-based evaluation platforms. The vehicle receives external environmental information collected by roadside devices and the cloud platform in real time, and performs joint evaluation in conjunction with onboard system status data, ensuring that no potential risk points are overlooked in complex environments, thereby enhancing the coordination and responsiveness of the autonomous driving system.

[0020] (3) Existing autonomous driving guidance methods are mostly based on voice prompts and lack multimodal guidance methods and dynamic adjustment mechanisms, which cannot meet the demand for rapid and intuitive prompts in complex scenarios. This invention achieves multi-channel output of information through multimodal guidance methods (such as roadside projection, sound and light prompts, in-vehicle voice and steering wheel vibration), enabling drivers to obtain timely guidance information through multiple sensory means in complex scenarios. This method can effectively improve the driver's understanding and reaction speed to the disengagement guidance strategy, thereby reducing the driver's takeover error rate in high-risk environments.

[0021] (4) In existing technologies, risk assessment of autonomous driving systems typically relies solely on local sensor information, lacking a comprehensive and accurate integrated risk assessment mechanism. To address this deficiency, this invention proposes a comprehensive risk assessment method based on traffic flow information, road surface condition information, and multi-source data from both vehicle and roadside sources. This method extracts vehicle trajectories through a simulation platform, utilizes full-type traffic conflict detection technology to identify single-vehicle conflicts, two-vehicle conflicts, and multi-vehicle conflicts, and uniformly converts different types of conflicts into equivalent traffic conflict numbers (UETCN). Subsequently, it calculates the risk index RI based on statistical data from the entire road, with the formula RI = (UETCN of a certain road segment − average value) / standard deviation, and classifies the risk into three levels: low, medium, and high, according to the magnitude of RI. In this way, the system can achieve comprehensive and dynamic risk assessment of autonomous vehicles in complex scenarios and generate corresponding active departure guidance strategies accordingly, thereby achieving a more accurate and comprehensive risk assessment. Through this technology, the system can generate corresponding guidance strategies based on the comprehensive risk level, ensuring that vehicle departure guidance in complex scenarios is more scientific and efficient.

[0022] (5) In the prior art, the status monitoring and feedback mechanism for disengagement operations is not perfect and often relies on manual intervention, leading to operational delays and disengagement failures. This invention monitors the disengagement execution status through an in-vehicle interactive device and transmits real-time feedback information to a cloud-based evaluation platform. The cloud platform adjusts its strategy based on the feedback results to ensure continuous optimization of the disengagement operation. For example, if the vehicle successfully disengages or fails, the system automatically adjusts the guidance strategy, such as extending the disengagement time or changing the target lane, and sends the new strategy to the roadside and in-vehicle devices. This mechanism makes the disengagement process more flexible and precise, greatly improving the system's stability and emergency response capabilities in complex environments. Attached Figure Description

[0023] Figure 1 This is a flowchart of the autonomous driving active disengagement guidance method according to an embodiment of the present invention; Figure 2 This is a model diagram of an autonomous driving active departure guidance system according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] Example 1: This embodiment provides a method for active disengagement guidance in complex autonomous driving scenarios, such as...Figure 1 As shown, it includes the following steps: Step S1: Deploy roadside active detachment guidance devices in complex scene areas of the target road according to the scene type, specifically including: Complex scenarios: These refer to scenarios where the perception accuracy of an autonomous driving system is below 90% and the decision response latency exceeds 100ms due to environmental dynamism, terrain constraints, weather interference, or traffic uncertainties. Specifically, they include the following four categories: Complex road morphology scenarios: such as sharp bends (curve radius ≤ 200m), continuous slopes (slope ≥ 5% and length ≥ 500m), tunnel entrances and exits (sudden change in illumination ≥ 100,000 lux), and unprotected intersections (no traffic light control and convergence of traffic from multiple directions). Complex traffic incident scenarios: such as road construction areas (occupying ≥1 lane and having temporary barriers), traffic accident scenes (vehicles stuck for ≥30 seconds or occupying ≥1 lane), and traffic congestion (average vehicle speed ≤20km / h and queue length ≥500m). Complex meteorological scenarios: such as heavy rain (precipitation ≥ 50 mm / h), dense fog (visibility ≤ 200 m), icy and snowy roads (road friction coefficient ≤ 0.3), and strong crosswinds (wind speed ≥ 10.8 m / s); Special operational scenarios: such as emergency rescue sites (where there are rescue vehicles and dense crowds and no fixed traffic guidance), temporary traffic control sections (temporary lane switching or closure), and traffic control areas around large-scale events (where traffic density is ≥30pcu / km and pedestrians are crossing in a disorderly manner).

[0026] For complex road morphology scenarios, one roadside device is deployed 300m ahead of the starting point of the scenario, and a second device is deployed 150m ahead, forming a dual early warning mechanism; complex road morphology scenarios include sharp bends, continuous slopes, long downhill slopes, and tunnel exits. For complex scenarios involving traffic incidents, the first unit is deployed 500m ahead of the incident's impact area, the second unit is deployed 300m ahead, and the third unit is deployed 100m ahead, covering all stages of early warning, including long-range, medium-range, and short-range warnings. Complex scenarios involving traffic incidents include construction areas, traffic accident areas, and traffic congestion areas. For complex environmental and meteorological scenarios, one roadside device is deployed every 1000m along the road to ensure continuous coverage; complex environmental and meteorological scenarios include heavy fog, rainstorms, and icy and snowy roads. For complex scenarios involving special operations, one unit is deployed in each of the four directions (east, south, west, and north) within a 500m radius around the rescue area to form a ring-shaped guidance system. These complex scenarios include emergency rescue areas and temporary traffic control areas.

[0027] And perform system initialization: Trigger collaborative initialization of the cloud platform, roadside devices and vehicle-mounted equipment through the mobile APP to complete the following operations: ① GPS positioning calibration of roadside devices and synchronization of scene location to the cloud platform; ② Communication connection test between vehicle-mounted equipment, roadside devices and cloud platform (send test data packets to ensure packet loss rate ≤1%); ③ Initialization of cloud platform algorithm parameters (such as setting TTC threshold and MRM probability benchmark value according to road section speed limit).

[0028] Step S2: The cloud-based evaluation platform receives in real time external environmental information collected by the roadside active detachment guidance device and autonomous vehicle operating status information collected by the in-vehicle interaction device. Roadside active detachment guidance device: External environmental information includes traffic flow information, meteorological information, road surface condition information, and accident and operation information collected by the core sensing module of the roadside active detachment guidance device; among them, traffic flow information is collected by radar devices, including vehicle speed, distance, and vehicle type; meteorological information is collected by environmental sensors, including visibility and road surface friction conditions; road surface condition information is collected by camera devices and environmental sensors, including road smoothness and obstacle location; accident and operation information is acquired in real time through vibration sensors and infrared thermal imaging devices. When vehicle vibration reaches a preset threshold or vehicle dwell time exceeds a preset threshold, emergency data collection is triggered to increase the collection frequency.

[0029] In-vehicle interactive device: The operating status information of autonomous vehicles includes the autonomous driving system status information, vehicle driving data and MPR data collected in real time through the in-vehicle interactive device; among them, the autonomous driving system status information includes the working status of each sensor and the decision response delay; the vehicle driving data includes vehicle speed, lane position and steering angle; the MPR data is used to reflect the minimum identifiable capability of the autonomous driving system to perceive the environment and make decisions, including system perception delay, decision calculation delay and execution response delay.

[0030] Step S3: Based on external environment information and operational status information, the cloud-based assessment platform conducts a risk assessment and determines the overall risk level, specifically including: Step A1: Road segment division and simulation modeling: Divide the entire road into several analysis segments, including construction sections, ramp sections, straight sections or curved sections; build a traffic simulation model for each road segment on the simulation platform, and input information such as road geometry, traffic flow, vehicle type ratio, speed limit, and warning zone length; consider the mixed traffic environment of manually driven vehicles and autonomous vehicles in the simulation. Step A2: Setting Influencing Factors: In the simulation of each road segment, consider the following key factors: Disengagement threshold of autonomous driving system: that is, the time threshold at which the system triggers a takeover request when encountering a complex scenario, and is set to three levels: short, medium and long. Driver takeover style: divided into aggressive, normal, and conservative; aggressive drivers maintain high attention during takeover with minimal performance degradation; conservative drivers experience significant performance degradation during takeover; normal drivers fall somewhere in between. Traffic flow levels are categorized into low, medium, and high flow. Large vehicle ratio: This refers to the proportion of large vehicles such as trucks and buses in traffic flow, and is set at three levels: low, medium, and high. Warning zone length: The length of the warning zone in front of the construction area is set to three levels: short, medium, and long. Speed ​​limits in construction zones: set at three levels: low speed, medium speed, and near-normal speed limit; Autonomous vehicle penetration rate: This refers to the proportion of autonomous vehicles in traffic flow, ranging from fully manual driving to fully autonomous driving.

[0031] Step A3: Conflict Detection and Equivalent Traffic Conflict Count. Vehicle trajectories are extracted in the simulation. Using a full-type traffic conflict detection method, single-vehicle conflicts, two-vehicle conflicts, and multi-vehicle conflicts are identified. The conflict count is converted into an equivalent traffic conflict count. Taking into account the probability and severity of accidents, different types of conflicts are unified into a quantitative index. The single-vehicle conflict includes vehicles deviating from the road or colliding with fixed objects. Step A4: Statistical analysis of the entire road: Calculate the equivalent number of traffic conflicts for each road segment; summarize the results of all road segments and calculate the overall mean, variance, and standard deviation; Step A5: Risk Index and Risk Level, defining the risk index. for: in, This represents the equivalent number of traffic conflicts on a certain road segment; These represent the mean and standard deviation of the equivalent number of traffic conflicts for the entire road, respectively.

[0032] According to the risk index Risk level classification: Low risk: Risk index Less than or equal to 1.5 times the standard deviation; Medium risk: Risk index Between 1.5 and 2.5 standard deviations; High risk: Risk index Greater than 2.5 times the standard deviation; Step A6: Real-time application: Input real-time road information into the simulation model or regression equation; quickly calculate the predicted equivalent number of traffic conflicts for the road segment; dynamically update the risk level based on the risk index to provide decision support for traffic management departments; the real-time road information includes traffic flow, proportion of large vehicles, speed limit, length of construction zone, and proportion of autonomous vehicles.

[0033] Step S4: The cloud-based assessment platform generates corresponding active detachment guidance strategies based on the comprehensive risk level and distributes them to the roadside active detachment guidance device and the vehicle-mounted interactive device, respectively. If the overall risk level is "high risk", the cloud platform generates an inducement strategy, including: the timing of the exit trigger (e.g., "start exiting when 200m away from the starting point of the scene"), the target exit lane (e.g., "change from the first lane on the left to the second lane on the right"), the safe speed range (e.g., "maintain speed of 40-60km / h during the exit"), and the multimodal guidance method (e.g., "roadside projection + in-vehicle voice + steering wheel vibration"). If the risk level is "medium risk", warning information will only be displayed through roadside devices and reminders will be pushed by vehicle-mounted equipment, without triggering active guidance; if the risk level is "low risk", the system will maintain routine monitoring and will not intervene.

[0034] Step S5: The roadside active detachment guidance device performs multimodal detachment guidance prompts for the autonomous vehicle, and the in-vehicle interactive device simultaneously outputs in-vehicle detachment guidance information; The roadside active detachment guidance device activated: Long-distance warning (500-300m ahead): The LED matrix screen displays "High-risk area ahead, autonomous driving please prepare to leave", the flashing light switches to yellow constant flashing, and the loudspeaker broadcasts "Entering a complex scene 500m ahead, autonomous driving vehicle please plan your exit route in advance"; Mid-range guidance (300-100m ahead of the scene): The LED matrix screen displays a dynamic arrow (pointing to the target exit lane) and a countdown (such as "enter the exit lane in 30 seconds"), the road projection unit projects "target lane + speed limit sign", and the strobe lights switch to red-yellow alternating flashing; Close-range confirmation (within 100m ahead of the scene): The LED matrix screen displays "Immediate departure, target lane: right lane 2", the loudspeaker plays "Attention, immediately change lanes to the right to complete the departure", and the flashing lights switch to red-blue alternating flashing to enhance the warning effect; In-vehicle interactive device actions: Warning phase (30 seconds before departure): The vehicle screen displays the scenario risk assessment result (e.g., "Overall risk score 75 points, high risk"), the target route planning map, and a voice prompt: "High-risk scenario of entering construction zone ahead. You need to leave to the second lane on the right in 30 seconds. Please be prepared." Disengagement execution phase (10 seconds before disengagement to disengagement completion): The vehicle screen displays the real-time disengagement progress (e.g., "50m has been traveled, 150m remaining to reach the target lane"), the steering wheel vibrates at a frequency of 2Hz, and the voice prompts every 5 seconds, "Please maintain a speed of 50km / h and change lanes to the right." Emergency Phase (Driver not responding for ≥10 seconds): Trigger emergency mode, automatically reduce vehicle speed to 40km / h, turn on hazard lights, display "Emergency: No takeover operation detected, please change lanes to the right immediately", and send an "emergency signal" to the roadside device. The roadside device's strobe lights switch to blue constant flashing, and the loudspeaker plays "Vehicles ahead require emergency assistance, please give way to surrounding vehicles".

[0035] Step S6: The in-vehicle interactive device monitors the disengagement status of the autonomous vehicle and feeds it back to the cloud evaluation platform. The cloud evaluation platform controls whether to terminate the guidance or adjust the guidance strategy based on the feedback results.

[0036] Status feedback: The in-vehicle interactive device collects disengagement operation data in real time (such as whether the lane change is completed and whether the vehicle speed is within a safe range), and synchronizes it to the cloud platform and roadside devices; if the vehicle successfully enters the target lane, the vehicle speed is stable and there is no risk of conflict, the in-vehicle device sends a "disengagement successful" signal. Strategy Adjustment: If the cloud platform receives the "disengagement successful" signal, it will issue a "stop guidance" command. The roadside device will turn off the flashing lights and loudspeakers and resume normal monitoring. The vehicle-mounted equipment will stop issuing warnings and display "disengagement completed, current scene is safe". If a "failed to disengage" signal is received (e.g., the vehicle does not change lanes according to the planned route or the speed exceeds the safe range), the cloud platform reassesses the environmental risks, adjusts the guidance strategy (e.g., extends the disengagement time to 40 seconds, changes the target lane), and re-sends it to the roadside and vehicle-mounted devices. If an "emergency assistance" signal is received, the cloud platform will coordinate with nearby roadside devices to expand the warning range (e.g., activate the warning on roadside devices within 1000m behind the notification scene), and at the same time push "emergency event" information to nearby autonomous vehicles to remind them to avoid the area.

[0037] Compared with the prior art, the present invention has the following significant advantages: Systematized definition and assessment of complex scenarios: For the first time, the classification and quantification standards for complex scenarios are clearly defined, and algorithms such as the full range of traffic conflict detection methods and equivalent conflict numbers are integrated to achieve accurate assessment of environmental complexity and risk. The accuracy rate of high-risk scenario identification is ≥95%, which is 60% higher than the existing single-scenario identification scheme. Strong collaborative capability of proactive departure guidance: Constructs a three-level collaborative system of "cloud-roadside-vehicle", realizes real-time data interaction through standardized communication protocols, and the departure guidance decision generation time is ≤100ms, which is 80% better than the non-collaborative solution (decision delay ≥500ms), ensuring that autonomous vehicles receive early warning and guidance in advance; Multimodal guidance has wide adaptability: Combining LED screen display, audio and visual prompts, road projection, in-vehicle interaction and other methods, in complex environments such as heavy rain (visibility ≤200m) and ice and snow (road friction coefficient ≤0.3), the driver's recognition rate of guidance information is ≥90%, which is 50% higher than that of voice prompts alone; Balancing safety and flexibility: Through the emergency takeover assistance module and dynamic strategy adjustment, the failure rate of disengagement when the driver does not respond is reduced to ≤5%. The roadside device supports deployment in multiple scenarios, and the vehicle-mounted device is compatible with mainstream autonomous driving models, which can meet diverse needs such as emergency rescue and temporary construction. The deployment efficiency is 4 times higher than that of traditional warning devices.

[0038] Example 2: This embodiment provides an active disengagement guidance system for autonomous driving in complex scenarios, such as... Figure 2 As shown, it includes a cloud-based assessment platform, a roadside active detachment guidance device, and an in-vehicle interaction device; the three interact with each other through a standardized communication protocol (supporting LTE-V2X and 5G); Cloud-based assessment platform: Core functions: Responsible for data access, risk assessment, strategy generation, and historical data management, specifically including the following modules: Data access module: Receives environmental data (road conditions, weather, traffic flow) uploaded by roadside devices and vehicle-mounted equipment, supports parallel access of ≥1000 data entries / second, and has a transmission latency of ≤50ms.

[0039] Multi-algorithm fusion evaluation module: Integrates environmental complexity assessment, hazard prediction, and evacuation necessity determination, specifically including: First, vehicle trajectories are extracted using a simulation platform. A comprehensive traffic conflict detection method is employed to identify single-vehicle, two-vehicle, and multi-vehicle conflicts. These are then assessed using indicators such as Traffic Conflict Detection (TTC) and Traffic Response Detection and Accident Detection (DRAC). Different types of conflicts are uniformly converted into equivalent traffic conflict numbers (UETCNs) to reflect the probability and severity of accidents. Subsequently, statistical analysis is performed on the UETCNs of each road segment, calculating the average and standard deviation for the entire road. Based on this, a risk index (RI) is defined, calculated as: RI = (UETCN of a road segment − average) / standard deviation. Risk levels are categorized according to the RI: RI less than or equal to 1.5 is considered low risk, RI between 1.5 and 2.5 is considered medium risk, and RI greater than 2.5 is considered high risk. Only when a high-risk scenario is identified does the system trigger an active disengagement guidance strategy to ensure that the autonomous vehicle can complete the disengagement operation promptly and safely in complex scenarios.

[0040] Decision generation and distribution module: Generates guidance strategies (such as exit timing, target lane, safe speed) based on the comprehensive risk level and distributes them to roadside devices and vehicle-mounted equipment. Decision generation time ≤100ms; Historical data management module: Stores scenario evaluation results, induction strategies and evacuation effect data, supports querying by scenario type and time range, data storage period is ≥1 year, and can be used for algorithm iteration and optimization.

[0041] Roadside active detachment guidance device: The roadside active detachment guidance device includes a main support structure and a core sensing and guidance module, adaptable to deployment requirements in various complex scenarios: Main support structure: Made of high-strength carbon fiber, the height of the upright can be adjusted from 1.8m to 3.5m, the base has a built-in switchable counterweight (suitable for fixing on asphalt, cement and icy roads), the functional compartment has an IP protection rating of ≥IP67, and supports operation in a wide temperature range of -40℃ to 85℃. Core perception module: Multi-source environmental sensors: integrate differential GPS (positioning accuracy ≤0.5m), millimeter-wave radar (detection distance 5m-100m, speed measurement accuracy ±0.5km / h), high-definition camera (resolution ≥4 million pixels, supports backlight compensation and night vision mode), and meteorological sensor (detects precipitation, visibility, and road friction coefficient), which can collect scene location, traffic flow, meteorological and road condition data in real time; Accident and work identification sensors: including vibration sensors (detection sensitivity ≥0.05g, capable of identifying collision vibrations ≤3km / h) and infrared thermal imagers (resolution ≥1280×1024, capable of identifying vehicles or construction workers stranded at night), triggering an alarm for an accident or work scene within 10ms; Multimodal induction module: Dynamic warning sign unit: An LED matrix screen (size ≥1.2m×0.6m, brightness ≥8000cd / ㎡) is installed on the top, which can display text information such as "Complex scene ahead, please leave in advance for autonomous driving" and "Target lane: the second lane on the right", as well as dynamic arrows (compliant with GB5768.2-2022), with a minimum visibility distance of ≥20m under midday sunlight; Audio-visual guidance unit: A directional tweeter is installed on the side (volume 60dB-130dB, supports custom voice, such as "Attention autonomous vehicles, a high-risk area is 500m ahead, please move to the right lane to leave"), and a strobe light is installed at the bottom (supports switching between red, yellow and blue, red light wavelength 620nm-660nm, yellow light 580nm-595nm, blue light 450nm-475nm); Road surface projection unit: adopts laser projection technology (resolution ≥1080P, brightness ≥8000 lumens, projection distance 5m-30m), can project dynamic graphics such as "deceleration lane", "left / right departure" and "speed limit 40km / h", and the visibility distance on icy and snowy roads is ≥15m; Data transmission and power supply unit: includes a 5G / BeiDou dual-mode communication module (BeiDou short message latency ≤8s in the absence of public network), a solar panel (power ≥50W, conversion efficiency ≥24%) and a lithium iron phosphate battery (capacity ≥200Ah, cycle life ≥3000 times, supports 72 hours of operation without sunlight).

[0042] In-vehicle interactive devices: The in-vehicle interactive device is installed inside the autonomous vehicle to enable interaction with roadside devices and cloud platforms, and to provide driver prompts, including: Vehicle-road cooperative communication module: Supports LTE-V2X and 5G communication, can receive scenario risk data (such as construction area location and accident type) pushed by roadside devices and guidance strategies issued by the cloud platform, with a communication latency of ≤30ms; Risk warning and interaction unit: It adopts a 7-inch touch screen (resolution ≥1920×1080), which can display the scene risk level, the exit countdown (such as "exit operation must be started in 10 seconds"), and the target path planning map. At the same time, it can provide prompts to the driver through voice (supporting multiple languages, such as Chinese and English) and steering wheel vibration (vibration frequency adjustable, 1-5Hz). Disengagement status feedback module: collects vehicle disengagement operation data in real time (such as lane change status and speed adjustment) and feeds it back to the cloud platform and roadside device. When the vehicle completes disengagement (such as entering a safe lane and stabilizing speed), it automatically sends a "disengagement successful" signal. Emergency Takeover Assist Module: If the driver fails to respond to the exit warning in time, the emergency mode can be triggered, automatically reducing the vehicle speed (maximum deceleration ≤ 3m / s²). 2Turn on the hazard lights and display a brief lane departure guide on the vehicle's infotainment screen (such as "Emergency: Please change lanes to the right immediately"), while simultaneously sending an "Emergency assistance required" signal to the roadside device.

[0043] Example 3: Due to road surface repairs at kilometer marker K200 of a certain expressway, two lanes on the left (out of a total of four lanes) are closed. This falls under the category of a "complex traffic incident scenario," and the implementation steps are as follows: Deployment and initialization: Deploy one roadside active disengagement guidance device at 500m, 300m and 100m in front of the construction area. Complete GPS positioning (synchronize the location to the cloud platform) and communication test through the mobile APP (ensure normal connection with surrounding autonomous vehicles and cloud platform backend). Environmental Assessment and Decision-Making: Roadside devices collect traffic flow data (current traffic density 25 pcu / km, average vehicle speed 80 km / h) and meteorological data (visibility 1500m, road surface dry) every 5 seconds. No anomalies were detected by the accident sensors. The cloud-based assessment platform first establishes a simulation model based on parameters such as road geometry, traffic flow, large vehicle ratio, speed limits, and warning zone length, and then runs it in a mixed traffic environment involving both manually driven and autonomous vehicles. During the simulation, vehicle trajectories are extracted, and single-vehicle, two-vehicle, and multi-vehicle conflicts are identified and uniformly converted into the equivalent traffic conflict number (UETCN). Subsequently, the risk index (RI) for the construction section is calculated based on statistical data from the entire road. When the RI for the road section is greater than 2.5, it is classified as high-risk. Based on this, the platform generates an emergency evacuation strategy including: starting evacuation 200m ahead of the construction zone, with the target lane being the second lane from the right, and a safe speed range of 40–60 km / h during the evacuation process, using multimodal guidance (roadside projection, audio-visual prompts, in-vehicle voice prompts, and steering wheel vibration). If the scenario is identified as "high-risk," an induction strategy is generated: "start evacuation 200m from the construction zone, with the target lane being the second lane from the right, maintaining a speed of 40-60 km / h during the evacuation process, using LED screens + projections + audio-visual guidance on the roadside, and using voice prompts + steering wheel vibration prompts in the vehicle." Multimodal guidance: Roadside device at 500m: LED screen displays "Construction ahead, autonomous driving please prepare to leave", yellow flashing lights flash continuously, voice broadcast "Construction ahead 500m, need to leave to the second lane on the right"; Roadside device at 300m: LED screen displays dynamic right-hand arrow and "Leaving will begin in 20s", road surface projection shows "Right arrow + 50km / h speed limit", red-yellow flashing lights flash alternately; In-vehicle device: Vehicle screen displays the location of the construction area and the target path, voice prompt "Leaving to the second lane on the right is required in 20s, current speed is 80km / h, please reduce to 50km / h", steering wheel vibrates at 2Hz; Disengagement Feedback and Adjustment: After receiving the instruction, the driver of an autonomous vehicle decelerated to 50 km / h as instructed and changed lanes from the first lane on the left to the second lane on the right within 15 seconds. The onboard device sent a "disengagement successful" signal. After receiving the signal, the cloud platform issued a "stop guidance" instruction, and the roadside device turned off the flashing lights and voice prompts and resumed normal monitoring. During the construction period, the disengagement success rate of autonomous vehicles on this section of road was 100%, and there were no scratches caused by untimely disengagement.

[0044] Example 4: A section of a highway at kilometer marker 350 experienced heavy rain (60 mm / h rainfall, 150 m visibility), which falls under the category of "complex meteorological environment scenario." The implementation steps are as follows: Deployment and initialization: Deploy one roadside device every 1000m along the K350-K355 section (5km in total) to complete positioning and communication tests (BeiDou short message module activated to cope with possible public network signal interruption). Environmental Assessment and Decision-Making: In a rainstorm scenario, the simulation platform inputs environmental parameters such as reduced visibility and decreased road surface friction coefficient, and models the system based on traffic flow and the proportion of autonomous vehicles. The equivalent traffic conflict number (UETCN) for the road segment is obtained through trajectory conflict detection, and the risk index (RI) is calculated by combining the average and standard deviation of the entire road segment. When the RI for the road segment is greater than 2.5, it is considered high-risk, and an guidance strategy is generated: "Start disengagement 1000m from the low-visibility section, target the middle lane (avoiding the leftmost flooded area), maintain a speed of 30-40km / h, enhance roadside projection brightness (activate laser illumination), and increase the steering wheel vibration frequency to 3Hz." Multimodal guidance: Roadside device: LED screen displays "Heavy rain, low visibility, please move to the middle lane for autonomous driving", laser projection brightness is increased to 10,000 lumens (ensure the road arrow is visible), red-blue strobe lights flash alternately; In-vehicle device: Vehicle screen displays "High risk of heavy rain, move to the middle lane after 1000m, it is recommended to turn on the windshield wipers at the highest setting", voice reminder every 3 seconds, steering wheel vibrates at 3Hz. Disengagement Feedback and Adjustment: Due to heavy rain, the driver of an autonomous vehicle failed to respond in time. Ten seconds later, the onboard device triggered the emergency mode, automatically decelerating to 35 km / h, activating the hazard lights, and displaying "Emergency: Please change lanes to the middle lane immediately" on the vehicle screen. At the same time, an "emergency signal" was sent to the roadside devices. After receiving the signal, the roadside devices within 1000m switched their strobe lights to a constant blue flashing light and announced "The vehicle ahead needs emergency assistance. Please slow down and give way to surrounding vehicles." Finally, the driver successfully disengaged after receiving the emergency prompts, and the onboard device sent a "disengagement successful" signal, and the system resumed normal monitoring.

[0045] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0046] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An active decoupling method for autonomous driving in a complex scenario, characterized in that, The method comprises the following steps: deploying a roadside active decoupling induction device in a complex scene area of a target road according to a scene type; a cloud evaluation platform receives external environment information collected by the roadside active decoupling induction device and running state information of an autonomous vehicle collected by a vehicle-mounted interactive device in real time; risk assessment is performed by the cloud evaluation platform based on the external environment information and the running state information, and a comprehensive risk level is determined; the cloud evaluation platform generates a corresponding active decoupling induction strategy according to the comprehensive risk level and respectively issues the active decoupling induction strategy to the roadside active decoupling induction device and the vehicle-mounted interactive device; a multi-modal decoupling induction prompt for the autonomous vehicle is executed by the roadside active decoupling induction device, and in-vehicle decoupling guidance information is synchronously output by the vehicle-mounted interactive device; the vehicle-mounted interactive device monitors a decoupling execution state of the autonomous vehicle and feeds back to the cloud evaluation platform, and the cloud evaluation platform controls whether to terminate the induction or adjust the induction strategy according to the feedback result. 2.The method of claim 1, wherein, The roadside active decoupling induction device deployed in the complex scene area of the target road specifically comprises: for a road form type complex scene, at least two roadside active decoupling induction devices are arranged at different distances in front of the scene starting point; the road form type complex scene includes sharp bends, continuous slopes, long downhills, and tunnel exits; for a traffic event type complex scene, multiple roadside active decoupling induction devices are arranged at multiple distance positions in front of an event influence range; the traffic event type complex scene includes construction areas, traffic accident areas, and traffic congestion areas; for an environmental meteorological type complex scene, roadside active decoupling induction devices are arranged at a preset interval along a road section; the environmental meteorological type complex scene includes heavy fog, heavy rain, and icy road surfaces; for a special operation type complex scene, multiple roadside active decoupling induction devices are arranged in different directions within a preset range outside a target operation area; the special operation type complex scene includes emergency rescue areas and temporary traffic control areas. 3.The method of claim 1, wherein, The roadside active decoupling induction device comprises a main support structure, a core perception module, a multi-modal induction module, and a data transmission and power supply unit; The main support structure is used to carry the functional units of the device and provide outdoor deployment capability, including an adjustable vertical pole and a base suitable for different road conditions. The core perception module comprises a positioning device for collecting position and road conditions, a radar device for collecting traffic flow and obstacle information, a camera device for collecting image information, an environmental sensing device for detecting weather and road surface environment, and a vibration sensor and an infrared thermal imaging device for identifying accidents and operation activities. The multi-modal induction module comprises a display unit for displaying graphical and textual prompts, an audio-visual prompt unit for outputting sound and light signals, and a projection unit for presenting graphical guidance on the road surface. The data transmission and power supply unit is used to realize communication between the roadside active decoupling induction device and the cloud evaluation platform and the vehicle-mounted interactive device, and to provide the energy required for continuous operation of the device.

4. The method of claim 1, wherein, The vehicle-mounted interactive device comprises a vehicle-road cooperation communication module, a risk warning and interaction unit, a decoupling state feedback module, and an emergency takeover assistance module. The vehicle-road cooperation communication module is configured to realize data interaction between the autonomous vehicle and the roadside active decoupling guiding device and the cloud evaluation platform, receive external environment information and active decoupling guiding strategies; The risk warning and interaction unit is configured to output visual, voice or tactile prompts to the driver, display scene risk level, decoupling operation timing and driving path information, and provide a human-computer interaction interface to assist the driver in understanding the guiding strategy; The decoupling state feedback module is configured to collect the decoupling execution state of the vehicle and feed back to the cloud evaluation platform and the roadside active decoupling guiding device, to indicate whether the vehicle completes the decoupling operation according to the guiding strategy; The emergency takeover assistance module is configured to trigger an emergency takeover prompt when the driver does not respond to the decoupling warning within a specified time limit, control the vehicle to slow down and signal for warning, and send identification information of the need for emergency assistance to the roadside active decoupling guiding device.

5. The method of claim 1, wherein, The external environment information includes traffic flow information, weather information, road surface state information and accident and operation information collected by the core perception module of the roadside active decoupling guiding device; The traffic flow information is collected by a radar device and includes vehicle speed, vehicle distance and vehicle type; The weather information is collected by an environmental sensing device and includes visibility and road surface friction condition; The road surface state information is collected by a camera device and an environmental sensing device and includes road flatness and obstacle position; The accident and operation information is obtained by real-time monitoring of a vibration sensor and an infrared thermal imaging device, and when the vehicle vibration reaches a preset threshold or the vehicle stays for more than a preset threshold, emergency data collection is triggered to increase the collection frequency.

6. The method of claim 1, wherein, The autonomous vehicle running state information includes autonomous driving system state information, vehicle driving data and MPR data collected by a vehicle-mounted interaction device in real time; The autonomous driving system state information includes sensor working state and decision response delay; The vehicle driving data includes vehicle speed, lane position and steering angle; The MPR data is used to reflect the minimum identifiable ability of the autonomous driving system to environmental perception and decision response, including system perception delay, decision calculation delay and execution response delay.

7. The method of claim 1, wherein, The cloud evaluation platform performs risk assessment and determines the comprehensive risk level based on the external environment information and the running state information, specifically including: Step A1: road segment division and simulation modeling: divide the entire road into several analysis road segments, including construction sections, ramp sections, straight sections or curve sections; build a traffic simulation model for each road segment on a simulation platform, input road geometry, traffic flow, vehicle proportion, speed limit, warning zone length, etc.; consider the mixed traffic environment of manual driving vehicles and autonomous vehicles in the simulation; Step A2: influence factor setting: in the simulation of each road segment, the following key factors are considered: Decoupling threshold of autonomous driving system: the time threshold for the system to trigger takeover request when encountering complex scenes, set to short, medium and long levels; Driver takeover style: divided into aggressive, normal and conservative; aggressive drivers maintain high attention during takeover, with less performance degradation; conservative drivers have significant performance degradation during takeover; normal is between the two; Traffic flow level: divided into low, medium and high flow; Large vehicle ratio: the proportion of large vehicles such as trucks and buses in the traffic flow, set to low, medium and high levels; Warning zone length: the length of the warning zone before the construction zone, set to short, medium and long levels; Construction zone speed limit: set to low, medium and close to normal speed limit; Autonomous vehicle penetration rate: the proportion of autonomous vehicles in the traffic flow, from different levels between complete manual driving and complete autonomous driving; Step A3: Conflict detection and equivalent traffic conflict number In the simulation, the vehicle trajectory is extracted, and the single vehicle conflict, double vehicle conflict and multi-vehicle conflict are identified by using the full type traffic conflict detection method; The conflict number is converted into equivalent traffic conflict number, the accident probability and severity are considered comprehensively, and different types of conflicts are unified into a quantitative index; The single vehicle conflict includes vehicle deviation from the road or collision with fixed objects; Step A4: Statistical analysis of the whole road: calculate the equivalent traffic conflict number for each section; The results of all sections of the whole road are summarized, and the average, variance and standard deviation of the whole are calculated; Step A5: Risk Index and Risk Rating, defining the risk index is: According to the risk index Risk level is divided: Low risk: Risk Index Less than or equal to 1.5 standard deviations; Medium risk: Risk index Between 1.5 and 2.5 times the standard deviation; High risk: Risk index Greater than 2.5 standard deviations; Step A6: Real-time application: input the real-time information of the road into the simulation model or regression equation; Quickly calculate the predicted equivalent traffic conflict number of the section; According to the risk index, dynamically update the risk level, and provide decision support for traffic management departments; The real-time information of the road includes traffic flow, large vehicle ratio, speed limit, construction zone length, and autonomous vehicle ratio. 8.The method of claim 1, wherein, The corresponding active disengagement induction strategy is generated according to the comprehensive risk level, specifically including: When the cloud evaluation platform determines that the comprehensive risk level is high risk, generate an emergency disengagement strategy, including: The setting of disengagement trigger time, the planning of target disengagement lane, the safety speed range of disengagement process, and the application of multi-modal guidance mode; The application of multi-modal guidance mode includes roadside projection, sound and light prompt, vehicle-mounted voice and steering wheel vibration; For medium risk state, generate pre-warning disengagement strategy, including early warning driver to slow down, lane keeping optimization and multi-modal induction prompt; For low risk state, generate monitoring strategy, only information prompt and real-time state monitoring, no active intervention. 9.The method of claim 1, wherein, The disengagement execution state of the autonomous vehicle is monitored by the vehicle-mounted interactive device and fed back to the cloud evaluation platform, and the cloud evaluation platform controls whether to terminate the induction or adjust the induction strategy according to the feedback result, specifically including: The vehicle-mounted interactive device collects disengagement operation data of the autonomous vehicle in real time, including whether the vehicle completes lane change, whether the vehicle speed in the disengagement process is within the safety range, and whether there is potential conflict risk, and synchronizes the disengagement operation data to the cloud evaluation platform and the roadside active disengagement induction device; When the cloud evaluation platform receives a feedback signal that the vehicle has successfully entered the target lane and the speed is stable and there is no conflict risk, it issues a stop induction command. The road-side active disengagement device stops the sound and light prompts and road projection, and the vehicle-mounted interactive device stops the warning and displays: disengagement complete, current scene is safe; When the cloud evaluation platform receives a feedback signal that the vehicle has not changed lanes according to the planned route or the speed has exceeded the safe range, it re-evaluates the comprehensive risk level and adjusts the induction strategy, including extending the disengagement time, changing the target lane, etc., and issues the updated induction strategy to the road-side active disengagement device and the vehicle-mounted interactive device; When the cloud evaluation platform receives an emergency assistance signal sent by the vehicle-mounted interactive device, it links the surrounding road-side devices to expand the warning range and pushes emergency event information to surrounding autonomous vehicles to prompt surrounding vehicles to take evasive measures.

10. A system for the active induced method of autonomous driving away from complex scenes, according to any one of claims 1-9, characterized in that, It includes: A road-side active disengagement device for deploying and implementing multi-modal disengagement induction in complex scene areas of target roads, including a main support structure, a core perception module, a multi-modal induction module, and a data transmission and power supply unit, for collecting external environment information and providing disengagement guidance to autonomous vehicles; A vehicle-mounted interactive device installed inside an autonomous vehicle for real-time reception of induction strategies issued by road-side active disengagement devices and cloud evaluation platforms, providing disengagement prompts to drivers or autonomous driving systems, and collecting vehicle operating status information and disengagement execution status feedback to the cloud evaluation platform; A cloud evaluation platform for receiving external environment information collected by road-side active disengagement devices and vehicle operating status information collected by vehicle-mounted interactive devices, evaluating the risk of autonomous vehicles and external environment, generating a comprehensive risk level, and generating an active disengagement induction strategy according to the risk level, issuing it to road-side devices and vehicle-mounted interactive devices, while receiving vehicle disengagement feedback and adjusting the induction strategy according to the feedback result.

Images