# US 11,308,330 B2 · Method and device for constructing autonomous driving test scenes, terminal and readable storage media

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

- Number: US 11,308,330 B2
- Document ID: US11308330B2
- Status: Granted
- Kind: B2
- Application number: 17/134,399
- Filed: 2020-12-26
- Published: 2022-01-06
- Granted: 2022-04-19
- Earliest claimed: 2020-12-26
- Assignees: Chang'an University
- Applicants: Chang'an University
- Inventors: Xiangmo Zhao; Runmin Wang; Yu Zhu; Wenshuai Zhou; Zhigang Xu; Zhanwen Liu; Jingjun Cheng; Lan Yang; Pengpeng Sun
- Web page: https://patentgrep.com/patent/US11308330B2

## CPC

- G06F 18/29: {Graphical models, e.g. Bayesian networks}
- G06V 20/41: {Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items (segmenting video sequences G06V20/49)}
- B60W 60/001: {Planning or execution of driving tasks}
- G06F 18/2415: based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06V 20/56: exterior to a vehicle by using sensors mounted on the vehicle
- G06F 30/15: Vehicle, aircraft or watercraft design
- G06F 30/20: Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30)
- G06F 2111/08: Probabilistic or stochastic CAD
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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

## Abstract

Provided are a method and a device for constructing autonomous driving test scenes, a terminal and a readable storage media. Primitive scenes are extracted from real traffic scenes to establish a primitive scene description model. The values of the description variables are selected from distribution intervals of the description variables of the primitive scenes, and the description variables are randomly sampled based on the distribution of the description variables to generate test primitive scenes by adopting an importance sampling based Monte Carlo method. The test primitive scenes are recombined according to the parameters of the elements of the test scene to generate the test scene. The present invention can directly simulate the actual complex system, so as to avoid the result distortion caused by the simplification of the complex system.

## Claims

### Claim 1

A method for constructing autonomous driving test scenes based on primitive scenes, comprising:
S1: extracting primitive scenes from real traffic scenes to establish a primitive scene description model, and estimating the distribution of description variables of the primitive scenes in the primitive scene description model using Gaussian Mixture Model;
S2: selecting values of the description variables from distribution intervals of the description variables of the primitive scenes according to the primitive scene description model, and randomly sampling the description variables based on the distribution of the description variables to generate a test primitive scene by adopting an importance sampling based Monte Carlo method; and
S3: determining parameters of elements of a test scene according to a test task, and recombining the test primitive scenes generated in S2 according to the parameters of the elements of the test scene to generate the test scene.

### Claim 2 (depends on 1)

The method of claim 1, wherein the real traffic scenes are decomposed into a plurality of independent primitives, and the independent primitives are modeled and parameterized.

### Claim 3 (depends on 1)

The method of claim 1, wherein the primitive scene description model is established to describe a probability distribution of the description variables of the primitive scenes; the Gaussian Mixture Model is adopted to estimate the distribution of the description variables of the primitive scenes;
X is supposed to be a random variable to be estimated, and the Gaussian Mixture Model of X is expressed as:
p
⁡
(
x
)
=
∑
k
=
1
N
⁢
⁢
α
k
⁢
N
⁡
(
x
❘
μ
k
,
Σ
k
)
;
wherein N(x|μk,Σk) is the kth component in the Gaussian Mixture Model, αk is the mixing coefficient which satisfies
∑
k
=
1
N
⁢
⁢
α
k
=
1
,
α
k
∈
[
0
,
1
]
;
during the parameter estimation, data generally obeys the Gaussian mixture distribution, and the probability distribution is represented by the Gaussian Mixture Model; the primitive scene description model contains N groups of unknown parameters (μk,Σk,αk); and the values of parameters of the primitive scene description model are determined according to the sample data.

### Claim 4 (depends on 3)

The method of claim 3, wherein the unknown parameters are subject to iterative calculation through an EM (Expectation-Maximization) algorithm to obtain optimal unknown parameters (μ,Σ,α).

### Claim 5 (depends on 4)

The method of claim 4, wherein the iterative calculation comprises:
1) setting an initial value of (μ,Σ,α) for each sample point xi;
2) calculating a posterior probability Qi(zk)=(zk|xi;μ,Σ,α) of each sample point xi according to the value of (μ,Σ,α);
3) calculating and updating the value of (μ,Σ,α) according to the posterior probability; and
4) calculating a likelihood function l by using the updated parameters of (μ,Σ,α) if the likelihood function converges to maximum likelihood estimation, obtaining optimal value of (μ,Σ,α); otherwise, returning to step 2 for iterative calculation.

### Claim 6 (depends on 5)

The method of claim 5, wherein when calculating the posterior probability for each sample point xi, the posterior probability is fixed to solve for the maximum likelihood estimation of P(x;μ,Σ,α);
a likelihood function of P(x;μ,Σ,α) is expressed as:
l
=
∑
i
=
1
n
⁢
⁢
∑
k
=
1
N
⁢
⁢
Q
i
⁡
(
z
k
)
⁢
⁢
ln
⁢
P
⁡
(
x
i
,
z
k
;
μ
,
Σ
,
α
)
Q
i
⁡
(
z
k
)
;
a derivative of l with respect μk is found, and the derivative of l is set to 0 to solve for μk:
μ
k
=
∑
i
=
1
n
⁢
⁢
Q
i
⁡
(
z
k
)
⁢
x
i
∑
i
=
1
n
⁢
⁢
Q
i
⁡
(
z
k
)
;
similarly, the derivative of l with respect to Σk and αk is found, respectively, and the derivative of l is set to 0 to solve for Σk and αk:
Σ
k
=
∑
i
=
1
n
⁢
⁢
Q
i
⁡
(
z
k
)
⁢
(
x
i
-
μ
k
)
⁢
(
x
i
-
μ
k
)
T
∑
i
=
1
n
⁢
⁢
Q
i
⁡
(
z
k
)
;
α
k
=
∑
i
=
1
n
⁢
⁢
Q
i
⁡
(
z
k
)
n
.

### Claim 7 (depends on 1)

The method of claim 1, wherein Monte Carlo method is adopted for the random sampling to generate the primitive scenes;
a random variable X is supposed to represent scenes generated by the Monte Carlo method; x is a scene sample of X and obeys the distribution of a probability density function ƒ(x); W is supposed to be a set of high-risk scenes; in order to select the high-risk scenes from the generated scenes, an indicator function h(x) of the high-risk scenes is defined as:
h
⁡
(
x
)
=
{
1
,
x
∈
W
0
,
x
∉
W
;
an expectation of the indicator function h(x) in the overall scene C is μ, and a variance is σ2;
under the condition of importance sampling, the expectation μ is expressed as:
μ
=
∫
h
⁡
(
x
)
⁢
f
⁡
(
x
)
⁢
dx
=
∫
h
⁡
(
x
)
⁢
f
⁡
(
x
)
g
⁡
(
x
)
⁢
g
⁡
(
x
)
⁢
dx
=
∫
(
h
⁡
(
x
)
⁢
ω
⁡
(
x
)
)
⁢
g
⁡
(
x
)
⁢
dx
;
a probability of occurrence of the high-risk scenes in the overall scene is supposed to be p, and the expectation μ of h(x) is equal to p, and the variance σ2 is equal to p(1−p); g(x) is a distribution with a higher probability of occurrence of the high-risk scenes, where
ω
⁡
(
x
)
=
f
⁡
(
x
)
g
⁡
(
x
)
is called an importance ratio;
the Monte Carlo method is adopted to generate n groups of scene samples xi, i=1, . . . , n; when a sample size is large, in terms of the law of Large Numbers, a sample mean {circumflex over (μ)} of h(x) is converged to the expectation μ with the probability of 1, and a Monte Carlo estimator μIS, of the expectation μ of the h(x) is expressed as:
=
1
n
⁢
∑
i
=
1
n
⁢
⁢
h
⁡
(
x
i
)
⁢
ω
⁡
(
x
i
)
,
x
i
∼
g
⁡
(
x
)
.

### Claim 8

A device for constructing autonomous driving test scenes based on primitive scenes, comprising:
a primitive extraction module, configured to extract primitive scenes from real traffic scenes to establish a primitive scene description model;
a variable estimation module, configured to estimate the distribution of description variables of the primitive scenes in the primitive scene description model by using a Gaussian Mixture Model;
a primitive scene generation module, configured to select values of description variables from distribution intervals of the description variables of the primitive scenes according to the primitive scene description model, and randomly sample the description variables by adopting an importance sampling based Monte Carlo method to generate test primitive scenes; and
a test scene generation module, configured to determine parameters of elements of a test scene according to a test task, and recombine the test primitive scenes to generate the test scene.

### Claim 9 (depends on 1)

A terminal, comprising:
a storage device;
a processor; and
a computer program stored in the storage device and executed by the processor;
wherein the computer program, when executed by the processor, causes the processor to execute the method of claim 1.

### Claim 10 (depends on 1)

A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program causes the processor to execute the method of claim 1.