A smart evidence collection method, system, and electronic device for courier delivery accidents.

Abstract

This invention relates to the technical field of intelligent evidence collection solutions for courier delivery accidents, specifically to an intelligent evidence collection method, system, and electronic device for courier delivery accidents. The method includes: triggering multi-source evidence collection by detecting acceleration mutations through the delivery terminal's IMU; simultaneously using dual cameras and microphones to collect video streams, depth maps, and environmental audio; analyzing courier actions and cargo trajectories based on a spatiotemporal attention mechanism; detecting collisions and arguments through audio spectrum analysis; fusing IMU, video, and audio data to construct a time-stamped causal event graph; using a pre-trained responsibility quantification model to output responsibility coefficients for the courier, third parties, and environmental factors; writing encrypted data and geofencing information into the blockchain; and generating an intelligent report containing a 3D reconstruction animation and a responsibility matrix. This invention achieves automatic alignment of multimodal data and blockchain-based evidence consolidation, significantly reducing errors in accident responsibility determination.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent evidence collection schemes for courier delivery accidents, specifically to an intelligent evidence collection method, system, and electronic equipment for courier delivery accidents. Background Technology

[0002] The current express delivery industry largely relies on manual review to handle delivery accidents, which has significant technical shortcomings. First, evidence collection methods are passive and outdated: traditional methods depend on retrieving surveillance footage or user complaints after the fact, and crucial dynamic information at the moment of the accident (such as the trajectory of goods slipping from the hands, the timing of external impacts) is permanently lost because equipment fails to activate in time. Second, the chain of evidence is fragmented: video, audio, and motion sensor data are stored independently with misaligned timestamps, making it difficult to establish multimodal event correlations and making it impossible to trace third-party interference. Finally, liability determination is highly subjective: the lack of automated analysis models makes manual assessment susceptible to experience bias, especially inaccurate estimation of environmental risk weights such as slippery roads and slopes. Although some solutions use continuous recording via dashcams, this continuous recording mode generates a large amount of redundant data, increasing terminal power consumption and making it difficult to retrieve effective accident footage. Furthermore, existing blockchain evidence storage is mostly limited to simple data on-chaining and lacks integration with 3D physical simulation, resulting in limited legal validity. Therefore, there is an urgent need to develop an evidence collection solution that can be proactively triggered, multi-source synchronized, and intelligently attributed to address the core pain points of insufficient evidence completeness and inaccurate liability quantification.

[0003] Therefore, existing technologies still need further development. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide an intelligent evidence collection method, system, and electronic device for courier delivery accidents, so as to solve the problems existing in the prior art.

[0005] To achieve the above-mentioned technical objectives, according to a first aspect of the present invention, the present invention provides an intelligent evidence collection method for courier delivery accidents, comprising:

[0006] S100: The inertial measurement unit built into the delivery terminal monitors the courier's movement status in real time. When a sudden acceleration change is detected that exceeds the preset threshold, the multi-source evidence collection module is automatically activated, and the front camera, rear camera and microphone of the delivery terminal are triggered simultaneously to collect multi-angle video streams, depth map data and environmental audio of the accident scene.

[0007] S200 performs real-time frame sequence analysis on the video stream, identifies the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; and detects key event points such as collision sounds and human arguments based on the audio spectrum features collected by the microphone.

[0008] S300: Align IMU data, video motion features, and audio event points across modalities to construct a causal event graph with synchronized timestamps; call a pre-trained accident liability quantification model to output the courier liability coefficient, third-party liability coefficient, and environmental liability coefficient based on the causal event graph;

[0009] S400 encrypts multi-source evidence data, its timestamps, and geofence information and writes them into the blockchain distributed ledger; it generates an intelligent evidence report containing a 3D scene reconstruction animation, a responsibility coefficient matrix, and a blockchain evidence number.

[0010] Specifically, the spatiotemporal attention mechanism includes:

[0011] Track key points of the human skeleton in a video frame sequence and calculate the rate of change of the torso tilt angle of the courier.

[0012] The trajectory of a parabolic motion is detected using the YOLO model, and the coordinates of its collision area with the ground are marked.

[0013] Specifically, the cross-modal alignment employs a dynamic time warping algorithm, and the alignment method satisfies:

[0014] The deviation between the peak moment of IMZ axial acceleration and the frame in the video where the goods leave the hand is controlled within ±100ms;

[0015] The time synchronization error between the audio voiceprint mutation point and the video collision frame is less than 50% of the audio frame length.

[0016] Specifically, the training method for the accident liability quantification model includes:

[0017] Generative adversarial networks were used to synthesize hundreds of thousands of delivery accident scenario samples.

[0018] The weakly supervised learning framework is used to label the three types of responsibility: deliveryman's operational errors, third-party collisions, and slippery road surfaces.

[0019] Specifically, it also includes:

[0020] When a verbal argument is detected, the voice emotion analysis module is activated to extract the anger index and frequency of threatening keywords from the voiceprint features, and to generate a chain of evidence of third-party provocative behavior.

[0021] Specifically, the voice emotion analysis employs:

[0022] Mel frequency cepstral coefficients are combined with two-stream feature extraction from the BERT speech encoder;

[0023] The time interval of abusive statements is located by activating gradient class mapping.

[0024] Specifically, the generation of the 3D scene reconstruction animation includes:

[0025] Constructing a point cloud model of the accident scene using parallax maps from binocular cameras;

[0026] The physics engine is used to simulate the transfer of rotational momentum during the cargo dropping process.

[0027] Specifically, it also includes:

[0028] The Beidou / GPS dual-mode positioning module of the delivery terminal is invoked to match the location data with the electronic fence database and identify the scene attributes of the accident, such as steps, slopes or no-parking zones.

[0029] According to a second aspect of the present invention, an intelligent evidence collection system for courier delivery accidents is provided, comprising:

[0030] An embedded sensing module, including an inertial measurement unit built into the delivery terminal, is used to monitor the courier's movement status in real time;

[0031] The multi-source evidence collection module includes a front-facing camera, a rear-facing camera, and a microphone at the delivery terminal, which respectively collect multi-angle video streams, depth map data, and environmental audio from the accident scene;

[0032] The control module automatically activates the multi-source evidence collection module when a sudden acceleration change exceeds a preset threshold; it performs real-time frame sequence analysis on the video stream, identifying the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; it detects key event points such as collision sounds and verbal arguments based on the audio spectrum features collected by the microphone; it performs cross-modal alignment of IMU data, video action features, and audio event points to construct a causal event graph with synchronized timestamps; it calls a pre-trained accident liability quantification model to output the courier's liability coefficient, third-party liability coefficient, and environmental liability coefficient based on the causal event graph; it encrypts the multi-source evidence collection data, its timestamps, and geofence information and writes them into the blockchain distributed ledger; and it generates an intelligent evidence collection report containing a 3D scene reconstruction animation, a liability coefficient matrix, and a blockchain evidence number.

[0033] According to a third aspect of the present invention, an electronic device is provided, comprising: a memory; and a processor, wherein the memory stores computer-readable instructions, which, when executed by the processor, implement the above-described intelligent evidence collection method for courier delivery accidents.

[0034] Beneficial effects:

[0035] (1) Innovation in evidence capture capability: By linking IMU thresholds to activate multiple sensors synchronously, the time blind spot of passive evidence collection is broken through, ensuring dynamic coverage of the entire accident process;

[0036] (2) Intelligent determination of responsibility: The causal event graph constructed by cross-modal alignment eliminates spatiotemporal bias, and the pre-trained responsibility model integrates physical laws and behavioral patterns to output accurate three-party responsibility coefficients;

[0037] (3) Enhanced legal effect: The combination of blockchain evidence storage and three-dimensional physical engine enables video actions, environmental parameters and voiceprint features to form an immutable closed-loop evidence chain.

[0038] (4) Embedded sensing module achieves millisecond-level response: The six-axis IMU and dual cameras work together to meet the real-time processing requirements of high dynamic scenes;

[0039] (5) Strong terminal compatibility: It can be adapted to the hardware configuration of mainstream delivery terminals, and the Beidou / GPS dual-mode positioning ensures the recognition of scene attributes in complex urban environments;

[0040] (6) Excellent power consumption control: The full-function module is activated only when there is a sudden acceleration change, which reduces energy consumption by 70% compared to the continuous recording solution.

[0041] Overall improvement: It has greatly improved the efficiency of accident liability determination, significantly shortened the judicial dispute resolution cycle, and significantly optimized the operating costs of express delivery companies. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the intelligent evidence collection method for courier delivery accidents provided in a specific embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the system composition of the intelligent evidence collection system for courier delivery accidents provided in a specific embodiment of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0045] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0046] Please see Figure 1 This invention provides an intelligent evidence collection method for delivery accidents involving couriers, comprising:

[0047] S100 monitors the courier's movement status in real time through the inertial measurement unit built into the delivery terminal. When a sudden acceleration change is detected that exceeds the preset threshold, the multi-source evidence collection module is automatically activated, and the front camera, rear camera and microphone of the delivery terminal are triggered simultaneously to collect multi-angle video streams, depth map data and environmental audio of the accident scene.

[0048] Specifically, step S100 includes:

[0049] ①IMU activation threshold setting:

[0050] A six-axis IMU (including a 3-axis accelerometer and a 3-axis gyroscope) is used, with a sampling frequency ≥100Hz. Acceleration mutation threshold: activation is triggered when the combined three-axis acceleration value meets the following conditions:

[0051]

[0052] in:

[0053] , , Instantaneous measurements of the accelerometer along the X / Y / Z axes (unit: g)

[0054] Triaxial composite acceleration modulus;

[0055] : Rate of change of acceleration (unit: g / s).

[0056] Preferred values ​​are based on:

[0057] Based on IMU data analysis of 500 real express delivery incidents:

[0058] Fall scenario: peak acceleration 4.2±1.3g, rate of change 28±7g / s;

[0059] Projectile scenario: peak acceleration 5.8±2.1g, rate of change 32±9g / s;

[0060] The threshold is set higher than the upper limit of normal movement (maximum 2.5g / 12g / s for running).

[0061] ② The multi-source data acquisition parameters are shown in Table 1:

[0062] Table 1 Multi-source data acquisition parameters

[0063]

[0064] S200 performs real-time frame sequence analysis on the video stream, identifies the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; and detects key event points such as collision sounds and arguments based on the audio spectrum features collected by the microphone.

[0065] Specifically, the spatiotemporal attention mechanism includes:

[0066] Track key points of the human skeleton in a video frame sequence and calculate the rate of change of the torso tilt angle of the courier.

[0067] The trajectory of a parabolic motion is detected based on the YOLO-X model, and the coordinates of its collision area with the ground are marked.

[0068] It should be further explained that, regarding step S200, the solution designed in this invention includes:

[0069] ① Design a human skeleton tracking solution:

[0070] Seventeen key points were extracted using the Open Pose model to calculate the torso tilt angle:

[0071]

[0072] in:

[0073] : Coordinates (pixels) of key points of the left / right hip joint;

[0074] : Coordinates of the center point of the shoulder / hip;

[0075] Angle between the torso and the vertical direction (unit: degrees).

[0076] Judgment threshold:

[0077] Imbalance state: Normal walking ;

[0078] Basis: Biomechanical experiments show The center of gravity shifts beyond the stability boundary.

[0079] ② Cargo trajectory detection:

[0080] The YOLO-V10 model has an input size of 640×640 and a confidence threshold of 0.7.

[0081] Equation of the parabolic trajectory:

[0082]

[0083] in:

[0084] The initial velocity upon release (calculated by the displacement difference between adjacent frames);

[0085] Represents the real-time position coordinates of the object within a three-dimensional space;

[0086] Represents the initial position vector;

[0087] Represents a time variable;

[0088] This represents the gravitational acceleration vector.

[0089] ③ Audio event detection:

[0090] Key acoustic feature extraction:

[0091]

[0092] Specifically:

[0093] The collision sound determination criteria are: >80dB;

[0094] The criteria for determining the sound of an argument are: .

[0095] Specifically, the physical meaning and numerical basis of the components of the above collision sound formula are shown in Table 2:

[0096] Table 2. Physical meaning and numerical settings of the components of the collision sound formula.

[0097]

[0098] Specifically, the physical meaning and numerical basis of the components of the above-mentioned argument formula are shown in Table 3:

[0099] Table 3. Physical meanings and numerical settings of the components of the argument formula.

[0100]

[0101] S300: Align IMU data, video motion features, and audio event points across modalities to construct a causal event graph with synchronized timestamps; call a pre-trained accident liability quantification model to output the courier liability coefficient, third-party liability coefficient, and environmental liability coefficient based on the causal event graph.

[0102] Specifically, the cross-modal alignment employs a dynamic time warping algorithm, and the alignment method satisfies:

[0103] The deviation between the peak moment of IMZ axial acceleration and the frame in the video where the goods leave the hand is controlled within ±100ms;

[0104] The time synchronization error between the audio voiceprint mutation point and the video collision frame is less than 50% of the audio frame length.

[0105] It should be further noted that the multimodal alignment scheme designed in this invention includes:

[0106] Design a Dynamic Time Warping (DTW) algorithm:

[0107] Define the distance function:

[0108]

[0109] in: Cumulative distance matrix;

[0110] Euclidean distance of eigenvectors ;

[0111] Video feature vectors ;

[0112] : Y-coordinate of the bottom edge of the bounding box;

[0113] , : The speed of horizontal / vertical movement of the package;

[0114] IMU feature vectors ;

[0115] Z-axis acceleration;

[0116] : Y-axis angular velocity;

[0117] Feature vector ;

[0118] Synchronization constraints: .

[0119] Specifically, the training method for the accident liability quantification model includes:

[0120] Generative adversarial networks were used to synthesize hundreds of thousands of delivery accident scenario samples.

[0121] The weakly supervised learning framework is used to label the three types of responsibility: deliveryman's operational errors, third-party collisions, and slippery road surfaces.

[0122] It should be further explained that, regarding the accident liability quantification model, the scheme designed in this invention includes:

[0123] ①GAN sample generation:

[0124] Generator structure: 5-layer ResNet, latent space dimension z=128;

[0125] Loss function:

[0126] in:

[0127] Discriminator;

[0128] : Generator;

[0129] Gradient penalty coefficient (optimized value 0.001);

[0130] : Noise distribution (Gaussian distribution N(0,1)).

[0131] 2. Calculation of Responsibility Coefficient

[0132] Graph Neural Network Inference:

[0133]

[0134] in:

[0135] The responsibility coefficient for delivery personnel;

[0136] The liability coefficient for third parties;

[0137] Environmental responsibility coefficient;

[0138] Accident feature vector, including: maximum acceleration, fall height, and road slope;

[0139] First layer weight (256×300);

[0140] Output layer weights (300×3);

[0141] Softmax: Normalization process;

[0142] Penalty coefficient (Preventing gradient explosion);

[0143] This is the activation function.

[0144] Specifically, it also includes:

[0145] When a verbal argument is detected, the voice emotion analysis module is activated to extract the anger index and frequency of threatening keywords from the voiceprint features, and to generate a chain of evidence of third-party provocative behavior.

[0146] Specifically, the voice emotion analysis employs:

[0147] Mel frequency cepstral coefficients are combined with two-stream feature extraction from the BERT speech encoder;

[0148] The time interval of abusive statements is located by activating gradient class mapping.

[0149] It should be further explained that, regarding voice emotion analysis, the solution designed in this invention includes:

[0150] 1. Design a dual-stream feature extraction scheme:

[0151] Dual-stream feature: MFCC+BERT speech coding;

[0152] MFCC stands for Mel-frequency cepstral coefficients, used to characterize acoustic physical features such as fundamental frequency and formants; BERT speech coding is a Transformer-based semantic vector used to capture linguistic contextual information (such as threat semantics).

[0153] 2. Design an anger index model:

[0154] Anger Index: ;

[0155] Specifically, the meaning and technical function of the components of the anger index formula are shown in Table 4;

[0156] Table 4. Meaning and Technical Function of the Components in the Anger Index Formula

[0157]

[0158] Weight settings: (Insulting terms such as "break" / "pay up") (A sudden rise in tone)

[0159] It needs to be further explained that the weight The allocation logic is shown in Table 5:

[0160] Table 5. Meaning and Technical Function of the Components in the Anger Index Formula

[0161]

[0162] Judgment rules:

[0163] Ordinary disputes: ;

[0164] Malicious conflict: (Triggering additional evidence storage).

[0165] ④ Voice sentiment analysis:

[0166] The following specific examples illustrate this:

[0167] Scenario: A dispute arises between the deliveryman and the recipient;

[0168] ① Event: Timeline: 0-2.4 seconds (8 300ms segments);

[0169] ②Key events:

[0170] t3: The recipient says "damaged", so we get w3=0.6, ||Grad-CAM||=0.82;

[0171] t5: Pitch rises sharply (Δf=52Hz / ms), resulting in w5=0.4, ||Grad-CAM||=0.73;

[0172] Other time periods: ||Grad-CAM||≈0.1 (stable mood).

[0173] ③ Anger Index Calculation:

[0174]

[0175] System response:

[0176] A value greater than 30 is considered a malicious conflict, and video evidence will be automatically saved.

[0177] Technical advantages:

[0178] 1. Semantic-acoustic dual-drive: MFCC captures intonation changes + BERT understands aggressive semantics;

[0179] 2. Enhanced legal effect: The weighting is set to match the Civil Code's standards for determining "verbal insults";

[0180] 3. Resource optimization: Data is encrypted and stored only during high-weight periods (significantly reducing blockchain storage requirements).

[0181] S400 encrypts multi-source evidence data, its timestamps, and geofence information and writes them into the blockchain distributed ledger; it generates an intelligent evidence report containing a 3D scene reconstruction animation, a responsibility coefficient matrix, and a blockchain evidence number.

[0182] Specifically, the generation of the 3D scene reconstruction animation includes:

[0183] Constructing a point cloud model of the accident scene using parallax maps from binocular cameras;

[0184] The physics engine is used to simulate the transfer of rotational momentum during the cargo dropping process.

[0185] It should be further explained that, regarding 3D scene reconstruction, the solution designed in this invention includes designing a collision physics model:

[0186]

[0187] in:

[0188] Total energy loss (unit: joules);

[0189] Material elasticity coefficient - Cardboard box: N / m (tested according to GB / T6544 standard);

[0190] Deformation (measured via depth map);

[0191] Friction coefficient - asphalt pavement: (GB50009-2012);

[0192] Package quality (obtained from the waybill database);

[0193] Gravitational acceleration;

[0194] Gliding distance (measured visually);

[0195] Rotational energy conversion coefficient (optimized value 0.4);

[0196] : Moment of inertia of the enclosure - cuboid:

[0197]

[0198] in:

[0199] h represents the length of the cuboid, h represents the height of the cuboid, and m represents the mass of the cuboid.

[0200] Angular velocity (calculated by differential analysis of consecutive frames).

[0201] Specifically, it also includes:

[0202] The Beidou / GPS dual-mode positioning module of the delivery terminal is invoked to match the location data with the electronic fence database and identify the scene attributes of the accident, such as steps, slopes or no-parking zones.

[0203] Furthermore, the method also includes:

[0204] 1. Dynamic adjustment of responsibility coefficient:

[0205] Input: Scene tags matched from the geofencing database;

[0206] The correction rules are shown in Table 6:

[0207] Table 6 Correction Rules

[0208]

[0209] Output: Updated responsibility coefficient matrix .

[0210] The following specific examples illustrate this:

[0211] The slope has a gradient of 12°. Increase by 0.6 (original value 0.24, corrected to 0.84).

[0212] 2. Enhanced 3D Physical Modeling

[0213] ①Staircase scene:

[0214] Simulates the multi-stage collision trajectory of a falling package:

[0215]

[0216] in:

[0217] The total energy loss of the package is calculated, and the results are used to quantify the degree of collision damage.

[0218] Number of steps (identified by depth map), the number of steps is identified by ToF depth map;

[0219] The i-th level drop height is calculated using the parallax of the binocular cameras.

[0220] Step inclination angle, measured by an IMU gyroscope;

[0221] The elastic coefficient of the packaging material is a preset value (for corrugated cardboard boxes: k=1200N / m).

[0222] For package quality, retrieve from the waybill database;

[0223] g is the acceleration due to gravity;

[0224] The coefficient of friction of the step surface is preset based on the material (cement step: μ=0.3).

[0225] Let be the sliding distance of the i-th step, calculated by tracking the displacement of the bounding box.

[0226] ②Slope scene:

[0227] Gliding distance calculation:

[0228] Dynamic friction coefficient (μ=0.15 for wet and slippery surfaces).

[0229] in:

[0230] The total sliding distance covered by the slope is calculated, and the result is used to locate the collision point and determine liability.

[0231] The initial velocity at the start of gliding is included, and the displacement difference between adjacent video frames is calculated (accuracy ±0.2m / s).

[0232] g is the acceleration due to gravity;

[0233] μ is the dynamic friction coefficient, which is taken as 0.15 for wet and slippery road surfaces, according to the road friction coefficient standard "JT / T715-2008" (measured value of wet asphalt is 0.12-0.18).

[0234] θ is the angle between the slope and the horizontal plane, measured by an IMU gyroscope;

[0235] The cosine of the slope angle is used to calculate the vertical component effect of the friction force.

[0236] The sine of the slope angle is used to calculate the gravitational component effect.

[0237] ③No-stopping zone:

[0238] Overlay parameters for duration of illegal stay: This triggers a full liability determination.

[0239] 3. Blockchain-based targeted evidence storage.

[0240] ①Important documentation for no-parking zones:

[0241] Stores the electronic map geofence registration number + courier's stop timestamp;

[0242] ②Slope / Staircase Scene:

[0243] Secure key evidence: Hash value of the measurement certificate for slope / step height (compliant with GB50026-2020 Engineering Surveying Standard).

[0244] 4. Intelligent report generation:

[0245] ①Enhanced scene identification:

[0246] Label the geographic risk in the 3D reconstruction animation (example: slope gradient 12°).

[0247] ② The responsibility matrix has been rearranged, as shown in Table 7:

[0248] Table 7. Schematic diagram of responsibility matrix rearrangement

[0249]

[0250] The report concludes: "The accident occurred on a slope (12°), and environmental responsibility accounts for 84%. It is recommended that the courier be exempted from compensation."

[0251] Advantages of the technical solution:

[0252] 1. Dynamic responsibility mapping:

[0253] Scene attributes directly drive responsibility weights (step height / slope, linear increment);

[0254] No-stopping zones trigger punitive weights (breaking the Softmax constraint).

[0255] 2. Enhanced physics engine for improved credibility:

[0256] The slope sliding formula is based on Newtonian mechanics, which greatly reduces calculation errors;

[0257] The energy accumulation model of step collisions was verified by ABAQUS simulation.

[0258] 3. Closed loop of legal effect:

[0259] National standards for evidence preservation;

[0260] The combination of 3D animation and correction coefficient matrix forms a strong chain of evidence.

[0261] Please see Figure 2 The present invention provides another embodiment, which provides an intelligent evidence collection system for courier delivery accidents, the intelligent evidence collection system for courier delivery accidents includes:

[0262] The embedded sensing module 100 includes an inertial measurement unit built into the delivery terminal for real-time monitoring of the courier's movement status.

[0263] The multi-source evidence collection module 200 includes a front-facing camera, a rear-facing camera, and a microphone at the delivery terminal, which respectively collect multi-angle video streams, depth map data, and environmental audio from the accident scene;

[0264] The control module 300 is used to automatically activate the multi-source evidence collection module when a sudden acceleration change exceeds a preset threshold; to perform real-time frame sequence analysis on the video stream, and to identify the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; to detect key event points such as collision sounds and human voice arguments based on the audio spectrum features collected by the microphone; to perform cross-modal alignment of IMU data, video action features and audio event points to construct a causal event graph with synchronized timestamps; to call a pre-trained accident liability quantification model and output the courier liability coefficient, third-party liability coefficient and environmental liability coefficient according to the causal event graph; to encrypt the multi-source evidence collection data and its timestamps and geofence information and write it into the blockchain distributed ledger; and to generate an intelligent evidence collection report containing a 3D scene reconstruction animation, a liability coefficient matrix and a blockchain evidence number.

[0265] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising:

[0266] The device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the intelligent evidence collection method for courier delivery accidents. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0267] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0268] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0269] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0270] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent evidence collection in delivery accidents involving couriers, characterized in that, The method includes: S100: The inertial measurement unit built into the delivery terminal monitors the courier's movement status in real time. When a sudden acceleration change is detected that exceeds the preset threshold, the multi-source evidence collection module is automatically activated, and the front camera, rear camera and microphone of the delivery terminal are triggered simultaneously to collect multi-angle video streams, depth map data and environmental audio of the accident scene. S200 performs real-time frame sequence analysis on the video stream, identifies the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; and detects key event points such as collision sounds and human arguments based on the audio spectrum features collected by the microphone. S300: Align IMU data, video motion features, and audio event points across modalities to construct a causal event graph with synchronized timestamps; call a pre-trained accident liability quantification model to output the courier liability coefficient, third-party liability coefficient, and environmental liability coefficient based on the causal event graph; S400: Encrypt multi-source evidence data, its timestamps, and geofence information, and write them into the blockchain distributed ledger; generate an intelligent evidence report containing a 3D scene reconstruction animation, a responsibility coefficient matrix, and a blockchain evidence number; The training method for the accident liability quantification model includes: Generative adversarial networks were used to synthesize hundreds of thousands of delivery accident scenario samples. The three-element liability label is labeled by using a weakly supervised learning framework to identify courier operational errors, third-party collisions, and slippery road surfaces. Specifically, the liability coefficient is calculated using the following formula: in: The responsibility coefficient for delivery personnel; The liability coefficient for third parties; Environmental responsibility coefficient; Accident feature vector, including maximum acceleration, fall height, and road slope; First-level weights; Output layer weights; Softmax: Normalization process; This is the activation function.

2. The intelligent evidence collection method for delivery accidents by couriers according to claim 1, characterized in that, The spatiotemporal attention mechanism specifically includes: Track key points of the human skeleton in a video frame sequence and calculate the rate of change of the torso tilt angle of the courier. The trajectory of a parabolic motion is detected using the YOLO model, and the coordinates of its collision area with the ground are marked.

3. The intelligent evidence collection method for delivery accidents by couriers according to claim 1, characterized in that, The cross-modal alignment employs a dynamic time warping algorithm, and the alignment method satisfies: The deviation between the peak moment of IMZ axial acceleration and the frame in the video where the goods leave the hand is controlled within ±100ms; The time synchronization error between the audio voiceprint mutation point and the video collision frame is less than 50% of the audio frame length.

4. The intelligent evidence collection method for courier delivery accidents according to claim 1, characterized in that, Also includes: When a verbal argument is detected, the voice emotion analysis module is activated to extract the anger index and frequency of threatening keywords from the voiceprint features, and to generate a chain of evidence of third-party provocative behavior.

5. The intelligent evidence collection method for delivery accidents by couriers according to claim 4, characterized in that, The voice sentiment analysis uses: Mel frequency cepstral coefficients are combined with two-stream feature extraction from the BERT speech encoder; The time interval of abusive statements is located by activating gradient class mapping.

6. The intelligent evidence collection method for delivery accidents by couriers according to claim 1, characterized in that, The generation of the 3D scene reconstruction animation includes: A point cloud model of the accident scene was constructed using parallax maps from binocular cameras. The physics engine is used to simulate the transfer of rotational momentum during the cargo dropping process.

7. The intelligent evidence collection method for delivery accidents by couriers according to claim 1, characterized in that, Also includes: The Beidou / GPS dual-mode positioning module of the delivery terminal is invoked to match the location data with the electronic fence database and identify the scene attributes of the accident, such as steps, slopes or no-parking zones.

8. A smart evidence collection system for delivery accidents involving couriers, characterized in that, The intelligent evidence collection method for courier delivery accidents according to any one of claims 1 to 7 includes: An embedded sensing module, including an inertial measurement unit built into the delivery terminal, is used to monitor the courier's movement status in real time; The multi-source evidence collection module includes a front-facing camera, a rear-facing camera, and a microphone at the delivery terminal, which respectively collect multi-angle video streams, depth map data, and environmental audio from the accident scene; The control module automatically activates the multi-source evidence collection module when a sudden acceleration change exceeds a preset threshold; it performs real-time frame sequence analysis on the video stream, identifying the courier's body movements and the trajectory of goods being thrown through a spatiotemporal attention mechanism; it detects key event points such as collision sounds and verbal arguments based on the audio spectrum features collected by the microphone; it performs cross-modal alignment of IMU data, video action features, and audio event points to construct a causal event graph with synchronized timestamps; it calls a pre-trained accident liability quantification model to output the courier's liability coefficient, third-party liability coefficient, and environmental liability coefficient based on the causal event graph; it encrypts the multi-source evidence collection data, its timestamps, and geofence information and writes them into the blockchain distributed ledger; and it generates an intelligent evidence collection report containing a 3D scene reconstruction animation, a liability coefficient matrix, and a blockchain evidence number.

9. An electronic device, characterized in that, include: Memory; The memory stores computer-readable instructions that, when executed by the processor, implement the intelligent evidence collection method for courier delivery accidents according to any one of claims 1 to 7.

Images