External human-machine interface position determination method based on pedestrian visual search features

Abstract

The present invention discloses a method for determining the position of an external human-machine interface based on pedestrian visual search characteristics, which specifically relates to the field of traffic safety technology, including: dividing a plurality of sight-focusing areas according to the vehicle shape, conducting a pedestrian visual search test and collecting test data; performing a difference analysis on the data to determine the factors that have a significant impact on the tester's gaze behavior; using a Markov chain to analyze the visual transfer characteristics of pedestrians under different influencing factors, and quantifying the probability of pedestrians transferring between different sight-focusing areas under different influencing factors; by quantifying the impact of factors on visual search, providing strong support for the arrangement of the vehicle's external human-machine interface position, and the arranged external human-machine interface position can be quickly noticed by pedestrians.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic safety, and in particular to a method for determining the position of an external human-machine interface based on pedestrian visual search features. Background Art

[0002] With the rapid development of autonomous driving technology, self-driving vehicles are gradually being integrated into mixed traffic environments. However, traditional driver-pedestrian interactions, such as eye contact and gestures, are becoming less effective in autonomous driving scenarios. Pedestrians rely on communication with drivers to determine whether it is safe to cross the road. When drivers are absent or engaged in activities unrelated to driving, pedestrians become uneasy and hesitant, which not only affects traffic efficiency but also may reduce public acceptance of self-driving vehicles.

[0003] To address this issue, external human-machine interfaces (HMIs) have been proposed as an alternative communication method to enhance interaction between autonomous vehicles and other traffic participants, particularly pedestrians. These interfaces complement traditional interaction methods by displaying relevant information, such as parking intentions or vehicle status, on the vehicle's exterior, helping pedestrians understand the vehicle's intentions. Research has shown that the use of external HMIs can significantly improve pedestrian safety and crossing efficiency, particularly in traffic scenarios with undefined right-of-way.

[0004] The optimal placement of external human-machine interfaces (HMIs) remains controversial. Previous approaches, often based on vehicle design, have proposed fixed locations for HMIs, such as the windshield, grille, and roof. However, pedestrians crossing the road are affected by numerous factors, such as distance between the pedestrian and the vehicle, vehicle color, and type. Fixed HMI placement can easily lead to pedestrians missing information.

[0005] Based on this, the present invention aims to provide a method for determining the position of an external human-machine interface that is more in line with the visual search characteristics of pedestrians, so that pedestrians can easily notice the external human-machine interface at first glance. Summary of the Invention

[0006] In order to make up for the shortcomings of the existing technical problems, the present invention provides a method for determining the position of an external human-machine interface based on the visual search characteristics of pedestrians. By collecting the visual search data of pedestrians when crossing the street, the data is differentially analyzed to obtain the factors that have a significant impact on the tester's gaze behavior. For different influencing factors, the Markov chain is used to quantify the probability of a person's transition between different visual focus points under different influencing factors. The external human-machine interface arranged based on this is more in line with the visual search characteristics of pedestrians, enabling pedestrians to focus on the external human-machine interface more quickly.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for determining the position of an external human-machine interface based on pedestrian visual search features comprises the following steps:

[0009] (1) Based on the vehicle's appearance, multiple visual focus areas were divided and pedestrian visual search tests were conducted. The test subjects' gaze time and number of gazes on different visual focus areas of vehicles of different colors and types at different longitudinal distances were collected;

[0010] (2) Conduct variance analysis on the data to identify factors that significantly influence the test subjects’ gaze behavior;

[0011] (3) Use Markov chains to analyze the visual transfer characteristics of pedestrians under different influencing factors, and quantify the probability of pedestrians' gaze stability at different visual focus points and the probability of transfer between different visual focus points under different influencing factors;

[0012] (4) Based on the results obtained in step (3), determine the location of the vehicle's external human-machine interface.

[0013] In the present invention, the sight focusing parts include the windshield, the hood, the grille and the bumper.

[0014] In the present invention, the pedestrian visual search test includes:

[0015] Step 1: The test scene is set at a two-way four-lane pedestrian crossing that is not controlled by traffic lights. The tester wears an eye tracker to record eye movements.

[0016] Step 2: Have multiple testers observe the traffic conditions and begin crossing the street when they deem it safe. If the conditions are unsafe, the testers must stop and wait for the opportunity to cross.

[0017] Step 3: The tester interacts with the motor vehicle and crosses the crosswalk before the vehicle approaches, completing the crossing behavior, which is considered a valid crossing. The test ends after completing multiple valid crossings.

[0018] In the present invention, the non-safe conditions include the following three conditions:

[0019] 1) When there are no motor vehicles on the road in the direction of oncoming traffic;

[0020] 2) When the tester's vision is blocked by parked vehicles on the road;

[0021] 3) When the approaching vehicle is a vehicle with special work nature.

[0022] In the present invention, the Mann-Whitney U or Kruskal-Wallis H is used to determine factors that significantly affect the test subject's gaze behavior.

[0023] In the present invention, a certain influencing factor is divided into two categories, and the Mann-Whitney U test is used to determine whether the vehicle type factor has a significant impact on the number and time percentage of pedestrians looking at different AOIs of the vehicle: the looking time and number of looking times collected in step (1) are divided into two groups according to the above categories;

[0024] ① Propose a hypothesis:

[0025] Null hypothesis H 0: The two sets of data come from the same population, that is, the distribution is the same and the median is equal;

[0026] ②Merge and sort data:

[0027] Merge the two sets of data into one data set, sort all data points in ascending order, and assign a rank to each value; if there are identical values, assign the average rank;

[0028] ③Calculate the rank sum of each group:

[0029] The sample size of the first group is n 1, the sample size of the second group is n 2. Calculate the rank sum of the two groups separately R 1 and R 2;

[0030] ④Calculation U Statistics:

[0031] Use the following formula to calculate the statistics U1 and U2 of the two groups of sample sizes:

[0032] ;

[0033] ;

[0034] The final U value is U 1 and U The smaller of the two, that is ;

[0035] ⑤ Determine significance:

[0036] Small sample :Check the Mann-Whitney U test critical value table, if , then reject H 0;

[0037] Large sample : Calculated using the normal approximation Z value:

[0038] ;

[0039] according to Z The value is determined by looking up the standard normal distribution table p value; the correction term is the correction of the same value; if there are a lot of same values ​​in the data, U The standard deviation of the value is corrected; the formula for the correction term is:

[0040] ;

[0041] in, t j Indicates the j The number of occurrences of repeated values;

[0042] ⑥Interpretation of results:

[0043] like p Value < threshold , reject the original hypothesis and think that there is a significant difference in the distribution or median of the two groups of data; if , the null hypothesis is not rejected.

[0044] In the present invention, a certain influencing factor includes k situation, ; The percentage of pedestrians’ gaze times and gaze time are calculated based on k The situations are divided into k The Kruskal-Wallis H test was used to determine whether the vehicle type factor had a significant impact on the number and time percentage of pedestrians looking at different AOIs of the vehicle. The specific steps are as follows:

[0045] ① Propose a hypothesis:

[0046] Null hypothesis H 0: The distribution of each group of data is the same, and the median is equal;

[0047] ②Merge data:

[0048] All k Merge the group data into one data set; Arrange all data points in ascending order, assign a rank to each value, and take the average rank for the same values;

[0049] ③Calculate the rank sum of each group and the total sample size:

[0050] Record The sample size of the group is n i , the rank sum is R i , total sample size ;

[0051] ④Calculation H Statistics:

[0052] ;

[0053] If there are a lot of identical values, H The value is corrected, the formula is:

[0054] ;

[0055] in, t j Indicates the j The number of occurrences of repeated values;

[0056] ⑤ Determine significance:

[0057] H The statistic is approximately subject to the degrees of freedom k Chi-square distribution of -1 ;

[0058] like or p Value < threshold ,reject H 0, it is considered that at least one group of data is different from the other groups; otherwise, it is not rejected H 0；

[0059] ⑥ If H 0 is rejected, and a post hoc test is performed:

[0060] Determine which groups have differences, perform multiple corrections, and use Bonferroni correction to adjust the significance level of post hoc comparisons; the specific steps of Bonferroni correction are:

[0061] (a) Specify how many hypothesis tests are needed;

[0062] (b) Setting the overall significance level value;

[0063] (c) Calculate the corrected significance level or adjust p value;

[0064] Method 1: Adjust the significance level :

[0065] ;

[0066] in, m The number of inspections required;

[0067] Method 2: Adjustment p value:

[0068] ;

[0069] in, The original inspection p value.

[0070] (d) Perform statistical tests on each comparison to obtain the original p value;

[0071] (e) Comparison p Value and threshold,

[0072] If the adjusted ,when , reject the null hypothesis;

[0073] If the adjusted p Value, when , reject the null hypothesis;

[0074] (f) Indicate whether the Bonferroni correction was used and indicate the adjusted significance level or p value.

[0075] In the present invention, a Markov chain is used to quantify the probability of pedestrians transitioning between different sight focus points under different influencing factors, as follows:

[0076] First, the eye movement data of pedestrians crossing the street were grouped according to different influencing factors, and the data with the same influencing factors were grouped together. Each group of data was analyzed using a Markov chain.

[0077] Assuming that the gaze behavior of pedestrians in crossing the street is a random sequence, we use To represent the fixation moment, the model construction steps are as follows;

[0078] (1) Calculate the probability of a pedestrian's gaze shifting in one step, as follows:

[0079] (a) Using the eye movement data of pedestrians crossing the street, we can calculate the probability matrix of gaze transitions. Each gaze focus point needs to be defined as a state of a Markov chain. Let A, B, C, and D be the four states of the gaze point, which are associated with the windshield, hood, grille, and bumper areas in the AOI. S for ，F uv From the state u Transfer to status v The frequency of , then F 11 Represents the frequency of the current gaze point being in state A and the next point being in state A; F 23 Represents the frequency of the current gaze point being in state B, the next point being in state C, and so on;

[0080] (b) Markov chain time Step transition probability Represents known Always in the state u , After a unit of time, it is in the state v The probability of the state transition of the Markov chain is only related to the departure and arrival states. u and v , transfer steps It is related to the starting time of the transfer, so Record p uv ;

[0081] (c) Assume , then the transition probability expression of state to state is as follows:

[0082] ;

[0083] in, F u From the state u The total frequency of transitions into any state; F uv From the state u Transfer to status v The frequency of ;

[0084] (d) In the state space S In any two states u and v There is a transition probability between p uv , and satisfy , organize all transition probabilities between states in the state space into a two-dimensional matrix P in order:

[0085] ;

[0086] (e) In summary, the one-step transition probability matrix of the pedestrian gaze point Markov chain can be obtained. P Draw a state transition probability histogram;

[0087] (2) Calculate the stable probability distribution of pedestrian gaze behavior; the details are as follows:

[0088] (a) The pedestrian's gaze point presents different states when it is in each visual focus area. The visual focus area of ​​the next gaze point is only related to the visual focus area of ​​the current gaze point. Therefore, it belongs to a typical homogeneous Markov chain. Sis the state space, whose probability distribution , if:

[0089] ;

[0090] Then it is called Markov chain stationary distribution; is the status u The stationary probability distribution under is the status v A stationary Markov chain is a strictly stationary random process whose evolution is independent of the time sequence.

[0091] (b) The ergodic nature of the Markov chain indicates that over a period of time, the system reaches a stable stage, from the initial state u Transition to the state at the corresponding time v The probability trend , the expression is:

[0092] ;

[0093] (c) Based on the probability of a pedestrian’s gaze shifting in one step, a set of four linear equations is established:

[0094] ;

[0095] in, is the transition probability p uv The transposed matrix of diag(1,1,1,1) is the identity matrix. Solve the steady-state probability distribution of the pedestrian's gaze at the focus points A, B, C, and D. , is a four-dimensional column vector, , each component Indicates the part of the pedestrian's vision that is focused during the gaze process The steady-state probability of Corresponding to the four visual focus areas A, B, C, and D.

[0096] In the present invention, the position of the external human-machine interface is determined based on the visual transfer characteristics, specifically as follows:

[0097] (1) For each combination of influencing factors, the stationary probability distribution of pedestrian gaze under that condition is extracted from the Markov chain analysis results;

[0098] (2) Compare the stability probabilities of the four regions under the same combination of influencing factors and select the region with the highest stability probability as the preferred location of the external human-machine interface;

[0099] (3) If the difference between the stationary probabilities of the region with the highest stationary probability and the region with the second highest stationary probability is less than the threshold under the same combination of influencing factors, the state transition probability matrix is ​​further analyzed:

[0100] (a) Calculate the sum of the probabilities of each region being the target of gaze shift and being shifted to other regions;

[0101] (b) Select the area with the highest total probability of being transferred as the location of the external human-machine interface;

[0102] (4) According to the rules of (1) to (3) above, the optimal position of the external human-machine interface is determined for each combination of influencing factors, and a position mapping table is generated, which records the optimal external human-machine interface positions corresponding to different influencing factors.

[0103] Compared with the prior art, the present invention has the following beneficial effects:

[0104] 1. This invention breaks through the limitations of the traditional fixed position of the external human-machine interface. It takes into account that the visual search characteristics of pedestrians when crossing the street may be affected by vehicle type, vehicle color and longitudinal distance. Through pedestrian visual search tests and data analysis, it derives the factors that will have a significant impact on the visual search strategy of pedestrians when crossing the street, providing data support for exploring the determination of the external human-machine interface position.

[0105] 2. The present invention combines Markov chains to analyze the overall distribution trend of visual search during long-term gaze when pedestrians cross the street. Based on this result, the location of the external human-machine interface is determined, which is more in line with the visual search characteristics of pedestrians and enables pedestrians to focus on the external human-machine interface more quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 Flowchart of the pedestrian visual search test in the present invention.

[0107] Figure 2 It is the focus point of the vehicle's vision in the present invention. DETAILED DESCRIPTION

[0108] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0109] The present invention discloses a method for determining the position of an external human-machine interface based on pedestrian visual search features, which is described in detail below.

[0110] (1) According to the appearance of the vehicle, multiple sight focus areas are divided, and the vehicle parts are divided into four parts: windshield, hood, grille and bumper, such as Figure 2As shown, a pedestrian visual search test was conducted to collect the testers' gaze time and gaze frequency on different sight focus points of vehicles of different colors and types at different longitudinal distances;

[0111] Step 1: The scene was set at a two-way, four-lane pedestrian crossing that was not controlled by traffic lights. The test subject wore Tobii Pro Glasses 2 to record eye movements and perform calibration.

[0112] Step 2: The tester observes the traffic conditions as in daily life and starts crossing the street when he / she deems it safe. The tester must stop and wait for the opportunity to cross the street when the following three conditions occur: 1) there are no motor vehicles coming from the oncoming direction; 2) there are parked vehicles on the road (such as a bus temporarily stopping at a station, or a taxi parked on the roadside waiting for passengers); 3) the approaching vehicle is a vehicle with special work nature (such as a police car, ambulance, engineering vehicle, bus, etc.).

[0113] Step 3: A valid crossing is considered one in which the participant interacts with a vehicle and crosses the crosswalk before the vehicle approaches. A total of 20 valid crossings are required. Researchers on the side of the road judge the participant's valid crossings and record them until 20 valid crossings are completed, at which point the test concludes. A valid crossing in this test is defined as: 1) the participant interacts with the vehicle before crossing, meaning the participant maintains their gaze on the vehicle; 2) the interacting vehicle is not a specialized vehicle; and 3) the interacting vehicle directly influences the participant's decision to cross, meaning the participant safely completes the crossing before the interacting vehicle reaches the participant's location. Figure 1 The entire test process is shown.

[0114] (2) Perform differential analysis on the data from step (1)

[0115] (1) The test subject's eye movement data was processed and key information was encoded, as shown in Table 1. To ensure the effectiveness of the vehicle longitudinal distance encoding, the distances of all easily recognizable markers on the road were measured, such as roadside trees and traffic signs on the ground. Using a consistent estimation standard across the test subjects, the range of gaze point distances relative to these markers can be roughly calculated.

[0116] Table 1: Key Information

[0117]

[0118] (2) Data processing revealed the participants' gaze duration and gaze frequency at different vehicle parts. The percentages of gaze duration and gaze frequency were calculated. Nonparametric tests were then used to analyze differences in the percentages of gaze duration and gaze frequency for pedestrians with different vehicle types, vehicle colors, and longitudinal distances.

[0119] (2.1) Since the vehicle types in the test were divided into two different levels (sedans and SUVs), the percentage of pedestrian fixations and the percentage of fixation time were also divided into two groups. The Mann-Whitney U test was used to determine whether the vehicle type factor had a significant impact on the number of fixations and the percentage of time pedestrians spent on different AOIs of the vehicle. The Mann-Whitney U test is used to compare whether the medians or distributions of two independent samples are different. The specific steps are:

[0120] ① Propose a hypothesis:

[0121] Null hypothesis H 0: The two sets of data come from the same population (same distribution, same median).

[0122] ②Merge and sort data:

[0123] Combine the two data sets into one, sort all data points in ascending order, and assign a rank to each value. If there are identical values, assign them the average rank.

[0124] ③Calculate the rank sum of each group:

[0125] The sample size of the first group is n 1, the sample size of the second group is n 2. Calculate the rank sum of the two groups separately R 1 and R 2.

[0126] ④Calculation U Statistics:

[0127] Use the following formula to calculate the statistics U1 and U2 of the two groups of sample sizes:

[0128] ;

[0129] ;

[0130] The final U value is U 1 and U The smaller of the two, that is .

[0131] ⑤ Determine significance:

[0132] Small sample :Check the Mann-Whitney U test critical value table, if , then reject H 0.

[0133] Large sample : Calculated using the normal approximation Z value:

[0134] ;

[0135] according to Z The value is determined by looking up the standard normal distribution table p value; the correction term is the correction of the same value; if there are a lot of same values ​​in the data, U The standard deviation of the values ​​is corrected. The formula for the correction term is:

[0136] ;

[0137] in, t j Indicates the j The number of occurrences of repeated values.

[0138] ⑥Interpretation of results:

[0139] like , reject the original hypothesis and think that there is a significant difference in the distribution or median of the two groups of data; if , the null hypothesis is not rejected.

[0140] (2.2) Since the test divided the vehicle color (white, black, and red) and longitudinal distance (0-20m, 20-40m, and 40-60m) into three different levels, the percentage of pedestrian gazes and the percentage of gaze time were also divided into three groups. The Kruskal-Wallis H test was used to determine whether there were significant differences in the number of gazes and the percentage of time that pedestrians spent on different AOIs on the vehicle when the vehicle was at different longitudinal distances or when the vehicle was of different colors. The Kruskal-Wallis H test is commonly used to compare whether there are differences between the average ranks of multiple independent samples. The specific steps are as follows:

[0141] ① Propose a hypothesis:

[0142] Null hypothesis H 0: The distribution of each group of data is the same (the medians are equal).

[0143] ②Merge data:

[0144] All k The data of the groups are merged into one data set (in this test data ), sort all data points in ascending order, assign a rank to each value, and take the average rank for the same values.

[0145] ③Calculate the rank sum of each group and the total sample size:

[0146] Record The sample size of the group is n i , the rank sum is R i , total sample size .

[0147] ④Calculation H Statistics:

[0148] ;

[0149] If there are a lot of identical values, H The value is corrected, the formula is:

[0150] ;

[0151] in, t j Indicates the j The number of occurrences of repeated values.

[0152] ⑤ Determine significance:

[0153] H The statistic is approximately subject to the degrees of freedom k Chi-square distribution of -1 .

[0154] like or ,reject H 0, it is considered that at least one group of data is different from the other groups; otherwise, it is not rejected H 0.

[0155] ⑥ Post-test (if H 0 rejected):

[0156] To determine which groups differ, multiple corrections are required, and the Bonferroni correction is used to adjust the significance level of post hoc comparisons. The specific steps of the Bonferroni correction are:

[0157] (a) Specify how many hypothesis tests you need to perform.

[0158] (b) Setting the overall significance level , usually choose .

[0159] (c) Calculate the corrected significance level or adjust p value.

[0160] Method 1: Adjust the significance level :

[0161] ;

[0162] in, is the number of times that the test needs to be performed. In this test data, .

[0163] Method 2: Adjustment p value( ):

[0164] ;

[0165] in, For the original inspection p value.

[0166] (d) Perform statistical tests on each comparison to obtain the original p value;

[0167] (e) Comparison p Values ​​and thresholds;

[0168] If the adjusted , , reject the null hypothesis.

[0169] If the adjusted p value, , reject the null hypothesis.

[0170] (f) Indicate whether the Bonferroni correction was used and indicate the adjusted significance level or p value.

[0171] (2.3) According to the results of Mann-Whitney U test and Kruskal-Wallis H test p The three factors in the value judgment test (vehicle type, vehicle color, and longitudinal distance) were determined to have a significant impact on pedestrian gaze behavior when crossing the street. If a factor did not have a significant impact, it was excluded; if it had a significant impact on the percentage of gaze time and the percentage of gaze counts, it was defined as an influencing factor.

[0172] (3) Construct a Markov chain gaze model of pedestrian crossing behavior to explore the characteristics of pedestrian visual shifts when looking at interacting vehicles under different influencing factors. First, group the pedestrian eye movement data according to different influencing factors, and group the data under the same influencing factor (e.g., all eye movement data of pedestrians under red vehicles). Each data set is analyzed using a Markov chain. Assuming that pedestrian gaze behavior during crossing is a random sequence, the model construction steps are as follows:

[0173] (1) Calculate the probability of a pedestrian's gaze shifting in one step, as follows:

[0174] (a) Using the eye movement data of pedestrians crossing the street, we can calculate the probability matrix of gaze transitions. Each gaze focus point needs to be defined as a state of a Markov chain. Let A, B, C, and D be the four states of the gaze point, which are associated with the windshield, hood, grille, and bumper areas in the AOI in turn. The state space S is ，F uv From the state u Transfer to status v The frequency of , then F 11 Represents the frequency of the current gaze point being in state A and the next point being in state A; F 23 It represents the frequency that the current gaze point is in state B, the next point is in state C, and so on.

[0175] (b) Markov chain time Step transition probability Represents known Always in the state u , After a unit of time, it is in the state v The probability of the state transition of the Markov chain is only related to the departure and arrival states. u and v , transfer steps It is related to the starting time of the transfer, so Record p uv .

[0176] (c) Assume , then the transition probability expression of state to state is as follows:

[0177] ;

[0178] in, F u From the state u The total frequency of transitions into any state; F uv From the state u Transfer to status v The frequency of ;

[0179] (d) In the state space S In any two states u and v There is a transition probability betweenp uv , and satisfy , organize all transition probabilities between states in the state space into a two-dimensional matrix in order, where u Row, No. v The elements of the column are p uv , let this two-dimensional matrix be P, then: ;

[0180] (e) In summary, the one-step transition probability matrix of the pedestrian gaze point Markov chain can be obtained. P Draw a histogram of state transition probabilities.

[0181] (2) Calculate the stable probability distribution of pedestrian gaze behavior. The details are as follows:

[0182] (a) When the pedestrian's gaze point is in different areas, it presents different states. The area where the next gaze point is located is only related to the area where the current gaze point is located. Therefore, it belongs to a typical homogeneous Markov chain. S is the state space, whose probability distribution , if:

[0183] ;

[0184] Then it is called Markov chain stationary distribution. is the status u The stationary probability distribution under is the status v A stationary Markov chain is a strictly stationary random process whose evolution is independent of the time sequence.

[0185] (b) The ergodic nature of the Markov chain indicates that over a period of time, the system reaches a stable stage, from the initial state u Transition to the state at the corresponding time v The probability trend , the expression is:

[0186] ;

[0187] (c) Based on the probability of a pedestrian’s gaze shifting in one step, a set of four linear equations is established:

[0188] ;

[0189] in, is the transition probability p uv The transposed matrix of diag(1,1,1,1) is the identity matrix. Solve the steady-state probability distribution of the pedestrian's gaze at the focus points A, B, C, and D. , is a four-dimensional column vector, , each component Indicates the part of the pedestrian's vision that is focused during the gaze process The steady-state probability of Corresponding to the four visual focus areas A, B, C, and D.

[0190] Through the above steps (1) to (2), the steady-state probability distribution of pedestrians’ gaze at the focus points A, B, C, and D under different influencing factors is solved. The stable probability distribution of pedestrian gaze at different areas under various influencing factors is visually compared.

[0191] (IV) Based on the calculated results of the probability of a pedestrian's gaze shifting in one step and the probability of stable gaze behavior, we can analyze the situation of pedestrians shifting their gaze points at various parts of the vehicle. Ultimately, we can obtain the gaze shift patterns of pedestrians in crossing behavior under different influencing factors, providing a theoretical basis for the layout of external human-computer interaction interfaces. The specific steps are as follows:

[0192] (1) For each combination of influencing factors (e.g., specific vehicle type, color, longitudinal distance), the stationary probability distribution of pedestrian gaze under that condition is extracted from the Markov chain analysis results;

[0193] (2) Compare the stability probabilities of the four regions under the same combination of influencing factors and select the region with the highest stability probability as the preferred location of the external human-machine interface;

[0194] (3) If the difference between the stationary probabilities of the region with the highest stationary probability and the region with the second highest stationary probability is less than the threshold value under the same combination of influencing factors ,and , then further analyze the state transition probability matrix:

[0195] (a) Calculate the sum of the probabilities of each region being the target of gaze shift and being shifted to other regions;

[0196] (b) Select the area with the highest total probability of being transferred as the location of the external human-machine interface;

[0197] (4) According to the rules of (1) to (3) above, the optimal position of the external human-machine interface is determined for each combination of influencing factors, and a position mapping table is generated, which records the optimal external human-machine interface positions corresponding to different influencing factors.

[0198] Table 2: Optimal external human-machine interface position mapping table

[0199]

[0200] The above is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for determining the position of an external human-machine interface based on pedestrian visual search features, characterized in that: The following steps are involved: (1) Based on the vehicle's appearance, multiple visual focus areas were divided and pedestrian visual search tests were conducted. The testers' gaze time and number of gazes on different visual focus areas of vehicles of different colors and types at different longitudinal distances were collected; (2) Conduct variance analysis on the data to identify factors that significantly influence the test subjects’ gaze behavior; (3) Use Markov chains to analyze the visual transfer characteristics of pedestrians under different influencing factors, and quantify the probability of pedestrians' gaze stability at different visual focus points and the probability of transfer between different visual focus points under different influencing factors; (4) Based on the results obtained in step (3), determine the location of the vehicle's external human-machine interface.

2. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 1, characterized in that: The sight focus areas include the windshield, hood, grille and bumper.

3. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 1, characterized in that: The pedestrian visual search test includes: Step 1: The test scene is set at a two-way four-lane pedestrian crossing that is not controlled by traffic lights. The tester wears an eye tracker to record eye movements. Step 2: Have multiple testers observe the traffic conditions and begin crossing the street when they deem it safe. If it is unsafe, the testers must stop and wait for the opportunity to cross. Step 3: The tester interacts with the motor vehicle and crosses the crosswalk before the vehicle approaches, completing the crossing behavior, which is considered a valid crossing. The test ends after completing multiple valid crossings.

4. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 3, characterized in that: The unsafe conditions include the following three conditions: 1) When there are no motor vehicles on the road in the direction of oncoming traffic; 2) When the tester's vision is blocked by parked vehicles on the road; 3) When the approaching vehicle is a vehicle with special work nature.

5. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 1, characterized in that: Mann-Whitney U or Kruskal-Wallis H was used to identify factors that significantly affected the subjects' gaze behavior.

6. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 5, characterized in that: For a certain influencing factor divided into two categories, the Mann-Whitney U test is used to determine whether the vehicle type factor has a significant impact on the number and time percentage of pedestrians looking at different AOIs of the vehicle: the gaze time and number of gazes collected in step (1) are divided into two groups according to the above categories; ① Propose a hypothesis: Null hypothesis H 0: The two sets of data come from the same population, that is, the distribution is the same and the median is equal; ②Merge and sort data: Merge the two sets of data into one data set, sort all data points in ascending order, and assign a rank to each value; if there are identical values, assign the average rank; ③Calculate the rank sum of each group: The sample size of the first group is n 1, the sample size of the second group is n 2. Calculate the rank sum of the two groups separately R 1 and R 2; ④Calculation U Statistics: Use the following formula to calculate the statistics U1 and U2 of the two groups of sample sizes: ； ； The final U value is U 1 and U The smaller of the two, that is ; ⑤ Determine significance: when When: Check the Mann-Whitney U test critical value table, if , then reject H 0; when When: Calculated using the normal approximation method Z value: ； according to Z The value is determined by looking up the standard normal distribution table p value; the correction term is the correction of the same value; if there are a lot of same values ​​in the data, U The standard deviation of the value is corrected; the formula for the correction term is: ； in, t j Indicates the j The number of occurrences of repeated values; ⑥Interpretation of results: like p Value < threshold , reject the original hypothesis and think that there is a significant difference in the distribution or median of the two groups of data; if , the null hypothesis is not rejected.

7. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 5, characterized in that: For a certain influencing factor, k situation, ; The percentage of pedestrians’ gaze times and gaze time is calculated based on k The situations are divided into k The Kruskal-Wallis H test was used to determine whether the vehicle type factor had a significant impact on the number and time percentage of pedestrians looking at different AOIs of the vehicle. The specific steps are as follows: ① Propose a hypothesis: Null hypothesis H 0: The distribution of each group of data is the same, and the median is equal; ②Merge data: All k Merge the group data into one data set; Arrange all data points in ascending order, assign a rank to each value, and take the average rank for the same values; ③Calculate the rank sum of each group and the total sample size: Record The sample size of the group is n i , the rank sum is R i , total sample size ; ④Calculation H Statistics: ； If there are a lot of identical values, H The value is corrected, the formula is: ； in, t j Indicates the j The number of occurrences of repeated values; ⑤ Determine significance: H The statistic is approximately subject to the degrees of freedom k Chi-square distribution of -1 ; like or p Value < threshold ,reject H 0, it is considered that at least one group of data is different from the other groups; otherwise, it is not rejected H 0； ⑥ If H 0 is rejected, and a post hoc test is performed: Determine which groups have differences, perform multiple corrections, and use Bonferroni correction to adjust the significance level of post hoc comparisons; the specific steps of Bonferroni correction are: (a) Specify how many hypothesis tests are needed; (b) Setting the overall significance level value; (c) Calculate the corrected significance level or adjust p value; Method 1: Adjust the significance level ; ； in, m The number of inspections required; Method 2: Adjustment p value: ； in, For the original inspection p value; (d) Perform statistical tests on each comparison to obtain the original p value; (e) Comparison p Value and threshold, If the adjusted ,when , reject the null hypothesis; If the adjusted p Value, when , reject the null hypothesis; (f) Indicate whether the Bonferroni correction was used and indicate the adjusted significance level or p value.

8. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 2, characterized in that: A Markov chain is used to quantify the probability of pedestrians transitioning between different sight focus points under different influencing factors, as follows: First, the eye movement data of pedestrians crossing the street were grouped according to different influencing factors, and the data with the same influencing factors were grouped together. Each group of data was analyzed using a Markov chain. Assuming that pedestrians’ gaze behavior during street crossing is a random sequence, the model construction steps are as follows; (1) Calculate the probability of a pedestrian's gaze shifting in one step, as follows: (a) Using the eye movement data of pedestrians crossing the street, we can calculate the probability matrix of gaze transitions. Each gaze focus point needs to be defined as a state of a Markov chain. Let A, B, C, and D be the four states of the gaze point, which are associated with the windshield, hood, grille, and bumper areas in the AOI in turn. The state space S for ,F uv From the state u Transfer to status v The frequency of ; (b) Markov chain time Step transition probability Represents known Always in the state u , After a unit of time, it is in the state v The probability of the Markov chain state transition probability is only related to the transfer of departure and arrival states. u and v , transfer steps It is related to the starting time of the transfer, so Record p uv ; (c) Assume , then the transition probability expression of state to state is as follows: ； in, F u From the state u The total frequency of transitions into any state; F uv From the state u Transfer to status v The frequency of ; (d) In the state space S In any two states u and v There is a transition probability between p uv , and satisfy , organize all transition probabilities between states in the state space into a two-dimensional matrix P in order: ； (e) In summary, we can get the one-step transition probability matrix of the pedestrian gaze point Markov chain, and according to P Draw a state transition probability histogram; (2) Calculate the stable probability distribution of pedestrian gaze behavior; the details are as follows: (a) The pedestrian's gaze point presents different states when it is in each visual focus area. The visual focus area of ​​the next gaze point is only related to the visual focus area of ​​the current gaze point. Therefore, it belongs to a typical homogeneous Markov chain. S is the state space, whose probability distribution , is the status u The stationary probability distribution under is the status v The stationary probability distribution under , if it satisfies: ； Then it is called Markov chain stationary distribution; a stationary Markov chain is a strictly stationary random process whose evolution is independent of time sequence; (b) The ergodic nature of the Markov chain indicates that over a period of time, the system reaches a stable stage, from the initial state u Transition to the state at the corresponding time v The probability trend , the expression is: ； (c) Based on the probability of a pedestrian’s gaze shifting in one step, a set of four linear equations is established: ； in, P T is the transition probability p uv The transposed matrix of diag(1,1,1,1) is the identity matrix. Solve the steady-state probability distribution of the pedestrian's gaze at the focus points A, B, C, and D. , is a four-dimensional column vector, , each component Indicates the part of the pedestrian's vision that is focused during the gaze process The steady-state probability of Corresponding to the four visual focus areas A, B, C, and D.

9. The method for determining the position of an external human-machine interface based on pedestrian visual search features according to claim 1, characterized in that: The position of the external human-machine interface is determined based on the visual transfer characteristics, as follows: (1) For each combination of influencing factors, the stationary probability distribution of pedestrian gaze under that condition is extracted from the Markov chain analysis results; (2) Compare the stability probabilities of the four regions under the same combination of influencing factors and select the region with the highest stability probability as the preferred location of the external human-machine interface; (3) If the difference between the stationary probabilities of the region with the highest stationary probability and the region with the second highest stationary probability is less than the threshold under the same combination of influencing factors, the state transition probability matrix is ​​further analyzed: (a) Calculate the sum of the probabilities of each region being the target of gaze shift and being shifted to other regions; (b) Select the area with the highest total probability of being transferred as the location of the external human-machine interface; (4) According to the rules of (1) to (3) above, the optimal position of the external human-machine interface is determined for each combination of influencing factors, and a position mapping table is generated, which records the optimal external human-machine interface positions corresponding to different influencing factors.

Images