CN111985386A - Method for identifying pedestrian illegal-passing behavior based on planned behavior theory - Google Patents

Method for identifying pedestrian illegal-passing behavior based on planned behavior theory Download PDF

Info

Publication number
CN111985386A
CN111985386A CN202010824900.3A CN202010824900A CN111985386A CN 111985386 A CN111985386 A CN 111985386A CN 202010824900 A CN202010824900 A CN 202010824900A CN 111985386 A CN111985386 A CN 111985386A
Authority
CN
China
Prior art keywords
pedestrian
violation
behavior
passing
scene
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010824900.3A
Other languages
Chinese (zh)
Other versions
CN111985386B (en
Inventor
郑四发
吴浩然
王裕宁
黄荷叶
王建强
许庆
李克强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202010824900.3A priority Critical patent/CN111985386B/en
Publication of CN111985386A publication Critical patent/CN111985386A/en
Application granted granted Critical
Publication of CN111985386B publication Critical patent/CN111985386B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
    • G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Multimedia (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Human Computer Interaction (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a method for identifying the illegal pedestrian crossing behavior based on a planned behavior theory, which comprises the following steps: step 1, collecting related perception information in a current traffic scene; step 2, identifying the degree of the pedestrian according to the individual characteristics of the pedestrian in the relevant perception information; step 3, identifying the crowd effect of the pedestrian according to the pedestrian population characteristics in the relevant perception information; step 4, acquiring main influence factors of the pedestrian violation trafficking behaviors according to historical pedestrian violation trafficking data similar to or identical to the current traffic scene; and 5, fusing the results obtained in the step 2, the step 3 and the step 4 through a planned behavior theory to obtain the violation passing intention of the pedestrian, further identifying the violation passing behavior of the pedestrian, and outputting an identification result. The invention can accurately identify the pedestrian violation trafficking intention in real time, effectively identify the pedestrian violation trafficking behavior and further support the driving decision.

Description

Method for identifying pedestrian illegal-passing behavior based on planned behavior theory
Technical Field
The invention relates to the technical field of safe driving of intelligent traffic systems, in particular to a method for identifying the illegal pedestrian crossing behavior based on a planned behavior theory.
Background
In recent years, with the gradual development of intelligent vehicles, the number of traffic accidents caused by automobiles is increasing, and the most serious traffic accidents caused by people and vehicles become the key point of social attention. The identification of the pedestrian violation-crossing behavior has important significance for ensuring the driving safety of the vehicle and reducing the traffic risk of the driver and the road pedestrian. Therefore, the identification of the intention of the pedestrian violation trafficking behavior and the identification of the violation trafficking behavior are necessary. The identification of the pedestrian violation-violating traffic behavior needs to comprehensively consider the coupling relationship among the factors of people, vehicles and roads, so as to identify the pedestrian traffic intention and provide support for driving decision.
In the prior art, Zhou et al propose a method for predicting the pedestrian violation trafficking behavior based on a planned behavior theory. According to the method, variables such as behavior concept, subjective regulation, observation behavior control, objective regulation, scene risk degree and the like of the pedestrian are obtained through a questionnaire investigation means, and the violation of the pedestrian is explained by a plan behavior theory. However, the questionnaire survey is too ideal, the actual vehicle cannot obtain relevant information, and the interaction between the actual scene person and the vehicle and the road is not considered, so that the questionnaire survey cannot be used in the actual scene and can only be used in the field of accident analysis. Dommes et al try to find the subjective and objective reasons that pedestrians pass violating the regulations by analyzing the factors that influence the decision of pedestrians running red light. The research extracts 13 behavior indexes (12 indexes before and during crossing the road and running red light), researches the effects of relevant variables of demographics, environment and actions, and has guiding significance for the research of the pedestrian violation-breaking passing decision process. However, the research is stopped in the analysis of the influence factors, the inherent reasons and the connection are not mined, the specific scene is emphasized, and a certain deviation exists between the specific scene and the actual scene of the pedestrian violation-breaking passing decision, so that the actual behavior of the pedestrian is difficult to identify, and the method is not suitable for actual application. Therefore, it is necessary to develop a method for identifying the pedestrian violation passing behavior based on the planning behavior theory.
Disclosure of Invention
The invention aims to provide a method for identifying the violation of regulations of a pedestrian based on a planned behavior theory, which can accurately identify the violation of regulations of the pedestrian in real time and effectively identify the violation of regulations of the pedestrian so as to support driving decision.
In order to achieve the aim, the invention provides a method for identifying the illegal pedestrian crossing behavior based on a planned behavior theory, which comprises the following steps:
step 1, collecting related perception information in a current traffic scene;
step 2, identifying the degree of the pedestrian according to the individual characteristics of the pedestrian in the relevant perception information;
step 3, identifying the crowd effect of the pedestrian according to the pedestrian population characteristics in the relevant perception information;
step 4, acquiring main influence factors of the pedestrian violation trafficking behaviors according to historical pedestrian violation trafficking data similar to or identical to the current traffic scene;
and 5, fusing the results obtained in the step 2, the step 3 and the step 4 through a planned behavior theory to obtain the violation passing intention of the pedestrian, further identifying the violation passing behavior of the pedestrian, and outputting an identification result.
Further, the pedestrian aggressiveness in step 2 can be specifically expressed by formula (1):
Anorm＝S(Av+Aa+Aw+Ab+Ad+Ao) (1)
in the formula (1), S represents a sigmoid function, AvIndicates the degree of acceleration due to average velocity, AaIndicates the degree of acceleration due to maximum acceleration, AwIndicating the degree of aggressiveness due to latency, AbIndicates the degree of acceleration due to backward movement, AdIndicates the degree of invasiveness caused by the interfering substance, AoIndicating the degree of aggressiveness resulting from observing traffic.
Further, Av、Aa、Aw、Ab、AdAnd AoThe variable is a continuous variable with a value in the range of 0-1, or a discrete variable with a value of 0 or 1.
Further, in the step 3, the acquisition manner of the crowd effect CE of the identified pedestrian is represented by formula (2):
Figure BDA0002635842130000021
in the formula (2), the pedestrian dependent effect CE is defined as the proportion of the pedestrian violating the regulations
Figure BDA0002635842130000022
And a binary function of the number of people N;
the dominant effect CE is obtained by normalizing the dominant effect CE by the following formula (9)norm：
CEnorm＝S(CE) (9)
In the formula (9), S is a sigmoid function and converts the conotoxic effect C into a standard variable between 0 and 1.
Further, the step 4 specifically includes:
step 4.1, using a data classification method to divide historical pedestrian data of the same or similar traffic scene as the current traffic scene into non-violation-passing pedestrians and violation-passing pedestrians, selecting a traffic scene which is possible to violate the traffic scene, and dividing a training set and a testing set according to a preset proportion to obtain main scene influence factors of the differential behavior of the two classes of people;
step 4.2, after the n-dimensional main scene influence factors of the group difference behaviors of the pedestrian who does not pass the violation and the pedestrian who passes the violation are obtained through the step 4.1, further, a main component analysis method is used to obtain main components of the group difference behavior scene influence factors, wherein the main components are expressed as a formula (10):
Figure BDA0002635842130000031
in the formula (10), YiRepresents the ith principal component, CiRepresenting the ith scene influencing factor; p denotes the principal component dimension, n denotes the dimension of the scene influencing factor, A is a principal component matrix of dimension p x n, the matrix elements of which are a11To apnThe final scene influencing factor principal component is expressed by equation (11):
Y＝[Y1，Y2，...，Yp]T (11)。
further, in step 4.1, acquiring main scene influence factors of the differential behaviors of the two types of people through a variance analysis method, defining the pedestrian class which does not pass the violation as class A, defining the pedestrian class which passes the violation as class B, and defining the scene influence factors as factor C;
case one, the factor C contains a significant variance difference in A, B people, and in the case of a confidence level of [0.95,1.00], it assumes that the test value P-value is [0,0.01], and it is considered to be one of the main scene influencing factors of the differentiated behavior of the two people;
in case II, the factor C does not contain a significant variance difference in A, B people, and in case that the confidence level is [0.95,1.00], the factor C assumes that the test value P-value is [0.1,1.0], and then the factor C is considered not to be one of the main scene influence factors of the difference behaviors of the two people;
case three, the factor C contains significant variance difference in the population of class a and B, but in the case of confidence level at [0.95,1.00], which assumes the test value P-value at (0.01,0.1), it is necessary to divide the training set and the test set multiple times to test the significance of the factor.
Further, the step 5 specifically includes:
step 5.1, obtaining the pedestrian excitation degree A according to the step 2, the step 3 and the step 4normFrom the dominant effect CEnormAnd obtaining the violation passing intention I of the pedestrian through a planning behavior theory by using the main component Y of the scene influence factor, expressing the violation passing intention I as formulas (12) to (14), and further solving unknown parameters { a, b, c and d }:
Io＝a+bAnorm_0+cCEnorm_0+dY0 (12)
It＝αIt-1+(1-α)(a+bAnorm_t+cCEnorm_t+dYt) (13)
It∈[0，1] (14)
in the formula IoInitial value representing pedestrian's intention to pass violation, Anorm_0、CEnorm_0、Y0Respectively representing the value of the main components of the radical degree, the slave effect and the scene influence factor of the pedestrian at the initial moment, ItRepresenting the illegal passing intention of the pedestrian at the time t, alpha is a preset empirical value, It0 represents that the pedestrian has no illegal passing intention at all, It1 represents the pedestrian's confident intention to pass through the traffic violation, ItThe violation passing intention is uncertain and is partially in line with the probability distribution of the violation passing of the pedestrian;
step 5.2, obtaining the pedestrian violation passing intention I according to the step 5.1tAnd scene influencing factor principal component YtAccording to the plan behavior theory, ACT is used for the illegal walk-through behavior of the pedestriantRepresented by formula (18):
ACTt＝Rule(It，Yt) (18)
wherein, Rule (I)t，Yt) Representing predefined traversal rules.
Due to the adoption of the technical scheme, the invention has the following advantages: 1. the method for identifying the pedestrian violation trafficking behavior comprehensively considering the coupling relation among the human, the vehicle and the road is established, the influence of the aggressiveness degree of the pedestrian and the subordinate effect is analyzed, the principal component of the influence factor of the human-vehicle road interaction environment on the pedestrian behavior decision is considered, and the pedestrian violation trafficking process can be objectively and really described. 2. According to the invention, through the incentive degree identification module, the method for identifying the illegal passing behaviors supporting different types of pedestrians is generated, and a foundation is laid for realizing differential decision of vehicles. 3. The invention is based on crowd effect identification, reflects the supervision effect brought by crowd gathering and the shelter effect brought by the excessive proportion of the violation pedestrians, and provides a feasible theoretical direction for researching the effect of the crowd on individuals. 4. The invention analyzes the main components of the traffic environment influence factors of the pedestrian who does not pass through the traffic violation and the pedestrian who passes through the traffic violation based on the real historical data of the traffic environment, reflects the objective influence of the pedestrian under the human-vehicle-road coupling environment, combines the influence of the intrinsic factors of the pedestrian based on the plan behavior theory, and realizes the separation and the fusion of the internal and external influence factors of the pedestrian passing through the violation. Compared with the existing identification method for the pedestrian violation-violating traffic behaviors, the method can more comprehensively reflect the decision mechanism of the pedestrian violation-violating traffic process, and also proves the effectiveness, feasibility and scientificity of the method.
Drawings
FIG. 1 is a schematic view of a violation identification process provided by an embodiment of the present invention;
FIG. 2 is a flow chart of information provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram of an algorithm result of a clustering algorithm for pre-classifying a population into two categories according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an application scenario provided by an embodiment of the present invention;
FIG. 5 is a schematic diagram of information fusion and violation behavior identification based on the planned behavior theory according to an embodiment of the present invention;
fig. 6 is a schematic diagram of information presentation of a driving assistance screen for a human-driven vehicle according to an embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings.
For example, when the intelligent vehicle enters an intersection with a traffic light and a zebra crossing from one side of a road, the pedestrian indicating light is in red, and the pedestrian is located at a curb. The vehicle needs to judge whether the pedestrian will generate the illegal behavior to perform the deceleration avoidance behavior, and the specific scene is shown in fig. 4.
As shown in fig. 1 and fig. 2, the method for identifying the pedestrian violation trafficking behavior based on the planning behavior theory provided by the invention comprises the following steps:
step 1, collecting related perception information in a current traffic scene. The "relevant perception information" is acquired by a vehicle-mounted perception system through sensors such as a vehicle-mounted camera and a radar, and the information can be understood as data which is relevant to time and is continuously updated.
And 2, identifying the pedestrian motivation degree according to the individual characteristics of the pedestrians in the relevant perception information. The "individual pedestrian characteristic" includes six items of information, i.e., the average speed, the maximum acceleration, the waiting time, whether to retreat, whether to have an interfering object, and whether to observe the traffic flow of the pedestrian. The average speed information and the acceleration information can be directly obtained through a laser radar, the waiting time information can be obtained through calculation of a timer, whether retreating information is obtained through calculation of pedestrian position information, whether interference object information is obtained through a camera, and whether traffic flow information is observed and the head orientation of a pedestrian is obtained through the camera. Each item of information is for each pedestrian, this item being an "individual" characteristic of the pedestrian. Specifically, for example: and averaging speed, so as to avoid error in data acquisition of the frame, and therefore, errors can be avoided by averaging the data of the current frame and the data of the previous five-six frames.
And 3, identifying the crowd effect of the pedestrian according to the pedestrian population characteristics in the relevant perception information. Wherein, the pedestrian population characteristics comprise the number of the population and the violation proportion of the population. The "from-crowd effect" may be a binary function of the number of people and the proportion of people violations. The "number of people" is understood to mean the number of pedestrians comprised by the group of people.
And 4, acquiring main influence factors of the violation traffics of the pedestrians according to historical violation traffics similar to or identical to the current traffic scene. The scene influence factor refers to the principal component of the pedestrian violation passing factor influenced by the scene through variance analysis and principal component analysis means. The historical violation data comprises a large number of public data sets, videos obtained by monitoring cameras, data obtained by aerial cameras and the like.
And 5, fusing the results obtained in the step 2, the step 3 and the step 4 through a planning behavior theory to obtain the violation passing intention of the pedestrian, further identifying the violation passing behavior of the pedestrian, and providing corresponding information on a driving auxiliary screen aiming at the manually driven vehicle.
The pedestrian violation behavior identification method comprehensively considers the intrinsic characteristics of pedestrians, the characteristics of crowds and environmental influence factors in the pedestrian violation behavior passing process, can be suitable for accurate identification of pedestrian violation behavior in actual traffic scenes, and provides support for intelligent decision making of vehicles.
As a preferred implementation of identifying the degree of pedestrian aggression in step 2, it can be represented by formula (1):
Anorm＝S(Av+Aa+Aw+Ab+Ad+Ao) (1)
in the formula (1), S represents a sigmoid function, AvIndicates the degree of acceleration due to average velocity, AaIndicates the degree of acceleration due to maximum acceleration, AwIndicating the degree of aggressiveness due to latency, AbIndicates the degree of acceleration due to backward movement, AdIndicates the degree of invasiveness caused by the interfering substance, AoIndicating the degree of aggressiveness resulting from observing traffic. A. thev、Aa、Aw、Ab、AdAnd AoThe value of (2) can be a continuous variable within the range of 0-1, and can also be a discrete variable, such as: 0 or 1. The specific value-taking methods of A with different subscripts comprise the following steps:
first, define Av、Aa、Aw、Ab、AdAnd AoWhether it is a discrete variable or a continuous variable; then, the degree of the pedestrian is excited according to the historical real data of the pedestrian and Av、Aa、Aw、Ab、AdAnd AoThe value of (A) is subjected to deep learning training to obtain Av、Aa、Aw、Ab、AdAnd AoThe true value of (d). The pedestrian historical real data comprises the historical real data of individual characteristics of pedestrians and true values of the historical pedestrian motivation rate (obtained by questionnaire survey or calculation of the danger degree of the pedestrians during passing).
The sigmoid function in the present embodiment may also be replaced with a selu function, a tanh function, or the like, in order to map the defined aggressiveness between (0,1), thereby eliminating the difference in input.
Of course, the parameters in equation (1) may also be expressed using other factors, such as: the motivation rate by age, the motivation rate by gender, and/or the motivation rate by interferents, etc. However, the parameters selected in equation (1) can be obtained by vehicle sensing and are easily applied to actual scenes. And can be proved by literature investigation: the influence of the selected parameters in the sigmoid function on the degree of pedestrian excitation is significant.
In one embodiment, the acquisition manner of the dominant effect CE identified in step 3 is represented by formula (2):
Figure BDA0002635842130000061
in the formula (2), the pedestrian dependent effect CE is defined as the proportion of the pedestrian violating the regulations
Figure BDA0002635842130000062
And a binary function of the number of people N. According to the proportion of the pedestrians violating the regulations with the increase of the number N of the crowd
Figure BDA0002635842130000063
The people group have the effect of monitoring each other or contracting each other. The "supervised/sheltered effect" is actually the comfort effect, that is, a statement of "crowd effect", refers to the tendency of the behavior of a single pedestrian to change with the behavior of other people, and currently, the crowd effect tendency of the pedestrian is generally obtained through questionnaire, but the embodiment intends to adopt the crowd quantity and the crowd violation proportion fitting, and uses the crowd quantity and the crowd violation proportion fitting to useAnd training the corresponding binary function by using the real data. The "supervised/shrouded effect" varies according to the proportion of people violations.
Specifically, the pedestrian dependent effect CE defines the number of people N and the proportion of pedestrians violating the regulations
Figure BDA0002635842130000078
Can be expressed as the following formulas (3), (4) and (5):
Figure BDA0002635842130000071
Figure BDA0002635842130000072
Figure BDA0002635842130000073
where a, b, c, d, e, f are all predefined positive coefficients, such as: a is 0.04, b is 1.1, d is 1, c is 2
Figure BDA0002635842130000074
There is no satellite effect. Of course, the specific values of a, b, c, d, e, f may be predefined as needed.
It can thus be seen that: the number N of the crowds takes 4 as a demarcation point, and within the range that N is more than or equal to 1 and less than or equal to 4, the supervision effect is enhanced along with the increase of the number N of the crowds, and the value of the crowd effect is increased; at N > 4, the value of the secondary effect no longer changes. Pedestrian proportion against traffic regulation
Figure BDA0002635842130000079
With 1/2 as a boundary point
Figure BDA00026358421300000710
In the range of (1), the subordinate effect is destroyed by someone violation, and the value of the subordinate effect is reduced; in that
Figure BDA00026358421300000711
In this case, the side effect that destroys the rule is formed, and the value of the side effect rises on the contrary.
Specifically, the pedestrian dependent effect CE defines the number of people N and the proportion of pedestrians violating the regulations
Figure BDA00026358421300000712
Can be expressed as the following formulas (6) and (7):
Figure BDA0002635842130000075
Figure BDA0002635842130000076
Figure BDA0002635842130000077
in the formula, abs (, x) is an absolute value function, and the trend is the same.
In this embodiment, sigmoid activation function is used, and the following formula (9) is used to normalize the dominant effect CE to obtain the dominant effect CEnorm：
CEnorm＝S(CE) (9)
In the formula (9), S is a sigmoid function, and the dominant effect CE is converted into a standard variable between 0 and 1 for subsequent processing.
The sigmoid function in this embodiment may also be replaced with a selu function or a tanh function. The parameters selected in the formula (8) can be obtained through vehicle perception, and are easily applied to actual scenes. The following are proved by literature research: the parameters selected in equations (6) - (8) can find evidence and corresponding data that significantly affect the crowdsourcing effect.
By standardizing the subordinate effect through an equation (9), the error brought by the input scale to the final neural network training for obtaining the violation behavior can be eliminated. Without normalization, if some parameters are too large (e.g., conus effect) and some parameters are too small (e.g., aggressiveness), the effect of aggressiveness may be directly ignored (too small).
In one embodiment, the above-mentioned crowd-sourcing CE may also be obtained by using a questionnaire, and the participants may score the questions to show their own tendency. That is, the secondary effects can also be expressed as a function of the score: ce (i), i represents the ith pedestrian. However, the method only can serve as a priori knowledge and is difficult to be used in practice, because pedestrians encountered on actual roads cannot be scored.
The public effect CE may also be defined using rules, such as: ce (i) ce (j) f (n), i denotes the i-th and j-th pedestrians among a group of pedestrians. Namely: when the ith pedestrian is the leader and breaks rules and regulations, the follower breaks rules and regulations similarly when the follower jth pedestrian effect CE is greater than the threshold value. The leading person and the following person are difficult to confirm, the threshold value is difficult to find, and the defined rule is difficult to achieve perfection.
In one embodiment, step 4 specifically includes:
step 4.1, as shown in fig. 3, using a data classification method, dividing the historical pedestrian data shown in fig. 4, which is the same as or similar to the current traffic scene, into: and defining the pedestrian which does not pass the violation as class A, and defining the pedestrian which passes the violation as class B. Only selecting traffic scenes which are possible to pass violations, dividing a training set and a testing set according to a preset proportion (such as a proportion of 3: 1), and obtaining main scene influence factors of the differential behaviors of the two classes of people through a variance analysis method. The data classification method may be a clustering algorithm, or a support vector machine, a synovial algorithm, or the like.
Case one, the C factor contains a significant variance difference in the A, B population, and in the case of a confidence level at [0.95,1.00], it assumes that the test value P-value is at [0,0.01], and is considered to be one of the main scene influencing factors of the differentiated behavior of the two populations.
Case two, factor C does not contain a significant variance difference in the A, B population, and in the case of a confidence level at [0.95,1.00], it assumes that the test value P-value is at [0.1,1.0], and it is considered not to be one of the main scenario influencing factors of the two population difference behavior.
Case three, the factor C contains significant variance difference in the population of class a and B, but in the case of confidence level at [0.95,1.00], which assumes the test value P-value at (0.01,0.1), it is necessary to divide the training set and the test set multiple times to test the significance of the factor.
Step 4.2, after learning the n-dimensional main scene influence factors of the A, B type crowd difference behaviors through step 4.1, further, obtaining the principal components of the crowd difference behavior scene influence factors by using a principal component analysis method, wherein the principal components are expressed as formula (10):
Figure BDA0002635842130000091
in the formula (10), YiRepresents the ith principal component, CiRepresenting the ith scene influencing factor; p denotes the principal component dimension, n denotes the dimension of the scene influencing factor, A is a principal component matrix of dimension p x n, the matrix elements of which are a11To apn. The final scene influencing factor principal component is expressed as formula (11):
Y＝[Y1，Y2，...，Yp]T (11)
the analysis of variance method in step 4.1 can also be replaced by clustering, t-test, or z-test methods.
The principal component analysis method in step 4.2 may be replaced by a factor analysis method or the like.
In one embodiment, step 5 specifically includes:
step 5.1, obtaining the pedestrian excitation degree A according to the step 2, the step 3 and the step 4normFrom the dominant effect CEnormThe main component Y of the scene influence factor is converted into three variables of behavior concept, subjective norm and observed behavior control, as shown in figure 5, the violation passing intention I of the pedestrian is obtained through a planning behavior theory and is expressed as formulas (12) to (14), and thenAnd solving for unknown parameters { a, b, c, d }:
Io＝a+bAnorm_0+cCEnorm_0+dY0 (12)
It＝αIt-1+(1-α)(a+bAnorm_t+cCEnorm_t+dYt) (13)
It∈[0，1] (14)
in the formula (12), IoThe initial value of the violation passing intention of the pedestrian is obtained by calculation according to the pedestrian excitation degree, the follower effect and the scene main component of the initial time frame, or an arbitrary value of 0-1 can be directly set, so that the violation passing intention of the pedestrian is iterated along with the increase of time to approach to a true value quickly; a. thenorm_0、CEnorm_0、Y0The values of the pedestrian aggressiveness degree, the follower effect and the scene influence factor principal component at the initial time are respectively represented, wherein the initial time is the time when the vehicle starts to observe the pedestrian, and the end time is the time when the sensor cannot sense the pedestrian. A. thenorm_0、CEnorm_0、Y0The specific numerical value is obtained by calculating according to the pedestrian characteristic, the group characteristic and the scene information sensed by the current frame sensor through the formula and the model provided previously.
In the formula (13), ItRepresenting the violation passing intention of the pedestrian at the moment t, and receiving the violation passing intention I at the previous momentt-1And main components A of pedestrian acceleration degree, audience effect and scene influence factor at the moment tnorm_t、CEnorm_t、YtThe influence ratio of (a) to (1-a); α is a preset empirical value, such as 0.5.
In the formula (14), It0 represents that the pedestrian has no illegal passing intention at all, It1 represents the pedestrian's confident intention to pass through the traffic violation, ItAnd (0,1) represents uncertain violation passing intentions and partially conforms to the probability distribution of violation passing of the pedestrian.
Or obtaining the violation passing intention I of the pedestrian through the primary quadratic terms expressed by the formulas (15) to (17), and further solving the unknown parameter { a }1，b1，b2，c1，c2，d1，d2}：
Figure BDA0002635842130000101
Figure BDA0002635842130000102
It∈[0，1] (17)
In the same way, the violation passing intention I of the pedestrian can be obtained through the items of three times and four times.
Step 5.2, obtaining the pedestrian violation passing intention I according to the step 5.1tAnd scene influencing factor principal component YtAccording to the plan behavior theory, ACT is used to determine the violation of the pedestrian, as also shown in fig. 5tRepresented by formula (18):
ACTt＝Rule(It，Yt) (18)
wherein, Rule (I)t，Yt) Representing predefined traversal rules. Such as:
Figure BDA0002635842130000103
and if and only if the violation passing intention of the pedestrian is more than half and the dangerous component of the main component of the scene influence factor is less than half, the pedestrian is considered to pass the violation, otherwise, the pedestrian is considered not to pass the violation.
Of course, ACT of the pedestrian's ACT of passing through violationstThe table can be easily judged according to actual conditions, but the accuracy is difficult to reach more than 80%. Can also be influenced by intention ItAnd scene factor YtAnd then training by using a neural network method, wherein similar methods comprise a Support Vector Machine (SVM), a k-means classification method, a neural network and the like.
In one embodiment, as shown in fig. 6, for a manually driven vehicle, the current recommended vehicle speed, the signal lamp state, the pedestrian violation probability and the recommended decision information need to be displayed on a driving assistance screen to assist the driver in making a reasonable decision, so as to avoid a collision accident caused by the pedestrian violation behavior.
Finally, it should be pointed out that: the above examples are only for illustrating the technical solutions of the present invention, and are not limited thereto. Those of ordinary skill in the art will understand that: modifications can be made to the technical solutions described in the foregoing embodiments, or some technical features may be equivalently replaced; such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (7)

1. A pedestrian violation passing behavior identification method based on a planned behavior theory is characterized by comprising the following steps:
step 1, collecting related perception information in a current traffic scene;
step 2, identifying the degree of the pedestrian according to the individual characteristics of the pedestrian in the relevant perception information;
step 3, identifying the crowd effect of the pedestrian according to the pedestrian population characteristics in the relevant perception information;
step 4, acquiring main influence factors of the pedestrian violation trafficking behaviors according to historical pedestrian violation trafficking data similar to or identical to the current traffic scene;
and 5, fusing the results obtained in the step 2, the step 3 and the step 4 through a planned behavior theory to obtain the violation passing intention of the pedestrian, further identifying the violation passing behavior of the pedestrian, and outputting an identification result.
2. The method for identifying the pedestrian violation passing behavior based on the planning behavior theory as claimed in claim 1, wherein the degree of aggressiveness of the pedestrian in the step 2 can be specifically expressed as formula (1):
Anorm＝S(Av+Aa+Aw+Ab+Ad+Ao) (1)
in the formula (1), S represents sigmoidFunction, AvIndicates the degree of acceleration due to average velocity, AaIndicates the degree of acceleration due to maximum acceleration, AwIndicating the degree of aggressiveness due to latency, AbIndicates the degree of acceleration due to backward movement, AdIndicates the degree of invasiveness caused by the interfering substance, AoIndicating the degree of aggressiveness resulting from observing traffic.
3. The method for identifying pedestrian violation traversal behavior based on planning behavior theory as claimed in claim 2, wherein A isv、Aa、Aw、Ab、AdAnd AoThe variable is a continuous variable with a value in the range of 0-1, or a discrete variable with a value of 0 or 1.
4. The method for identifying the pedestrian violation passing behavior based on the planning behavior theory as claimed in claim 1, wherein in the step 3, the acquisition mode of the identified pedestrian subordinate effect CE is represented by formula (2):
Figure FDA0002635842120000011
in the formula (2), the pedestrian dependent effect CE is defined as the proportion of the pedestrian violating the regulations
Figure FDA0002635842120000012
And a binary function of the number of people N;
the dominant effect CE is obtained by normalizing the dominant effect CE by the following formula (9)norm：
CEnorm＝S(CE) (9)
In the formula (9), S is a sigmoid function and converts the conotoxic effect C into a standard variable between 0 and 1.
5. The pedestrian violation passing behavior identification method based on the planning behavior theory as claimed in claim 1, wherein the step 4 specifically comprises:
step 4.1, using a data classification method to divide historical pedestrian data of the same or similar traffic scene as the current traffic scene into non-violation-passing pedestrians and violation-passing pedestrians, selecting a traffic scene which is possible to violate the traffic scene, and dividing a training set and a testing set according to a preset proportion to obtain main scene influence factors of the differential behavior of the two classes of people;
step 4.2, after the n-dimensional main scene influence factors of the group difference behaviors of the pedestrian who does not pass the violation and the pedestrian who passes the violation are obtained through the step 4.1, further, a main component analysis method is used to obtain main components of the group difference behavior scene influence factors, wherein the main components are expressed as a formula (10):
Figure FDA0002635842120000021
in the formula (10), YiRepresents the ith principal component, CiRepresenting the ith scene influencing factor; p denotes the principal component dimension, n denotes the dimension of the scene influencing factor, A is a principal component matrix of dimension p x n, the matrix elements of which are a11To apnThe final scene influencing factor principal component is expressed by equation (11):
Y＝[Y1,Y2,…,Yp]T (11)。
6. the method for identifying the pedestrian violation trafficking behavior based on the planning behavior theory as claimed in claim 5, wherein in step 4.1, the main scene influencing factors of the differential behaviors of the two classes of people are obtained through a variance analysis method, the pedestrian class which does not violate the regulations is defined as class A, the pedestrian class which does violate the regulations is defined as class B, and the scene influencing factors are defined as factor C;
case one, the factor C contains a significant variance difference in A, B people, and in the case of a confidence level of [0.95,1.00], it assumes that the test value P-value is [0,0.01], and it is considered to be one of the main scene influencing factors of the differentiated behavior of the two people;
in case II, the factor C does not contain a significant variance difference in A, B people, and in case that the confidence level is [0.95,1.00], the factor C assumes that the test value P-value is [0.1,1.0], and then the factor C is considered not to be one of the main scene influence factors of the difference behaviors of the two people;
case three, the factor C contains significant variance difference in the population of class a and B, but in the case of confidence level at [0.95,1.00], which assumes the test value P-value at (0.01,0.1), it is necessary to divide the training set and the test set multiple times to test the significance of the factor.
7. The pedestrian violation passing behavior identification method based on the planning behavior theory as claimed in claim 1, wherein the step 5 specifically comprises:
step 5.1, obtaining the pedestrian excitation degree A according to the step 2, the step 3 and the step 4normFrom the dominant effect CEnormAnd obtaining the violation passing intention I of the pedestrian through a planning behavior theory by using the main component Y of the scene influence factor, expressing the violation passing intention I as formulas (12) to (14), and further solving unknown parameters { a, b, c and d }:
Io＝a+bAnorm_0+cCEnorm_0+dY0 (12)
It＝αIt-1+(1-α)(a+bAnorm_t+cCEnorm_t+dYt) (13)
It∈[0,1] (14)
in the formula IoInitial value representing pedestrian's intention to pass violation, Anorm_0、CEnorm_0、Y0Respectively representing the value of the main components of the radical degree, the slave effect and the scene influence factor of the pedestrian at the initial moment, ItRepresenting the illegal passing intention of the pedestrian at the time t, alpha is a preset empirical value, It0 represents that the pedestrian has no illegal passing intention at all, It1 represents the pedestrian's confident intention to pass through the traffic violation, ItThe violation passing intention is uncertain and is partially in line with the probability distribution of the violation passing of the pedestrian;
step 5.2, according to step 5.1 the pedestrian passing violation intention ItAnd scene influencing factor principal component YtAccording to the plan behavior theory, ACT is used for the illegal walk-through behavior of the pedestriantRepresented by formula (18):
ACTt＝Rule(It，Yt) (18)
wherein, Rule (I)t，Yt) Representing predefined traversal rules.
CN202010824900.3A 2020-08-17 2020-08-17 Method for identifying pedestrian illegal-passing behavior based on planned behavior theory Active CN111985386B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010824900.3A CN111985386B (en) 2020-08-17 2020-08-17 Method for identifying pedestrian illegal-passing behavior based on planned behavior theory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010824900.3A CN111985386B (en) 2020-08-17 2020-08-17 Method for identifying pedestrian illegal-passing behavior based on planned behavior theory

Publications (2)

Publication Number Publication Date
CN111985386A true CN111985386A (en) 2020-11-24
CN111985386B CN111985386B (en) 2022-06-14

Family

ID=73435113

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010824900.3A Active CN111985386B (en) 2020-08-17 2020-08-17 Method for identifying pedestrian illegal-passing behavior based on planned behavior theory

Country Status (1)

Country Link
CN (1) CN111985386B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766538A (en) * 2020-12-26 2021-05-07 浙江天行健智能科技有限公司 Decision modeling method for pedestrian crossing road based on SVM algorithm

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016022108A1 (en) * 2014-08-06 2016-02-11 Robinson Kurt B Systems and methods involving features of adaptive and/or autonomous traffic control
CN107305626A (en) * 2016-04-25 2017-10-31 孙久耕 Pedestrian movement's track detection and analysis system violating the regulations
US20180174446A1 (en) * 2015-02-09 2018-06-21 Kevin Sunlin Wang System and method for traffic violation avoidance

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016022108A1 (en) * 2014-08-06 2016-02-11 Robinson Kurt B Systems and methods involving features of adaptive and/or autonomous traffic control
CN107077782A (en) * 2014-08-06 2017-08-18 李宗志 Adaptive and/or autonomous traffic control system and method
US20180174446A1 (en) * 2015-02-09 2018-06-21 Kevin Sunlin Wang System and method for traffic violation avoidance
CN107305626A (en) * 2016-04-25 2017-10-31 孙久耕 Pedestrian movement's track detection and analysis system violating the regulations

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Y. LI等: "Pedestrian Trajectory Prediction at Un-signalized intersection using probabilistic reasoning and sequence learning", 《2019 IEEE INTELLIGENT TRANSPORTATION SYSTEMS CONFERENCE (ITSC)》 *
辛秀颖: "基于行为的无信号灯控制下人车混行交通流研究", 《中国博士学位论文全文数据库（电子期刊） 工程科技II辑》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766538A (en) * 2020-12-26 2021-05-07 浙江天行健智能科技有限公司 Decision modeling method for pedestrian crossing road based on SVM algorithm

Also Published As

Publication number Publication date
CN111985386B (en) 2022-06-14

Similar Documents

Publication Publication Date Title
US10885777B2 (en) Multiple exposure event determination
CN112700470B (en) Target detection and track extraction method based on traffic video stream
CN103871242B (en) A kind of driving behavior overall evaluation system and method
CN112071061B (en) Vehicle service system based on cloud computing and data analysis
CN108944939B (en) Method and system for providing driving directions
Zhang et al. Pedestrian safety analysis in mixed traffic conditions using video data
CN111094095B (en) Method and device for automatically sensing driving signal and vehicle
CN112116263B (en) Traffic intersection risk level assessment method and device, electronic equipment and storage medium
Bouhsissin et al. Driver behavior classification: a systematic literature review
CN115035491A (en) Driving behavior road condition early warning method based on federal learning
CN113658427A (en) Road condition monitoring method, system and equipment based on vision and radar
Wang et al. Modeling aggressive driving behavior based on graph construction
CN111985386B (en) Method for identifying pedestrian illegal-passing behavior based on planned behavior theory
CN112559968A (en) Driving style representation learning method based on multi-situation data
Gao et al. An ADAS with better driver satisfaction under rear-end near-crash scenarios: A spatio-temporal graph transformer-based prediction framework of evasive behavior and collision risk
Bäumler et al. Report on validation of the stochastic traffic simulation (Part B)
Feraud et al. Are you a good driver? A data-driven approach to estimate driving style
Kamrani et al. Drivers’ reactions to connected vehicle forward collision warnings: Leveraging real-world data from the THEA CV pilot
CN114026574A (en) Running a trainable module while monitoring for departure from a trained application scope
Roohullah et al. Accident detection in autonomous vehicles using modified restricted Boltzmann machine
Jain et al. A computational model for driver risk evaluation and crash prediction using contextual data from on-board telematics
Wang et al. Driver modeling based on vehicular sensing data
Abdelwahab et al. Investigating driver injury severity in traffic accidents using fuzzy ARTMAP
CN117746632B (en) Bus driving risk assessment and early warning system and method based on intelligent network
Chen et al. The framework of vehicle emotion system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant