Response processing method for collision accidents, intelligent helmet and response processing platform

Abstract

The present disclosure provides a response processing method of a collision accident, an intelligent helmet and a response processing platform, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring a collision event signal from a target terminal associated with a delivery capacity; the target terminal comprises an intelligent helmet and / or a voice communication terminal; in response to the collision event signal, triggering a delivery management server to initiate an outbound call request to a target terminal with a call function associated with the delivery capacity to establish a voice call link; acquiring voice interaction data in the outbound call voice call link, the voice interaction data being used for analysis to assess and determine the injury level of the delivery capacity; the delivery management server generates a corresponding accident intervention action according to the injury level, and executes the accident intervention matched with the injury level. According to the embodiment of the present disclosure, the accident handling efficiency and the safety guarantee level of the delivery capacity can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a method for handling collision accidents, a smart helmet, and a response handling platform. Background Technology

[0002] With the rapid development of the on-demand delivery industry, delivery personnel (such as food delivery riders) face complex traffic environments during daily deliveries, posing a risk of traffic accidents such as collisions or falls. Once such an accident occurs, the personal safety of delivery personnel is directly threatened; therefore, a solution is urgently needed to address such incidents promptly. Summary of the Invention

[0003] This disclosure provides a method for responding to collision accidents, a smart helmet, and a response processing platform, which can improve the efficiency of accident handling and the safety level of delivery capacity.

[0004] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0005] According to one aspect of this disclosure, a method for responding to a collision accident is provided, comprising: acquiring a collision event signal from a target terminal associated with a delivery capacity, the target terminal including a smart helmet and / or a voice communication terminal; in response to the collision event signal, triggering a delivery management server to initiate an outbound call request to the target terminal with call functionality associated with the delivery capacity to establish a voice call link; acquiring voice interaction data in the outbound voice call link, the voice interaction data being analyzed to assess and determine the injury level of the delivery capacity; and the delivery management server generating a corresponding accident intervention action based on the injury level and executing an accident intervention matching the injury level.

[0006] In one embodiment of this disclosure, in response to a collision event signal, a delivery management server is triggered to initiate an outbound call request to a target terminal with call functionality associated with the delivery capacity to establish a voice call link and to acquire voice interaction data in the outbound voice call link. This includes: in response to the collision event signal, sending an outbound call request to a target terminal bound to the delivery capacity identifier based on the delivery capacity identifier; detecting that the target terminal has accepted the outbound call request and determining that a voice call link has been established; sending collision accident-related inquiry voice to the delivery capacity through the voice call link, the collision accident-related inquiry voice being used to guide the delivery capacity to provide feedback on its current status information; acquiring audio data collected by the target terminal, the audio data being used to assess the injury level of the delivery capacity; wherein, the audio data includes the delivery capacity's response voice data to the inquiry voice, and / or, environmental sound data at the accident scene.

[0007] In one embodiment of this disclosure, acquiring voice interaction data in a voice call link further includes: generating the next round of inquiry voice based on the voice interaction data of progressive inquiry dialogue and collision accident-related inquiry voice, until the injury level of the terminal user can be assessed based on the currently collected voice interaction data, and then stopping the generation of the next round of inquiry voice.

[0008] In one embodiment of this disclosure, analyzing voice interaction data to assess and determine the injury level of delivery capacity includes: determining semantic feature data of the voice interaction data through content-to-semantic analysis, and assessing the injury level of delivery capacity based on the semantic feature data of the voice interaction data; and / or determining acoustic speech feature data of the voice interaction data through analysis of the coherence characteristics and timbre characteristics of the speaker's voice dialogue in the voice interaction data, and assessing the injury level of delivery capacity based on the acoustic speech feature data of the voice interaction data.

[0009] In one embodiment of this disclosure, analyzing voice interaction data to assess and determine the injury level of delivery personnel includes: determining, based on semantic feature data of the voice interaction data, that the voice of delivery personnel in the voice interaction data contains one or more of the following: emergency help keywords, severe pain keywords, or painful groans, then determining the injury level as a serious injury; and / or, based on acoustic voice feature data of the voice interaction data, determining that the voice of multiple timbres in the voice interaction data contains injury-related keywords, then determining the injury level as a serious injury.

[0010] In one embodiment of this disclosure, assessing the injury level of delivery personnel based on semantic feature data of delivery personnel's speech includes: determining, based on semantic feature data of speech interaction data, that the speech of delivery personnel in the speech interaction data contains one or more of the following: keywords for autonomous injury processing, keywords for mild pain, or soft groans, then determining the injury level as minor injury; and / or, based on acoustic speech feature data of speech interaction data, determining that the coherence characteristics of the speech of delivery personnel in the speech interaction data meet the condition of unclear language expression, then determining the injury level as minor injury.

[0011] In one embodiment of this disclosure, assessing the injury level of delivery personnel based on semantic feature data of delivery personnel's voice includes: determining, based on semantic feature data of voice interaction data, that the voice of delivery personnel in the voice interaction data contains keywords indicating no injury, and thus determining the injury level as no injury; and / or, based on acoustic voice feature data of voice interaction data, determining that the coherence characteristics of the voice corresponding to the timbre of delivery personnel in the voice interaction data meet the conditions for normal dialogue, and thus determining the injury level as no injury.

[0012] In one embodiment of this disclosure, determining the injury level of the delivery personnel based on voice feature data further includes: analyzing the ambient sound data in the voice interaction data to determine the impact intensity at the time of the collision; and combining the impact intensity, the delivery personnel's response voice data, and / or the ambient sound data at the accident scene to determine the injury level.

[0013] According to another aspect of this disclosure, a smart helmet is provided, including a sensing module, a signal transmitting module, and an audio acquisition module.

[0014] The perception module is used to detect collision incidents; The signal sending module is used to send a collision event signal to the delivery management server after the sensing module detects a collision accident. This causes the delivery management server to respond to the collision event signal by initiating an outbound call request to a target terminal with communication function associated with the delivery capacity wearing the smart helmet, so as to establish a voice call link. The audio acquisition module is used to collect voice interaction data in the voice call link and send the voice interaction data to the delivery management server. The voice interaction data is used to analyze and assess the injury level of the delivery capacity and to perform accident interventions that match the injury level.

[0015] According to another aspect of this disclosure, a collision accident response processing platform is provided, including a signal receiving module, an outbound call triggering module, a data analysis module, and a response execution module.

[0016] The signal receiving module is used to acquire collision event signals from target terminals associated with the delivery capacity, including smart helmets and / or voice communication terminals; The outbound call triggering module is used to respond to collision event signals and trigger the management server to initiate an outbound call request to the target terminal with call function associated with the delivery capacity in order to establish a voice call link; The data analysis module is used to acquire voice interaction data in the outbound voice call link. The voice interaction data is used for analysis to assess and determine the severity of the injury to the delivery capacity. The response execution module is used to trigger the management server to generate corresponding accident intervention actions based on the injury level, and to execute accident interventions that match the injury level.

[0017] The technical solutions provided in this disclosure can include the following beneficial effects: This embodiment of the disclosure achieves automated and timely accident response by automatically sensing collision events at the target terminal and triggering the server to automatically initiate outbound calls, effectively overcoming the problem of lag in traditional manual response. Simultaneously, by establishing a voice call link to obtain voice interaction data for injury assessment, it can directly obtain real-time feedback on the delivery capacity at complex accident scenes, improving the accuracy and reliability of injury assessment. Finally, based on the determined injury level, it generates and executes matching accident intervention actions, achieving differentiated and precise intervention strategies. This ensures timely assistance to seriously injured delivery personnel while avoiding waste of rescue resources, comprehensively improving accident handling efficiency and the safety assurance level of delivery capacity.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0020] Obviously, the accompanying drawings described below are merely some embodiments of this disclosure. Those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0021] Figure 1 A flowchart of a collision accident response processing method according to an embodiment of this disclosure is shown; Figure 2 This diagram illustrates a response and processing scenario for a collision accident according to an embodiment of the present disclosure. Figure 3 This diagram illustrates a flowchart of the process for acquiring voice interaction data in an embodiment of this disclosure. Figure 4 This diagram illustrates the architecture of a collision incident response and processing system according to an embodiment of the present disclosure. Figure 5 This diagram illustrates a smart helmet according to an embodiment of the present disclosure; Figure 6 This diagram illustrates a collision incident response processing platform according to an embodiment of the present disclosure. Detailed Implementation

[0022] To facilitate understanding of the technical solutions of this disclosure, the disclosure will be further described below with reference to the accompanying drawings.

[0023] The terms "first" and "second," etc., used in this disclosure are only used to distinguish different objects and not to describe a specific order. The terms "comprising" and "having," and any variations thereof, in the embodiments of this disclosure are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may also include steps or units not listed.

[0024] Throughout this disclosure, the references to "embodiments" do not necessarily refer to the same embodiments, nor are they independent or alternative embodiments mutually exclusive with other embodiments. Those skilled in the art will understand, explicitly and implicitly, that the embodiments described in this disclosure can be combined with other embodiments.

[0025] In the embodiments of this disclosure, "at least one" refers to one or more, "more" refers to two or more, "at least two" refers to two or three or more, and "and / or" is used to describe the relationship between associated objects. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0026] The following detailed description of this exemplary implementation method is provided in conjunction with the accompanying drawings and embodiments.

[0027] Figure 1 This document illustrates a flowchart of a collision incident response handling method according to an embodiment of this disclosure. This method can be executed by a delivery management server. It should be noted that the data involved in this disclosure, including but not limited to the data itself, its acquisition, and its use, shall comply with relevant laws, regulations, and related provisions. Before using the technical solutions disclosed in the embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure through appropriate means, and user authorization should be obtained, in accordance with relevant laws and regulations.

[0028] like Figure 1 As shown, the collision accident response and processing method provided in this embodiment includes S101 to S104.

[0029] In S101, a collision event signal is acquired from a target terminal associated with the delivery capacity. The target terminal includes a smart helmet and / or a voice communication terminal.

[0030] Combination Figure 2The diagram illustrates the terminal device connection architecture. The smart helmet and the voice communication terminal (e.g., a smartphone) are pre-bound and associated with the same delivery capacity (i.e., the same rider). The smart helmet and the voice communication terminal can establish a communication connection via short-range communication methods such as Bluetooth or Wi-Fi. Simultaneously, the smart helmet and the voice communication terminal can also independently connect to the delivery management service terminal via wireless communication methods such as cellular networks. In some embodiments, the smart helmet can also indirectly connect to the delivery management service terminal via the voice communication terminal it communicates with, acting as a gateway.

[0031] Based on the above architecture, the delivery management server can obtain collision event signals in several ways: it can obtain collision event signals directly from the smart helmet; it can obtain collision event signals sent by the smart helmet indirectly through the voice communication terminal; or it can obtain collision event signals sent by the voice communication terminal itself directly.

[0032] In some embodiments, the smart helmet has built-in sensors (such as accelerometers, gyroscopes, etc.). Based on the data it senses, the smart helmet determines whether a collision has occurred at the edge. When a collision is determined to have occurred, the smart helmet directly sends a collision event signal to the delivery management server, or indirectly sends the signal to the delivery management server via a voice communication terminal. In this case, the collision event signal can be a separate control signal used to trigger subsequent processes, or it can be a data packet containing the original sensed data used to determine that a collision has occurred.

[0033] In other embodiments, the smart helmet itself does not perform collision detection, but directly sends the collected perception data to the delivery management server through the above-mentioned direct or indirect methods, and uses the perception data as a collision event signal; then, after receiving the signal, the delivery management server needs to perform certain analysis and calculation on the perception data it contains, and after confirming that a collision has indeed occurred, it executes the subsequent S102.

[0034] In S102, in response to the collision event signal, the delivery management server is triggered to initiate an outbound call request to the target terminal with call function associated with the delivery capacity in order to establish a voice call link.

[0035] After confirming a collision event, the delivery management server automatically triggers an outbound call engine (such as an AI voice robot) to initiate a voice call to a terminal with call functionality bound to the delivery capacity. Once the terminal connects, a two-way voice call link is established, achieving full automation and high timeliness in accident response. It can establish a direct voice communication channel with the delivery capacity in a very short time after an accident without relying on human customer service intervention, effectively overcoming the problem of delayed traditional manual response and buying valuable time for subsequent injury assessment and golden rescue.

[0036] In S103, voice interaction data from the outbound voice call link is acquired. This voice interaction data is used for analysis to assess and determine the severity level of delivery capacity.

[0037] After establishing a voice call link, inquiries can be sent to delivery personnel via this link, and voice interaction data such as responses from delivery personnel and ambient sounds from the accident scene can be collected. This voice interaction data includes not only semantic information at the text level, but also acoustic features such as timbre, intonation, and coherence, as well as ambient sound characteristics (the specific acquisition and analysis process of the voice interaction data will be described in detail later with reference to specific embodiments). This disclosure provides a rich data foundation for subsequent injury assessment based on multimodal analysis by acquiring multi-dimensional voice interaction data.

[0038] In S104, the delivery management server generates corresponding accident intervention actions based on the injury level and executes the accident intervention that matches the injury level.

[0039] After determining the injury level (e.g., serious injury, minor injury, no injury) of delivery personnel, the delivery management server automatically generates and executes corresponding accident intervention actions based on a preset tiered routing strategy. These actions include contacting the emergency center, notifying the station manager, or providing medical guidance (the specific logic for determining the injury level will be detailed later in conjunction with specific embodiments). This disclosure achieves precise scheduling and efficient utilization of rescue resources by executing differentiated accident interventions that match the injury level. While ensuring timely and effective assistance for seriously injured personnel, it avoids wasting rescue resources and comprehensively improves accident handling efficiency and the safety level of delivery personnel.

[0040] In some embodiments, in response to a collision event signal, the delivery management server is triggered to initiate an outbound call request to a target terminal with call functionality associated with the delivery capacity, in order to establish a voice call link and obtain voice interaction data in the outbound voice call link, which may include... Figure 3 S301 to S304 are shown.

[0041] In S301, in response to a collision event signal, an outbound call request is sent to the target terminal bound to the delivery capacity identifier based on the delivery capacity identifier.

[0042] Upon receiving a collision event signal, the delivery management server extracts the delivery capacity identifier (such as rider ID, registered mobile phone number, etc.) carried in the signal and queries the associated database for the target terminal with calling capabilities (such as smartphone, smart helmet, etc.) bound to that identifier. If the delivery capacity is bound to multiple target terminals, the delivery management server can send outbound call requests according to a preset priority strategy (e.g., prioritizing calls to smartphones, then calling smart helmets if the call is not answered, or simultaneously initiating calls to multiple terminals). Outbound call requests can be implemented by calling operator interfaces or third-party voice service platform APIs.

[0043] In S302, the target terminal is detected to have made an outbound call request, and a voice call link is established.

[0044] The delivery management server can detect whether the target terminal has been successfully connected by listening to the signaling interaction status or receiving status callback codes. If no connection is detected within a specified time (e.g., the terminal is muted, the call is rejected, no one answers, or there is a network abnormality), the delivery management server can trigger a retry mechanism to make another call, or transfer the call to a human operator according to a preset degradation strategy.

[0045] In S303, collision-related inquiry voice messages are sent to delivery personnel via a voice call link. These messages are used to guide delivery personnel to provide feedback on their current status.

[0046] After the link is established, the delivery management server dynamically synthesizes or plays pre-set audio to send inquiry voice messages to the delivery personnel. The content of these inquiry voice messages may include questions to confirm the delivery personnel's level of consciousness (such as "Can you hear me?") or questions to confirm their injury status (such as "Where do you feel uncomfortable right now?" or "Are you injured?"). This proactive and targeted voice guidance prompts the delivery personnel to respond verbally or make sounds. This disclosure, by proactively sending guiding inquiry voice messages, effectively stimulates voice feedback from the delivery personnel, avoiding the silence and lack of data caused by one-way calls, providing the necessary prerequisites for obtaining effective audio data subsequently, and improving the effectiveness of human-computer interaction.

[0047] In S304, audio data collected by the target terminal is acquired. The audio data is used to assess the injury level of the delivery personnel. The audio data includes the delivery personnel's response voice data to the inquiry voice feedback, and / or, environmental sound data of the accident scene.

[0048] Simultaneously or subsequently, the delivery management server controls the target terminal's microphone to continuously collect audio data and upload it in real time. The response voice data includes not only the textual semantic content of the delivery personnel's reply but also their acoustic characteristics (such as speech rate, volume, and whether there are gasps, groans, tremors, or pained tones). The environmental sound data from the accident scene includes background noise, surrounding voices, traffic collision sounds, or ambulance sirens. During collection and transmission, noise reduction and enhancement processing can be performed on the target terminal or in the cloud to extract clearer audio features.

[0049] This disclosure provides a rich and comprehensive data dimension for subsequent injury assessment by simultaneously acquiring response speech data containing semantic and acoustic features, as well as environmental sound data reflecting the objective situation on site. This enables injury assessment to go beyond simple text semantic analysis and comprehensively judge the rider's degree of physical pain and the degree of danger on site, significantly improving the accuracy, comprehensiveness and robustness of the assessment results.

[0050] In some embodiments, acquiring voice interaction data in a voice call link further includes generating the next round of inquiry voice based on progressive inquiry dialogue and voice interaction data based on collision accident-related inquiry voice, until the level of injury of the terminal user can be assessed based on the currently collected voice interaction data, at which point the generation of the next round of inquiry voice stops.

[0051] Progressive questioning can be based on a pre-set injury assessment logic tree or dialogue model, asking questions to delivery personnel in a progressive and hierarchical manner. For example, the first round of questioning is usually a simple confirmation of consciousness and status (such as "Can you hear me?" or "Are you safe now?"). After receiving an affirmative answer, the next round of questioning smoothly transitions to specific injury investigation (such as "Where do you feel pain now?" or "Can you move your limbs normally?").

[0052] When generating the next round of questions, the delivery management server (or cloud-based intelligent dialogue engine) performs semantic understanding and intent recognition on the voice interaction data collected in the previous round in real time, and dynamically adjusts the dialogue strategy accordingly. For example, if the delivery personnel responded "I feel dizzy and can't move my legs" in the previous round, key features such as "dizzy" and "can't move my legs" will be extracted to dynamically generate targeted follow-up questions (such as "Have you been hit on the head? Are you conscious now?"), instead of mechanically playing a preset fixed script.

[0053] The delivery management server can internally set a confidence threshold for injury assessment or a completion indicator for key information collection. When the currently collected voice interaction data (such as information on consciousness level, pain location, and limb mobility) meets the input requirements of the injury assessment model, i.e., the assessment confidence reaches the preset threshold, or all key question nodes under the current injury branch have been traversed, it is determined that "the injury level of the end user can be assessed." At this point, the generation of new inquiry voices stops, the dialogue loop ends, and the currently accumulated voice interaction data is directly input into the injury assessment model, jumping to the subsequent accident intervention stage.

[0054] By introducing a mechanism of progressive questioning and dynamically generating the next round of voice, the system can, on the one hand, conduct precise and adaptive follow-up questions based on real-time feedback from delivery capacity. This effectively overcomes the problem of incomplete information gathering or irrelevant answers in complex accident scenarios using traditional fixed scripts, significantly improving the relevance and completeness of obtaining key information about injuries. On the other hand, the progressive questioning approach, from simple to complex, aligns with the cognitive and expressive patterns of people experiencing shock or injury during a sudden accident, effectively lowering the communication threshold. Furthermore, the mechanism of stopping follow-up questioning once assessment conditions are met avoids secondary interference and energy depletion caused by ineffective or excessive questioning of injured delivery personnel, shortening the overall call duration and enabling a faster transition to the substantive accident intervention phase, significantly improving the overall efficiency of the rescue response.

[0055] In some embodiments, analyzing voice interaction data to assess and determine the severity of injury to delivery capacity includes: determining semantic feature data of the voice interaction data through content-to-semantic analysis, and assessing the severity of injury to delivery capacity based on the semantic feature data of the voice interaction data; and / or determining acoustic speech feature data of the voice interaction data through analysis of the coherence characteristics and timbre characteristics of the speaker's voice dialogue in the voice interaction data, and assessing the severity of injury to delivery capacity based on the acoustic speech feature data of the voice interaction data.

[0056] The aforementioned semantic feature data primarily reflects the textual content information of delivery personnel's feedback, such as keywords extracted through speech recognition (e.g., "leg pain," "dizziness," "bleeding") or complete sentence meanings, used to directly assess their subjective feelings and objective injuries. Acoustic speech feature data primarily reflects the physiological and physical state of delivery personnel when they speak, such as slower speech rate, voice trembling, decreased volume, and non-semantic sound features like panting and groaning extracted through speech signal processing, used to indirectly assess their degree of physical weakness or pain. In practical applications, injury assessment can be performed based on either of these feature data alone, or semantic feature data and acoustic speech feature data can be fused together for multimodal analysis to comprehensively determine the final injury level. By introducing a dual analytical dimension of semantic and acoustic speech features, on the one hand, semantic features can accurately obtain the subjective injury feedback of delivery personnel; on the other hand, acoustic features can capture their subconscious physiological reactions (e.g., voice trembling, panting) in a state of injury or pain. The combination of the two approaches, or their complementary nature, effectively overcomes the limitations of single-text analysis when riders are unclear in their expression, have blurred consciousness, or are unable to accurately describe their injuries, significantly improving the comprehensiveness, accuracy, and robustness of injury assessment.

[0057] In some embodiments, analyzing voice interaction data to assess and determine the injury level of delivery personnel includes: determining, based on semantic feature data of the voice interaction data, that the voice of delivery personnel in the voice interaction data contains one or more of the following keywords: first aid help keywords, severe pain keywords, or painful groans, and then determining the injury level as a serious injury level; and / or, based on acoustic voice feature data of the voice interaction data, determining, based on multiple timbre voices in the voice interaction data, that the voice contains injury-related keywords, and then determining the injury level as a serious injury level.

[0058] Keywords for emergency assistance include "call an ambulance," "help," and "call 120 quickly." Keywords for severe pain include "broken leg," "dizzy," "heavy bleeding," and "unable to breathe." Painful groans refer to high-decibel sounds identified through acoustic features, conveying obvious pain but lacking clear semantic meaning. Regarding "multiple timbre voices," in actual accident scenarios, voice interaction data may include not only the delivery person's voice but also the voices of bystanders and bystanders calling for help (i.e., multiple timbres identified through speaker separation technology). These bystander voices may contain injury-related keywords such as "he's bleeding" or "he was hit badly." Additionally, it could refer to the delivery person's own voice exhibiting hoarseness, distorted voice, or other abnormal timbre changes due to injury, combined with contextual injury-related keywords. By capturing extreme injury semantics and abnormal acoustic features (especially introducing sudden changes in bystander and self-timbre timbre as auxiliary judgments), critical situations can be quickly and accurately identified. This multi-tone, multi-dimensional serious injury assessment mechanism effectively avoids missed assessments due to the rider being unconscious, unable to speak, or unable to express himself clearly, providing a highly reliable basis for triggering the highest level of emergency rescue (such as directly contacting the 120 emergency center).

[0059] In some embodiments, assessing the injury level of delivery personnel based on semantic feature data of delivery personnel's speech includes: determining, based on semantic feature data of speech interaction data, that the speech of delivery personnel in the speech interaction data contains one or more of the following: keywords for autonomous injury processing, keywords for mild pain, or soft groans, then determining the injury level as minor injury; and / or, based on acoustic speech feature data of speech interaction data, determining that the coherence characteristics of the speech of delivery personnel in the speech interaction data meet the condition of unclear language expression, then determining the injury level as minor injury.

[0060] Keywords for self-treatment of injuries include "I'll apply some medicine myself," "No need to go to the hospital," and "It's not a big deal." Keywords for mild pain include "It hurts a little," "I scraped my skin," and "I twisted my hand." Soft groans refer to low-volume, relatively calm cries of pain. Regarding "coherence characteristics meeting the conditions of unclear language expression," this refers to delivery personnel exhibiting abnormally slow speech, excessively long pauses, logical incoherence, unclear pronunciation, stuttering, or slow reactions when answering inquiries. This is usually a subconscious physiological reaction resulting from a mild concussion, shock, or pain leading to difficulty concentrating and a temporary decline in cognitive ability. By identifying the semantics of minor injuries and the characteristics of unclear language expression due to pain / shock, it is possible to accurately distinguish between minor injuries and uninjured states. In particular, using "unclear expression" as the acoustic basis for minor injury assessment effectively captures the objective situation where riders may subjectively conceal or downplay their injuries, or believe they are unharmed but are physiologically affected. This compensates for the shortcomings of relying solely on semantic analysis and significantly improves the objectivity and accuracy of minor injury assessment.

[0061] In some embodiments, assessing the injury level of delivery personnel based on semantic feature data of delivery personnel's speech includes: determining the injury level as uninjured if the speech of delivery personnel in the speech interaction data contains keywords indicating no injury based on semantic feature data of speech interaction data; and / or determining the injury level as uninjured if the speech of delivery personnel in the speech interaction data has continuity characteristics that meet the conditions of normal dialogue based on acoustic speech feature data of speech interaction data.

[0062] Keywords indicating "I'm fine," "I'm not injured," and "Don't worry about me" are included. Regarding "coherence characteristics meeting the conditions of normal conversation," this means that during voice interaction, the delivery personnel exhibit stable vocal characteristics, a steady speaking speed, clear enunciation, logical coherence, and fluent responses, without obvious abnormal pauses, vocal tremors, or logical errors. They are able to clearly and accurately understand inquiries and provide reasonable responses. Through positive semantic confirmation and normal acoustic coherence verification, injuries can be efficiently and accurately ruled out, avoiding excessive rescue efforts. This not only saves medical and dispatch resources and reduces platform operating costs but also allows delivery personnel confirmed to be uninjured to quickly resume delivery tasks, minimizing the negative impact of accidents on platform fulfillment rates and user experience.

[0063] In some embodiments, determining the injury level of the delivery personnel based on voice feature data further includes: analyzing the ambient sound data in the voice interaction data to determine the impact intensity at the time of the collision; and combining the impact intensity, the delivery personnel's response voice data, and / or the ambient sound data at the accident scene to determine the injury level.

[0064] Analyzing ambient sound data in voice interaction data to determine the impact intensity at the time of a collision can be achieved by extracting acoustic physical features (such as sound pressure level / volume peak, spectral change characteristics, sound duration, and energy distribution) from the ambient sound at the moment of collision, thereby quantitatively estimating the physical impact force when a vehicle or object collides.

[0065] Determining the severity of injuries by combining impact intensity, delivery personnel's response voice data, and / or ambient sound data from the accident scene can employ a multi-dimensional cross-validation assessment strategy. For example, if the analyzed "impact intensity" is extremely high, but the delivery personnel's "response voice data" indicates minor injury or no injury (in which case the rider may be confused due to shock or mild concussion, or deliberately concealing injuries out of fear), a comprehensive assessment can be made to increase the severity of injuries or trigger a higher-level secondary confirmation mechanism. Furthermore, "ambient sound data from the accident scene" (such as continuous vehicle horns, shouts from surrounding people, loud rescue sounds, or the sounds of wind and rain) can also serve as contextual information to aid in judging the severity of the accident and the level of danger at the scene. By integrating and analyzing objective physical impact intensity, subjective rider voice feedback, and the atmosphere of the scene, a more realistic injury severity assessment can be generated.

[0066] Figure 4 The architecture of the collision incident response and processing system disclosed herein is shown below, with reference to... Figure 4 The system for responding to and handling public collision accidents and the corresponding methods are explained in detail.

[0067] like Figure 4 As shown, the system architecture includes an edge-side perception layer, a core processing layer, a decision analysis layer, and a business response layer.

[0068] The edge-side perception layer includes an environmental perception unit and a smart helmet sensing unit. The environmental perception unit is used to acquire ambient sound, location information, and movement status. The smart helmet sensing unit includes an accelerometer, a gyroscope, and a pressure sensor.

[0069] The core processing layer includes a collision detection module and an AI outbound call engine. The AI ​​outbound call engine's functions include progressive inquiry strategy, two-factor authentication, and multimodal analysis. Multimodal analysis includes speech recognition, timbre and emotion analysis, and ambient sound analysis.

[0070] The decision analysis layer includes a injury analysis engine. The functional modules of the injury analysis engine include an AF-level grading model and multi-dimensional assessment. The two models operate in parallel, and the assessment dimensions include state of consciousness, limb movement, and bleeding pain.

[0071] The business response layer includes an emergency response center and execution units. The emergency response center includes hierarchical strategy routing, and the execution units include a resource scheduling system and a human intervention hotline. The resource scheduling system includes hospital bed and ambulance dispatch, and the human intervention hotline includes police coordination and emergency contacts.

[0072] The aforementioned edge-side perception layer, decision analysis layer, and business response layer are also connected to a data storage center, which includes real-time data streams, injury case databases, and response record databases.

[0073] Based on the above architecture, this disclosed solution adopts an automatic triggering mechanism. After the helmet sensor detects an anomaly, an AI outbound call is initiated within a preset time (e.g., 120 seconds). Identity verification can employ a dual verification mechanism using the last few digits of the phone number and the rider's name. Accident verification can utilize a progressive questioning strategy, supporting the extraction of key information in multi-person dialogue scenarios. Injury assessment can be based on a multi-dimensional evaluation using voice features, semantic understanding, and timbre analysis.

[0074] As an example, the injury severity grading criteria can be shown in Table 1 below.

[0075] Table 1

[0076] In some embodiments, a dual-model parallel analysis can be employed, using a model with strong logical reasoning capabilities and a model with strong semantic understanding capabilities for joint decision-making. A dual-model architecture combining a general large language model and a reasoning-enhanced model is adopted, using the OR principle to determine incidents, and the AND principle for important incidents. Furthermore, inquiry strategies and scripts can be continuously optimized based on historical cases, a robust timeout retry mechanism and degradation scheme can be established, and the recognition accuracy can be continuously optimized based on actual cases.

[0077] In some specific application scenarios, the process of determining the severity of injuries and implementing accident intervention based on the above-mentioned voice interaction data analysis can be manifested in the following typical situations: In the first typical scenario, the system addresses the identification and response to clearly serious injuries. The system uses intelligent voice interaction to inquire whether a collision or fall has occurred with the delivery personnel. The delivery personnel confirm the accident and express pain (e.g., "There's been an accident," "Ouch," etc.), while explicitly requesting specific rescue intervention (e.g., requesting traffic police to the scene). During the data analysis phase, the system performs multi-dimensional analysis of the voice interaction data: at the text analysis level, it extracts key event words such as "accident"; at the timbre analysis level, it detects acoustic features such as painful groans, and the calculated pain index exceeds a preset threshold (e.g., greater than 0.8); at the semantic understanding level, it identifies the clear intention to request external professional rescue intervention. Based on these characteristics, the system classifies the injury as Level 1 (e.g., Grade A serious injury). Accordingly, the system executes accident intervention actions matching this level, including: immediately contacting the emergency center (e.g., 120) for medical rescue, notifying the delivery station manager to assist on-site, and activating the emergency plan to continuously track the rescue progress.

[0078] In the second typical scenario, timbre-assisted judgment is applied to complex environments. When complex background noise (such as third-party arguments) exists at the accident scene, making the delivery personnel's voice description unclear, relying solely on text analysis will not be sufficient to accurately identify injuries. In this case, the system employs a multimodal processing approach: first, noise reduction is performed on the collected voice interaction data to separate multi-person dialogues and extract the target voice of the delivery personnel; then, timbre analysis is performed to detect painful groaning features from the extracted voice; finally, a comprehensive understanding is achieved by combining the contextual background information of the accident. Based on the above analysis results, the system determines that the complexity of the current environment or injury exceeds the automatic processing threshold, thereby triggering a manual intervention mechanism, whereby a human customer service representative takes over the subsequent processing.

[0079] In the third typical scenario, accurate classification and response are addressed for minor injuries. The delivery worker reports a fall during voice interaction but explicitly states they are uninjured and can handle it themselves (e.g., "I tripped and fell, it's okay, I can handle it myself"). The system analyzes this voice interaction data: first, it confirms the objective fact of the fall; second, it performs timbre analysis, detecting no abnormal acoustic features such as painful groans; finally, it performs semantic understanding, recognizing the semantic intent of "no serious injury" and "can handle it independently." After comprehensive judgment, the system classifies it as a second-degree injury (e.g., Class B minor injury). Accordingly, the system executes matching accident intervention actions, including: sending simple medical or treatment instructions to the delivery worker, automatically initiating follow-up confirmation after a preset time (e.g., 2 hours), and notifying the delivery station manager to monitor the subsequent status of the delivery worker.

[0080] In the fourth typical scenario, the handling of abnormal interactions is addressed. When the delivery personnel's responses during voice interaction do not match the key questions posed by the system (such as whether the employee is injured), resulting in irrelevant answers, evasion of key questions, semantic confusion, or poor information quality (e.g., inquiring about irrelevant information such as the system's detection mechanism or equipment upgrade status), the system detects irrelevant answer patterns and missing key information through semantic and logical analysis. Based on this, the system classifies it as an abnormal level (e.g., "Other"). Accordingly, the system executes matching accident intervention actions, including: automatically transferring the call to a human operator, marking the interaction data as an abnormal case for subsequent evaluation model optimization training, and assigning a dedicated person for follow-up processing.

[0081] Based on the same inventive concept, this disclosure also provides a smart helmet, such as... Figure 5 As shown, the smart helmet includes a sensing module 501, a signal transmitting module 502, and an audio acquisition module 503.

[0082] Sensing module 501 is used to detect collision incidents; The signal sending module 502 is used to send a collision event signal to the delivery management server after the sensing module detects a collision accident, so that the delivery management server responds to the collision event signal to trigger an outbound call request to a target terminal with a call function associated with the delivery capacity wearing the smart helmet, so as to establish a voice call link. The audio acquisition module 503 is used to collect voice interaction data in the voice call link and send the voice interaction data to the delivery management server. The voice interaction data is used to analyze and assess the injury level of the delivery capacity and to perform accident interventions that match the injury level.

[0083] In this embodiment, at the moment an accident occurs, the helmet can proactively report the event and trigger the server to make an outbound call to the associated terminal without the rider needing to manually operate it. This effectively overcomes the limitation that riders cannot seek help in time due to injury, panic, or equipment limitations, and significantly shortens the emergency response time. At the same time, by relying on the helmet to directly collect interactive data in the established voice call link, first-hand voice information can be obtained in a low-interference state, providing reliable data support for accurately assessing the injury level, thereby ensuring that the server can execute graded intervention measures that are strictly matched to the injury.

[0084] Based on the same inventive concept, this disclosure also provides a collision accident response and processing platform, such as... Figure 6 As shown, the response and processing platform for the collision accident includes a signal receiving module 601, an outbound call triggering module 602, a data analysis module 603, and a response execution module 604.

[0085] The signal receiving module 601 is used to acquire collision event signals from target terminals associated with delivery capacity, including smart helmets and / or voice communication terminals. The outbound call triggering module 602 is used to respond to a collision event signal and trigger the management server to initiate an outbound call request to a target terminal with call function associated with the delivery capacity in order to establish a voice call link; The data analysis module 603 is used to acquire voice interaction data in the voice call link of outbound calls. The voice interaction data is used for analysis to assess and determine the injury level of delivery capacity. The response execution module 604 is used to trigger the management server to generate corresponding accident intervention actions based on the injury level, and to execute accident interventions that match the injury level.

[0086] The collision accident response and processing platform provided in this embodiment achieves a fully automated and intelligent closed-loop process from accident perception, communication establishment, injury assessment to rescue execution. Upon receiving a collision signal, the platform proactively initiates outbound calls to establish a voice link, effectively overcoming the limitation that delivery personnel may be unable to actively seek help after an accident due to injury, panic, or equipment limitations. Simultaneously, it intelligently assesses injury levels through in-depth analysis of voice interaction data, eliminating the subjectivity and lag of traditional manual inquiry and judgment, and achieving objective and accurate injury assessment. Finally, based on the assessment results, it triggers matching accident intervention actions, realizing on-demand allocation and precise hierarchical scheduling of rescue resources.

[0087] It should be noted that although several modules or units for action execution have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0088] Furthermore, some of the block diagrams shown in the attached figures are functional entities and do not necessarily correspond to physically or logically independent entities.

[0089] Those skilled in the art will understand that all or part of the steps of the above embodiments can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation, or an implementation combining hardware and software aspects.

[0090] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for responding to and handling collision accidents, characterized in that, include: Acquire collision event signals from target terminals associated with delivery capacity, the target terminals including smart helmets and / or voice communication terminals; In response to the collision event signal, the delivery management server is triggered to initiate an outbound call request to the target terminal with call function associated with the delivery capacity in order to establish a voice call link; Acquire voice interaction data in the outbound voice call link, and use the voice interaction data for analysis to assess and determine the injury level of the delivery capacity; The delivery management server generates corresponding accident intervention actions based on the injury level and executes the accident intervention that matches the injury level.

2. The method according to claim 1, characterized in that, The step of responding to the collision event signal by triggering the delivery management server to initiate an outbound call request to a target terminal with call functionality associated with the delivery capacity to establish a voice call link, and the step of obtaining voice interaction data in the outbound voice call link, includes: In response to the collision event signal, an outbound call request is sent to the target terminal bound to the delivery capacity identifier based on the delivery capacity identifier; Upon detecting that the target terminal has accepted the outbound call request, it is determined to establish the voice call link. The collision-related inquiry voice is sent to the delivery capacity through the voice call link. The collision-related inquiry voice is used to guide the delivery capacity to provide feedback on its current status information. The audio data collected by the target terminal is acquired and used to assess the injury level of the delivery personnel; wherein, the audio data includes the delivery personnel's response voice data to the inquiry voice feedback, and / or, environmental sound data of the accident scene.

3. The method according to claim 2, characterized in that, The step of obtaining the voice interaction data in the voice call link further includes: The next round of questioning voice is generated based on the progressive questioning dialogue and the voice interaction data based on the collision accident-related questioning voice, until the injury level of the terminal user can be assessed based on the currently collected voice interaction data, at which point the generation of the next round of questioning voice stops.

4. The method according to claim 1, characterized in that, Analyzing the voice interaction data to assess and determine the injury level of the delivery personnel includes: By performing semantic analysis on the content of the voice interaction data, the semantic feature data of the voice interaction data is determined, and the injury level of the delivery capacity is assessed based on the semantic feature data of the voice interaction data. And / or, by analyzing the coherence characteristics and timbre features of the speaker's voice dialogue in the voice interaction data, the acoustic speech feature data of the voice interaction data is determined, and the injury level of the delivery capacity is assessed based on the acoustic speech feature data of the voice interaction data.

5. The method according to claim 4, characterized in that, Analyzing the voice interaction data to assess and determine the injury level of the delivery personnel includes: Based on the semantic feature data of the voice interaction data, if it is determined that the voice of the delivery capacity in the voice interaction data contains one or more of the following keywords: emergency help keywords, severe pain keywords, or painful groans, then the injury level is determined to be a serious injury level. And / or, based on the acoustic speech feature data of the speech interaction data, if it is determined that the speech of multiple timbres in the speech interaction data contains injury-related keywords, then the injury level is determined to be a serious injury level.

6. The method according to claim 4, characterized in that, The injury level of the delivery personnel is assessed based on the semantic feature data of their voice recordings, including: Based on the semantic feature data of the voice interaction data, if it is determined that the voice of the delivery capacity in the voice interaction data contains one or more of the following keywords: self-processing injury keywords, mild pain keywords, or soft groans, then the injury level is determined to be a minor injury level. And / or, based on the acoustic speech feature data of the speech interaction data, if it is determined that the continuity characteristics of the delivery capacity speech in the speech interaction data meet the condition of unclear language expression, then the injury level is determined to be minor injury level.

7. The method according to claim 4, characterized in that, The injury level of the delivery personnel is assessed based on the semantic feature data of their voice recordings, including: Based on the semantic feature data of the voice interaction data, if it is determined that the voice of the delivery capacity in the voice interaction data contains the keyword "not injured", then the injury level is determined to be the "not injured" level. And / or, based on the acoustic speech feature data of the speech interaction data, if it is determined that the continuity characteristics of the speech corresponding to the delivery capacity timbre in the speech interaction data meet the conditions of normal dialogue, then the injury level is determined to be an uninjured level.

8. The method according to claim 2, characterized in that, The method of determining the injury level of the delivery personnel based on the voice feature data also includes: Analyze the ambient sound data in the voice interaction data to determine the impact intensity when the collision occurs; The severity of injury is determined by combining the impact intensity, the response voice data of the delivery capacity, and / or the environmental sound data of the accident scene.

9. A smart helmet, characterized in that, include: The perception module is used to detect collision incidents; The signal sending module is used to send a collision event signal to the delivery management server after the sensing module detects a collision accident, so that the delivery management server responds to the collision event signal to trigger an outbound call request to a target terminal with a call function associated with the delivery capacity wearing the smart helmet, so as to establish a voice call link. An audio acquisition module is used to acquire voice interaction data in the voice call link and send the voice interaction data to the delivery management server. The voice interaction data is used to analyze and assess the injury level of the delivery capacity and to perform accident interventions that match the injury level.

10. A collision accident response and processing platform, characterized in that, include: A signal receiving module is used to acquire collision event signals from target terminals associated with delivery capacity, the target terminals including smart helmets and / or voice communication terminals; The outbound call triggering module is used to respond to the collision event signal and trigger the management server to initiate an outbound call request to the target terminal with call function associated with the delivery capacity in order to establish a voice call link. The data analysis module is used to acquire voice interaction data in the voice call link of the outbound call, and the voice interaction data is used for analysis to assess and determine the injury level of the delivery capacity; The response execution module is used to trigger the management server to generate a corresponding accident intervention action based on the injury level, and to execute the accident intervention that matches the injury level.

Images