# US 2026/0060581 A1 · Human-factor intelligent driving behavior prediction method and system, and terminal device and storage medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0060581 A1
- Document ID: US20260060581A1
- Status: Published application
- Kind: A1
- Application number: 19/379,788
- Filed: 2025-11-05
- Published: 2026-03-05
- Earliest claimed: 2023-08-07
- Assignees: Kingfar International Inc.
- Applicants: Kingfar International Inc.
- Inventors: Qichao ZHAO; Ran YANG; Qingju WANG
- Web page: https://patentgrep.com/patent/US20260060581A1

## CPC

- A61B 5/18: for vehicle drivers {or machine operators}
- G06N 3/045: Combinations of networks
- A61B 5/7257: {using Fourier transforms}
- B60W 40/02: related to ambient conditions
- B60W 40/09: Driving style or behaviour
- B60W 50/0097: {Predicting future conditions}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0475: Generative networks
- G06N 3/08: Learning methods
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06V 10/82: using neural networks
- G06V 20/56: exterior to a vehicle by using sensors mounted on the vehicle
- G06V 20/597: {Recognising the driver's state or behaviour, e.g. attention or drowsiness}
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- B60W: CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A human-factor intelligent driving behavior prediction method, a terminal device and a storage medium are provided. The method includes: obtaining a physiological signal; performing fast Fourier transform on the physiological signal to generate an amplitude-frequency characteristic and obtaining a sampling frequency in the amplitude-frequency characteristic; performing, based on a period of the sampling frequency, multi-period decomposition on the physiological signal to generate a data decomposition result sample; performing two-dimensional spatial expansion on the data decomposition result sample to generate two-dimensional spatial data; performing prediction on target consecutive frames to generate an iteratively predicted future frame; merging the two-dimensional spatial data, the target consecutive frames, and the iteratively predicted future frame to generate a multi-scale three-dimensional feature; analyzing the multi-scale three-dimensional feature to generate a target output feature; and analyzing the target output feature to generate driving behavior description information and driving behavior inference information.

## Claims

### Claim 1

A human-factor intelligent driving behavior prediction method, comprising:
obtaining a physiological signal of a driver;
performing fast Fourier transform on the physiological signal to generate an amplitude-frequency characteristic, and obtaining a sampling frequency in the amplitude-frequency characteristic that meets a predetermined amplitude-frequency selection criterion;
performing, based on a period of the sampling frequency, multi-period decomposition on the physiological signal, to generate a data decomposition result sample;
performing, based on a multivariate time-series data encoding layer, two-dimensional spatial expansion on the data decomposition result sample, to generate two-dimensional spatial data;
performing prediction, based on a vehicle road scene video frame prediction layer, on target consecutive frames corresponding to a vehicle road scene video, to generate an iteratively predicted future frame;
performing, based on a multi-modal synchronous data fusion layer, merging operation on the two-dimensional spatial data, the target consecutive frames, and the iteratively predicted future frame, to generate a multi-scale three-dimensional feature;
performing, based on a three-dimensional backbone network layer, feature analysis processing on the multi-scale three-dimensional feature, to generate a target output feature; and
respectively performing, based on a human-factor intelligent driving behavior explanation layer and a human-factor intelligent driving behavior inference layer, analysis processing on the target output feature, to generate human-factor intelligent driving behavior description information and human-factor intelligent driving behavior inference information.

### Claim 2 (depends on 1)

The human-factor intelligent driving behavior prediction method according to claim 1, wherein said performing, based on the three-dimensional backbone network layer, the feature analysis processing on the multi-scale three-dimensional feature, to generate the target output feature comprises:
obtaining a three-dimensional feature segmentation rule in the three-dimensional backbone network layer; and
dividing, according to the three-dimensional feature segmentation rule, the multi-scale three-dimensional feature into H/4×W/4×((2+N+5×3)/6) sub-features.

### Claim 3 (depends on 2)

The human-factor intelligent driving behavior prediction method according to claim 2, wherein subsequent to said dividing, according to the three-dimensional feature segmentation rule, the multi-scale three-dimensional feature into the H/4×W/4×((2+N+5×3)/6) sub-features, the method further comprises:
obtaining a linear encoding rule in the three-dimensional backbone network layer; and
linearly mapping, according to the linear encoding rule, each of the sub-features to a vector C, where the vector C is of an arbitrary number of dimensions.

### Claim 4 (depends on 1)

The human-factor intelligent driving behavior prediction method according to claim 1, wherein the three-dimensional backbone network layer comprises a self-attention encoding rule, and said performing, based on the three-dimensional backbone network layer, the feature analysis processing on the multi-scale three-dimensional feature, to generate the target output feature comprises:
S1: performing spatial sampling on the multi-scale three-dimensional feature to obtain a first target feature;
S2: performing a Video Swin Transformer blocks operation on the multi-scale three-dimensional feature, to obtain a second target feature, wherein an MPL layer in a model corresponding to the Video Swin Transformer blocks operation is a 1×1 convolutional layer, wherein a number of convolutional kernels is equal to a number of dimensions of the sub-features input to the model;
S3: repeatedly performing S1 and S2; and
S4: repeatedly performing S3 for K times, wherein K is a predetermined positive integer.

### Claim 5 (depends on 1)

The human-factor intelligent driving behavior prediction method according to claim 1, wherein the multi-scale three-dimensional feature has a size of H×W×(2+N+5×3), and has a number of channels of (2+N+5×3), wherein H is a height of a feature map in the multi-scale three-dimensional feature, and W is a width of the feature map in the multi-scale three-dimensional feature.

### Claim 6 (depends on 5)

The human-factor intelligent driving behavior prediction method according to claim 5, wherein a value of N in the number of channels is selected according to a preset selection rule to cause the number of channels to be an integer multiple of 6.

### Claim 7 (depends on 5)

The human-factor intelligent driving behavior prediction method according to claim 5, wherein the multi-scale three-dimensional feature is decomposed into sub-features having a size of H×W×3×((2+N+5×3)/3), wherein the first three dimensions redefine each frame in the multi-scale three-dimensional features, and each frame contains H×W×3 pixels.

### Claim 8

A terminal device, comprising a memory and a processor, wherein the memory stores a computer instruction executable on the processor, and the processor is configured to, when loading and executing the computer instruction, implement a human-factor intelligent driving behavior prediction method, the method comprising:
obtaining a physiological signal of a driver;
performing fast Fourier transform on the physiological signal to generate an amplitude-frequency characteristic, and obtaining a sampling frequency in the amplitude-frequency characteristic that meets a predetermined amplitude-frequency selection criterion;
performing, based on a period of the sampling frequency, multi-period decomposition on the physiological signal, to generate a data decomposition result sample;
performing, based on a multivariate time-series data encoding layer, two-dimensional spatial expansion on the data decomposition result sample, to generate two-dimensional spatial data;
performing prediction, based on a vehicle road scene video frame prediction layer, on target consecutive frames corresponding to a vehicle road scene video, to generate an iteratively predicted future frame;
performing, based on a multi-modal synchronous data fusion layer, merging operation on the two-dimensional spatial data, the target consecutive frames, and the iteratively predicted future frame, to generate a multi-scale three-dimensional feature;
performing, based on a three-dimensional backbone network layer, feature analysis processing on the multi-scale three-dimensional feature, to generate a target output feature; and
respectively performing, based on a human-factor intelligent driving behavior explanation layer and a human-factor intelligent driving behavior inference layer, analysis processing on the target output feature, to generate human-factor intelligent driving behavior description information and human-factor intelligent driving behavior inference information.

### Claim 9 (depends on 8)

The terminal device according to claim 8, wherein said performing, based on the three-dimensional backbone network layer, the feature analysis processing on the multi-scale three-dimensional feature, to generate the target output feature comprises:
obtaining a three-dimensional feature segmentation rule in the three-dimensional backbone network layer; and
dividing, according to the three-dimensional feature segmentation rule, the multi-scale three-dimensional feature into H/4×W/4×((2+N+5×3)/6) sub-features.

### Claim 10 (depends on 9)

The terminal device according to claim 9, wherein subsequent to said dividing, according to the three-dimensional feature segmentation rule, the multi-scale three-dimensional feature into the H/4×W/4×((2+N+5×3)/6) sub-features, the method further comprises:
obtaining a linear encoding rule in the three-dimensional backbone network layer; and
linearly mapping, according to the linear encoding rule, each of the sub-features to a vector C, where the vector C is of an arbitrary number of dimensions.

### Claim 11 (depends on 8)

The terminal device according to claim 8, wherein the three-dimensional backbone network layer comprises a self-attention encoding rule, and said performing, based on the three-dimensional backbone network layer, the feature analysis processing on the multi-scale three-dimensional feature, to generate the target output feature comprises:
S1: performing spatial sampling on the multi-scale three-dimensional feature to obtain a first target feature;
S2: performing a Video Swin Transformer blocks operation on the multi-scale three-dimensional feature, to obtain a second target feature, wherein an MPL layer in a model corresponding to the Video Swin Transformer blocks operation is a 1×1 convolutional layer, wherein a number of convolutional kernels is equal to a number of dimensions of the sub-features input to the model;
S3: repeatedly performing S1 and S2; and
S4: repeatedly performing S3 for K times, wherein K is a predetermined positive integer.

### Claim 12 (depends on 8)

The terminal device according to claim 8, wherein the multi-scale three-dimensional feature has a size of H×W×(2+N+5×3), and has a number of channels of (2+N+5×3), wherein H is a height of a feature map in the multi-scale three-dimensional feature, and W is a width of the feature map in the multi-scale three-dimensional feature.

### Claim 13 (depends on 12)

The terminal device according to claim 12, wherein a value of N in the number of channels is selected according to a preset selection rule to cause the number of channels to be an integer multiple of 6.

### Claim 14 (depends on 12)

The terminal device according to claim 12, wherein the multi-scale three-dimensional feature is decomposed into sub-features having a size of H×W×3×((2+N+5×3)/3), wherein the first three dimensions redefine each frame in the multi-scale three-dimensional features, and each frame contains H×W×3 pixels.

### Claim 15

A computer-readable storage medium, storing a computer instruction, wherein the computer instruction is configured to, when loaded and executed by a processor, implement a human-factor intelligent driving behavior prediction method, the method comprising:
obtaining a physiological signal of a driver;
performing fast Fourier transform on the physiological signal to generate an amplitude-frequency characteristic, and obtaining a sampling frequency in the amplitude-frequency characteristic that meets a predetermined amplitude-frequency selection criterion;
performing, based on a period of the sampling frequency, multi-period decomposition on the physiological signal, to generate a data decomposition result sample;
performing, based on a multivariate time-series data encoding layer, two-dimensional spatial expansion on the data decomposition result sample, to generate two-dimensional spatial data;
performing prediction, based on a vehicle road scene video frame prediction layer, on target consecutive frames corresponding to a vehicle road scene video, to generate an iteratively predicted future frame;
performing, based on a multi-modal synchronous data fusion layer, merging operation on the two-dimensional spatial data, the target consecutive frames, and the iteratively predicted future frame, to generate a multi-scale three-dimensional feature;
performing, based on a three-dimensional backbone network layer, feature analysis processing on the multi-scale three-dimensional feature, to generate a target output feature; and
respectively performing, based on a human-factor intelligent driving behavior explanation layer and a human-factor intelligent driving behavior inference layer, analysis processing on the target output feature, to generate human-factor intelligent driving behavior description information and human-factor intelligent driving behavior inference information.

### Claim 16 (depends on 1)

An apparatus, comprising:
a processor configured to invoke and execute a computer program from a memory, to cause a device on which the apparatus is mounted to implement the human-factor intelligent driving behavior prediction method according to claim 1.

### Claim 17 (depends on 1)

A computer program product, comprising a computer program instruction, the computer program instruction is configured to cause a computer to implement the human-factor intelligent driving behavior prediction method according to claim 1.
