A system and method for improving driving confidence based on optimization of interactive interface scenario elements presentation

Abstract

The present invention provides a system for improving driving trust based on the optimization of the presentation of interactive interface scenario elements, wherein the sensor module is responsible for collecting the perceived surrounding environment and its own position information data, and then sending it to the data processing module; the data processing module includes a preprocessing module, an identification module, and a map module; the preprocessing module of the data processing module transmits the information to the identification module; the identification module identifies the preprocessed data after receiving it, and then transmits the identified data to the map module; the map module generates map information after receiving it, and then transmits the generated map information data to the element selection module and the interface module; the element selection module receives the map information of the map module and filters it, and the interface module presents the filtered information through the central control screen, and visualizes it to the driver. The present invention can help the driver analyze the scenario environment and relieve driving pressure, and enhance the driver's trust in the automatic driving system.

Description

Technical field:

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a system and method for improving driving confidence based on optimizing the presentation of scenario elements in an interactive interface. Background technology:

[0002] One of the main reasons that affects drivers' trust in autonomous driving systems is that the interaction of the human-computer interface is not efficient enough, that is, the interactive interface does not convey information to the driver accurately and efficiently, resulting in the driver's delay in receiving information or difficulty in understanding the true meaning of the system, thus causing the driver to lack trust in the autonomous driving system.

[0003] Among the existing technologies, the current technologies for improving driving trust are not efficient and practical. For example, patent CN105867595A improves the efficiency of human-computer interaction by sending control instructions in combination with voice information and gesture information. It can be said that it improves driving trust from the side, but the disadvantage is that this interaction technology is still relatively complex and requires continuous interaction to be realized, which is not efficient; CN115027495A collects and integrates vehicle perception information and navigation information through the intelligent driving system to inform users of the processing capabilities of its system, thereby reducing tension and improving trust, but its defect is that this technology is very dependent on the advancement of the system, otherwise it is difficult to accurately predict and process various dangerous details, and it is not suitable for ordinary cars; CN111813470B optimizes and visually enhances the text content in the interactive interface in groups, reduces the user's cognitive load on the interaction, and improves the user experience, but the defect of this technology is that the visual enhancement is text, and the user still needs to read the text to understand it, and it is impossible to transmit information efficiently and timely.

[0004] Therefore, how to efficiently and accurately improve interaction efficiency and user experience in the cockpit, thereby enhancing driving trust, is an urgent problem to be solved. Summary of the invention:

[0005] The purpose of the present invention is to design a system and method for improving driving trust based on optimizing the presentation of scenario elements in the interactive interface, so as to enhance the driver's situational awareness of the surrounding environment, improve the efficiency of human-computer interaction and user experience, thereby enhancing the driver's trust in the automatic driving system, and further provide a basis for efficient interaction between the driver and the automatic driving system.

[0006] The technical solution of the invention proposes a system for improving driving confidence based on the optimization of the presentation of scenario elements in the interactive interface, including a sensor module, a data processing module, an element selection module, and an interface module, and is characterized in that:

[0007] Among them, the sensor module includes a camera, a laser radar, an ultrasonic sensor, and a GPS positioning sensor; the sensor module is responsible for collecting the perceived surrounding environment and its own location information data, and then sending it to the data processing module;

[0008] The data processing module includes a preprocessing module, a recognition module, and a map module; the preprocessing module of the data processing module performs filtering, noise reduction, and segmentation preprocessing after receiving the data collected by the sensor, and then passes the preprocessed information to the recognition module; the recognition module recognizes roads, vehicles, pedestrians, and obstacles after receiving the preprocessed data, and then passes the recognized data to the map module; the map module generates map information after receiving it, including roads, pedestrians, vehicles, traffic infrastructure, and obstacles, and then transmits the generated map information data to the element selection module and the interface module

[0009] After receiving the map information from the map module, the element selection module filters the scenario elements that affect the current event and passes the filtered information to the interface module; the interface module presents the information filtered by the element selection module through the central control screen and visualizes it to the driver.

[0010] The invention also provides a method for improving driving trust based on the optimization of the presentation of scenario elements in an interactive interface, the method comprising the following steps:

[0011] Step 1: After receiving the information from the map module of the data processing module, the element selection module initializes the information and encodes it, converting each trip driven by the autonomous driving mode into a directed graph; wherein each node of each trip is encoded into a vector of K blocks, each block represents the factor feature affecting the corresponding node, and the similarity of all adjacent nodes based on the factor feature is calculated and normalized;

[0012] Step 2: Input the factors influencing each event after embedding in step 1 into the first GGNN layer of the first DIEL layer, use distance correlation to decompose similar factors and separate them, then propagate and update learning through multiple GGNN layers, and then connect and fuse the decomposed factors through the attention mechanism;

[0013] Step 3: After all events are embedded after updating factor by factor in step 2, the embeddings of all events are aggregated to complete the embedding of the trip set. During the aggregation process, attention weights are assigned to each factor of each event to evaluate the influence of the event factor on the event.

[0014] Step 4: After all events are aggregated and embedded into the itinerary in step 3, the itinerary S g,K and each event factor c i,kThe dot product of is used to obtain the probability of each event. Among the events of the same type, the maximum probability is taken as the representative event of this type of event. The factor of this representative event is highlighted, and then return to step 1 to repeat the above steps to ensure real-time update of environmental data, thereby achieving real-time targeted key presentation.

[0015] Advantages of the present invention:

[0016] 1) The present invention can help drivers analyze the situational environment and relieve driving pressure by selecting situational awareness elements for key presentation, greatly improving interaction efficiency and user experience, and enhancing drivers' trust in the autonomous driving system;

[0017] 2) Simplified situational awareness elements of the interactive interface, which helps drivers to correctly restore or improve their situational awareness, thereby improving safety and trust;

[0018] 3) By highlighting the situational awareness elements that affect the event at the time, users can quickly obtain important information and improve trust;

[0019] 4) It can improve the driver's driving interference caused by distrust due to failure to obtain information in a timely manner in emergency situations, thereby improving safety and trust in autonomous driving. Description of the drawings:

[0020] Figure 1 It is a system proposed by the present invention to improve driving confidence based on the optimization of presentation of scenario elements in the interactive interface;

[0021] Figure 2 A directed graph is obtained by converting a trip driven in an automatic driving mode according to an example of the present invention;

[0022] Figure 3 is a model diagram of a Disen_GNN algorithm for optimizing scene elements of an interactive interface to make them presented in a focused manner according to an example of the present invention;

[0023] Figure 4 is an in-degree and out-degree similarity matrix diagram obtained according to the characteristics of the directed graph of the example of the present invention;

[0024] Figure 5 It is a schematic flow chart of a method for improving driving confidence based on optimizing the presentation of scenario elements in an interactive interface according to an example of the present invention. Specific implementation method:

[0025] The following is combined with Figure 1-5 The technical solution of the present invention is described in detail.

[0026] like Figure 1As shown, this embodiment provides a system for improving driving confidence based on optimizing the presentation of scenario elements in an interactive interface, including a sensor module, a data processing module, and other modules.

[0027] Among them, the sensor module includes a camera, a laser radar, an ultrasonic sensor, and a GPS positioning sensor;

[0028] The data processing module includes a preprocessing module, a recognition module, and a map module;

[0029] Other modules include element selection module and interface module.

[0030] Among them, the sensor module is mainly responsible for collecting the perceived surrounding environment and its own location information data, and then sending it to the data processing module;

[0031] The preprocessing module of the data processing module receives the data collected by the sensor and performs preprocessing such as filtering, noise reduction, and segmentation, and then passes the preprocessed information to the recognition module;

[0032] After receiving the pre-processed data, the recognition module identifies roads, vehicles, pedestrians, and obstacles, and then passes the identified data to the map module;

[0033] After receiving the data, the map module generates map information, including roads, pedestrians, vehicles, traffic infrastructure, obstacles, etc., and then transmits the generated map information data to other modules;

[0034] After receiving the map information from the map module, the element selection module of other modules filters the scenario elements that affect the current event and passes the filtered information to the interface module; the interface module presents the information filtered by the element selection module through the central control screen and visualizes it to the driver.

[0035] In this embodiment, the surrounding environment information collected by the sensor module includes vehicles, roads, pedestrians, transportation infrastructure, buildings, vegetation, possible animals, etc.;

[0036] The self-position information collected by the sensor module is mainly used to monitor the position of the main vehicle on the road in real time and return this position information;

[0037] The information filtered by the element selection module in the other modules refers to the information after being filtered by the method proposed by the present invention, including the surrounding environment map information and the selected scene element information to be highlighted, and this information is finally sent to the interface module for visual presentation.

[0038] It can be seen that in this embodiment, the embedding of factors is to map these influencing factors into variables, the embedding of events is to map events into vectors, and the embedding of trips is to map trips into vectors. The embedding of events takes into account the factors that affect events, so all factors must be embedded and learned before events are embedded. Then, after embedding and learning all events on the trip, the trip is embedded. Finally, the probability of the representative event is calculated by the dot product of the factors of the event that occurred on the trip and the trip, so as to obtain the representative event. Then, the factors of the representative event can be highlighted and presented, and they must be updated in real time, because the factors that affect the event cannot remain unchanged.

[0039] like Figure 2 , Figure 3 As shown, this embodiment also provides a method for improving driving trust based on the optimization of the presentation of interactive interface scenario elements. The method is specifically implemented by an element selection module and specifically includes the following steps:

[0040] Step 1: After the element selection module receives the information of the map module of the data processing module, it initializes the information and encodes it, converting each trip driven by the automatic driving mode into a directed graph;

[0041] like Figure 3 As shown in the figure, each node of each trip will be encoded into a vector of K blocks, each block represents the factor features that affect this node. In addition, in order to obtain the similarity between an event and its neighboring events, so as to be able to predict the influencing factor of the most representative event later, it is necessary to calculate the similarity of all neighboring nodes (events) based on the factor features and normalize them.

[0042] In this implementation case, it is assumed that every 100 meters of the autonomous driving mode is a journey, and this journey is converted into a directed graph. In this 100-meter journey, it may experience left turns, right turns, left U-turns, overtaking, and going straight through intersections. These are called events on the journey, also known as nodes. These events will be encoded into K vectors, each of which is a factor affecting these events. Here, the factors are assumed to be seven categories: buildings, vegetation, animals, basic transportation facilities, pedestrians, roads, and other vehicles on the road. If the main event is a left turn, the factors affecting the left turn are pedestrians, vehicles, infrastructure, and buildings; the factors affecting the left U-turn are pedestrians, vehicles, infrastructure, and vegetation. Then the left U-turn is called the neighbor event of the main event left turn. Here, the factors that affect the events are taken into account to obtain the similarity between them, so each factor must be embedded and learned, and then the similarity between events is calculated based on the embedded learning of the factors, and finally the similarity is normalized.

[0043] Step 1 is achieved by:

[0044] Step 1.1: Assume that every 100 meters of driving in the autonomous driving mode is a journey, convert this journey into a directed graph, and construct a directed journey graph G s ,like Figure 3 As shown:

[0045] G s =(V s , E s ) (1)

[0046] Where V s Represents a set of nodes, each node represents an event that occurs during this journey, E s Represents an edge set.

[0047] Step 1.2: All n events encountered during this journey are represented by a set S':

[0048] S'={V s，1 , V s，2 , ... V s，n} (2)

[0049] Where V S，i ——represents the event i that occurs on this trip S. Each S，i-1 To V s，i The directed edge of represents the event V s，i In the event V s，i-1 That is, in this journey, if the event after the left turn is a right turn, it is called a right turn V s，2 Make a left U-turn s，1 After that, it indicates the order in which events occur.

[0050] Step 1.3: Encode each event encountered in this journey described in step 1.2 into K block vectors. Assume that each block represents the factor characteristics that affect the event. Combine these factors and construct a set:

[0051] C = {c1, ...c7} (3)

[0052] The present invention assumes that this influencing factor set is these seven categories of factors: buildings, vegetation, animals, basic transportation facilities, pedestrians, roads, and other vehicles on the road. These are the factors that affect the event.

[0053] Step 1.4: For the trip set S' in step 1.2 = {V s，1 , V s，2 , ... V s，n}, each event V s，i After being mapped into vectors, they can be represented by an embedding vector es，i So this trip set can also be represented as E s ={e s，1 , e s，2 ,...e s，n}.

[0054] Each event on this trip is encoded into k influencing factors, so to encode and embed this event, we must first encode and embed each factor that affects the event. The formula for embedding learning for the kth factor that affects event i on this trip set S' is:

[0055]

[0056] where c i,k ——Represents the embedding learning of the kth factor affecting event i;

[0057] ——are the weight matrix and parameter matrix of the kth factor respectively;

[0058] e s,i ——represents the embedding of event i in trip S';

[0059] Step 1.5: Use the factors learned by encoding embedding in step 1.4 to solve the similarity of adjacent events:

[0060]

[0061] in ——Indicates the similarity between adjacent nodes (events) i and j.

[0062] ——represents the embedding of the kth factor of nodes (events) i and j.

[0063] In this journey, the factors affecting the main event and the neighboring events are likely to have the same parts. For example, the factors affecting left turns are pedestrians, vehicles, infrastructure, and buildings; the factors affecting left U-turns are pedestrians, vehicles, infrastructure, and vegetation. The similarities between the two are pedestrians, vehicles, and infrastructure. Therefore, it is necessary to consider the factors affecting the events to determine the similarity between them.

[0064] Step 1.6: After calculating the similarity between each node and its neighboring nodes in step 1.5, the similarity calculated for each node must be normalized:

[0065] Here, we take the directed edge pointing outward (hereinafter referred to as outgoing) as an example, and its similarity normalization formula is:

[0066]

[0067] in ——It represents the normalized value of the similarity between adjacent nodes (events) i and j;

[0068] ——refers to the set of outgoing neighbors, and the incoming ones can be expressed as

[0069] ——refers to one of the outgoing neighbors.

[0070] Step 1.7: After calculating and normalizing the similarity of all nodes, we get the out-degree similarity matrix and the in-degree similarity matrix, which are like Figure 4 As shown, out-degree refers to the directed edge direction pointing outward, and in-degree refers to the directed edge direction pointing inward.

[0071] The implementation of this step is demonstrated below through a specific embodiment:

[0072] like Figure 3 As shown; after receiving the data from the map module, the element selection module starts to work, and first converts the 100-meter journey of the automatic driving mode into a directed graph; in this embodiment, the left turn in this journey is the main event, which is decomposed into 5 vectors, representing the five factors affecting the left turn: pedestrians, vehicles, infrastructure, buildings, and vegetation; the left turn event needs to be encoded and embedded, but the five factors affecting the left turn event must be considered, so these five factors need to be encoded and embedded first. Similarly, the factors of each event must be encoded and embedded; when encoding and embedding the factors of other events, it is found that the influencing factors of left turn and left turn are very similar. At this time, it is necessary to calculate their similarity and normalize them. Similarly, the same operation is performed on each event, so that it is found that the left turn and left turn have the highest similarity, so the left turn is made the first-level neighbor event of the left turn; at this point, the data initialization and factor embedding are completed, and the embedded factors are embedded in the event through the following step 2.

[0073] Step 2: Input the factors that affect each event after embedding in step 1 into the first GGNN layer of the first DIEL layer, use distance correlation to decompose similar factors and separate them, then propagate and update learning through multiple GGNN layers, and then connect and fuse the decomposed factors through the attention mechanism.

[0074] like Figure 2As shown in the figure, this step mainly realizes the encoding embedding of events through the factors embedded in step 1. Step 1 has completed the encoding embedding of factors. The factors that affect each event after embedding in step 1 are input into the first GGNN layer of the first DIEL layer. The distance correlation is used to decompose similar factors separately, and then the learning is updated through multiple GGNN layers. The decomposed factors are connected and fused through the attention mechanism. At this time, this is the output of the first DIEL layer, which will be the input of the second DIEL layer. The work of the first DIEL layer is repeated to finally complete the encoding embedding of all events. Figure 2 Event V i For example.

[0075] Furthermore, step 2 is implemented by the following method:

[0076] Step 2.1: Input the initialization embedding of step 1 into the GGNN layer of the first DIEL layer, and use the distance correlation method to decompose the different factors of each event so that the factors are independent of each other. The distance correlation method formula can be expressed as:

[0077]

[0078] in ——It means to separate the two factors and make them independent of each other;

[0079] ——represents the initial embedding representation of the kth factor of each event in the trip set S' of the first GGNN layer of the first DIEL layer;

[0080] ——It means and Initial embedding representation of neighboring factors.

[0081] dCor(·) is a function of distance correlation, and its formula is:

[0082]

[0083] Where dCov(·)——is the distance covariance between the two matrices;

[0084] dVar(·)——find the distance covariance of itself;

[0085] Step 2.2: After decomposing the different factors of each event in step 2.1, the characteristics of each factor are independent of each other. Then, multiple GGNN layers are used to propagate and update the factors one by one for the different factors of each event. The propagation update rules of factor embedding are as follows:

[0086]

[0087] in ——is the embedding propagation of the kth factor of event i in the gth GGNN layer;

[0088] ——respectively, the incoming and outgoing weight matrices to be learned;

[0089] ——is the k-th factor embedding representation of all events in the itinerary set S' of the g-1-layer GGNN layer;

[0090] b in , b out ——represent the incoming and outgoing parameter matrices of the i-th factor respectively;

[0091] ——They are respectively the in-degree similarity matrices and out-degree similarity matrix The i-th row of

[0092]

[0093]

[0094] in ——They are update gate and reset gate respectively;

[0095] W G , Q G , W R , Q R ——are their respective weight matrices;

[0096] ——represents the embedding of the i-th node after the g-1-th GGNN layer;

[0097]

[0098] in ——refers to the embedding of event i at the g-th GGNN layer;

[0099] W c , Q c ——represents the weight matrix.

[0100]

[0101] ——represents the embedding of the i-th node after the g-th GGNN layer;

[0102] when You can get the embedding of all nodes (events) after the g-layer GGNN layer.

[0103] Step 2.3: After updating and learning each factor in step 2.2, the attention mechanism is used to fuse and connect the decomposed and updated factors of each event to complete the learning of the update event based on the update factor. The attention mechanism formula can be expressed as:

[0104]

[0105]

[0106] where α L ——refers to the attention weight of the factor to be retained;

[0107] ——refers to the final event embedding output by the previous (L-1th) DIEL layer of event i in the trip set S';

[0108] — Concatenate the learned embeddings of all factors to the embedding of event i obtained from all GGNN layers;

[0109] ——refers to the final event embedding output by the Lth DIEL layer for event i in the trip set S';

[0110] W f ∈R d ——is the weight vector;

[0111] W p , W q ∈R d×d ——is the weight matrix.

[0112] Step 2.4, embed the factors of the update event completed in step 2.3 as the output of this DIEL layer, input it to the next DIEL layer, and return to step 2.1 to start execution and propagate the update again. The number of DIEL layers should not be too many, otherwise over-smoothing will occur. Therefore, the present invention is set to two DIEL layers.

[0113] It can be seen that in this embodiment, taking the left turn event as an example, if the factors affecting it at this time are pedestrians, vehicles, infrastructure, buildings, and vegetation, then these five factors are introduced into the first GGNN layer of the first DIEL layer, and the correlation between each other is eliminated through the distance correlation method formula (7) (8), and the factors are decomposed and separated. At this time, the factors are independent and concise, thereby avoiding information redundancy; the five decomposed and independent factors are continuously propagated and updated in multiple GGNN layers. This propagation update has its own propagation update rules. The formulas (9), (10), (11), (12), and (13) of step 2.2 are all; these five independent factors after propagation update learning The factors will be fused and connected through the attention mechanism formula (14); at this time, the five factors after fusion and connection are the output of the first layer of DIEL layer and are input to the second layer of DIEL layer. The second layer will repeat the work of the first layer; the original five factors affecting left turns are embedded and learned by two layers of DIEL layers to obtain five new factors. The new here is because the updated factors after propagation also consider the neighboring nodes of the node, for example, the main event left turn and the neighboring event left turn are considered, so it has stronger robustness. Through these five new factors, the encoding embedding of the left turn event is realized; similarly, the same operation is performed on other events, and finally the encoding embedding of all events can be completed.

[0114] Step 3: After all events are embedded after updating factor by factor in step 2, the embeddings of all events are aggregated to complete the embedding of the trip set. During the aggregation process, attention weights are assigned to each factor of each event to evaluate the impact of the event factor on the event.

[0115] The meaning of weight here can be understood as: if the influencing factors of left turn are nearby vehicles, roads, pedestrians, buildings, and vegetation, and the first three have the highest weights, it means that these three factors have a great impact on the left turn event.

[0116] Furthermore, step 3 is implemented by:

[0117] Step 3.1: To aggregate all the events experienced during this journey, we must first assign attention weights to each event factor. The formula is:

[0118] α i,k =Q T tanh(W1c n,k +W2c i,k +b) (16)

[0119] where α i,k ——is the attention weight of the kth factor of event i;

[0120] Q——weight vector;

[0121] b——is the parameter to be learned by the attention neural network;

[0122] ——represents the embedding of the kth factor of node (event) i;

[0123] ——represents the embedding of the kth factor of the last node (event) n.

[0124] Step 3.2: After getting the attention weights of the factors in step 3.1, we can generate the embedding of this trip by aggregating events based on the attention weights of the factors:

[0125]

[0126] S g,k ——is the trip embedding that aggregates all events based on the kth factor attention weight;

[0127] S′ g =[S g,1 ,S g,2 ,...,S g,K ] is the final itinerary embedding set.

[0128] It can be seen that in this embodiment, since the purpose of this step is to encode and embed the 100-meter journey, all events experienced in this journey and the factors affecting the events must be recorded, and the degree of influence of each factor is different, so the factors must first be assigned attention weight formula (16). For example, the influencing factors of left turn are nearby vehicles, roads, pedestrians, buildings, and vegetation. The weights of the first three are the largest, which means that these three factors have a great influence on the left turn event; the same operation is performed on other events in this journey to obtain the weight of each factor of all events; after obtaining the attention weights of the factors of these events, the encoding embedding of this 100-meter journey can be completed using this weight and the embedding formula (17) of the event factor.

[0129] Step 4: After all events are aggregated and embedded into the itinerary in step 3, the itinerary S g,K and each event factor c i,k The dot product of is used to obtain the probability of each event. Among the events of the same type, the maximum probability is taken as the representative event of this type of event. The factor of this representative event is highlighted, and then return to step 1 to repeat the above steps to ensure real-time update of environmental data, thereby achieving real-time targeted key presentation.

[0130] Furthermore, step 4 is implemented by the following method:

[0131] Step 4.1: According to the itinerary S g,K and the factor c for each event on this itinerary i,k The dot product of is used to get the probability of each event. Because this event occurs on this trip and is affected by the factor, the probability of the event here is calculated by using the dot product of the event factor and the trip:

[0132]

[0133] Step 4.2: Take the maximum value among the probabilities of similar events. That is, there are two left-turn events, which are called similar events. Take the one with the largest probability value as its representative event:

[0134] Y′ i =softmax(X′ i ) (19)

[0135] Step 4.3: After the representative event of each event is obtained in step 4.2, the factor of the representative event is presented. That is, if a left turn event is encountered in the future, the factor of the left turn representative event will be presented:

[0136] Y′ i =[y′1,y′2,...y′ M ] (20)

[0137] y′ m (1≤m≤M)——is the mth influencing factor of the representative event to be highlighted.

[0138] Step 4.4: After highlighting the influencing factors representing the events, real-time updates are required, and the process returns to step 1, and continues in sequence to achieve continuous dissemination and updates, and to present the influencing factors (scenario-aware elements) in a targeted manner, thereby optimizing the complex and unfocused scenario elements in the interactive interface.

[0139] It can be seen that in this embodiment, assuming that there are two left turns in this 100-meter journey, the factors affecting the first left turn are: vehicles, pedestrians, roads; the factors affecting the second left turn are: roads, vegetation, buildings. Then, the factors of the two events are respectively calculated by formula (18) and the dot product with this journey, and the probability of each event is obtained; then the maximum value of the probability of these two left-turn events is taken as the representative event of the left turn. If the probability value of the first left-turn event is the largest, then it is the representative event of the left turn, because when embedding the journey in step 3.2, formula (17) takes into account the factor weight, so the probability will be large if the factor weight is large, so the maximum probability is selected as the representative event; similarly, the same operation is performed for other events such as left turn, right turn, overtaking, etc. in this journey, and the probability of each representative event can be obtained; when the vehicle makes a left turn, a left turn, etc., the factors representing the left turn, left turn, etc. will be presented, that is, in the future on the road, once a left turn, a left turn, etc. is made, the central control screen will focus on presenting the representative events such as left turn, left turn, etc. The factors (situational awareness elements) representing the event are presented. This emphasis presentation can be understood as making these situational elements more prominent in color and enhancing the display. After presenting the first frame of the factors, it will return to step 1 and restart to continue executing downward according to the new information from the map module to update the influencing factors. By integrating this method into the element selection module of other modules in the system, the sensor module can be continuously relied on to obtain new information about the surrounding environment, and after processing, it will be passed to the element selection module to select the situational elements, and finally transmitted to the interface module for visualization through the central control screen. The effect that can be achieved is that as soon as the driver sees the highlighted situational elements displayed on the central control screen, he will know the important factors affecting the current event and focus on them, which greatly improves the interaction efficiency, thereby improving driving trust and improving driving safety.

[0140] The above is only an embodiment of the present invention, and the common sense such as the known specific structure and characteristics in the scheme is not described in detail here. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the attached claims rather than the above description, so it is intended to include all changes within the meaning and scope of the equivalent elements of the claims. Any figure mark in the claims should not be regarded as limiting the claims involved.

Claims

1. A method for improving driving trust based on optimizing the presentation of scenario elements in an interactive interface, characterized in that: The specific steps include: Step 1: The sensor module collects the surrounding environment and its own position information data perceived by the sensor module, and the data processing module performs filtering, noise reduction, segmentation preprocessing on the data collected by the sensor module, identifies roads, vehicles, pedestrians, obstacles, and generates map information, and then transmits the generated map information data to the element selection module and the interface module. After the element selection module receives the information of the map module of the data processing module, it initializes the information and encodes it, and converts each trip driven by the automatic driving mode into a directed graph; wherein each node of each trip is encoded into a vector of K blocks, each block represents the factor feature that affects the corresponding node, and the similarity of all adjacent nodes based on the factor feature is calculated and normalized; Step 1.1: Assume that every 100 meters of driving in the autonomous driving mode is a journey, convert this journey into a directed graph, and construct a directed journey graph G s : G s =(V s ,E s ) (1) Where V s Represents a set of nodes, each node represents an event that occurs during this journey, E s represents an edge set; Step 1.2: All n events encountered during this journey are represented by a set S': S'={V s,1 ,V s,2 ,...V s,n }(2) Where V s,i ——represents the event i that occurs on this trip S; each from V s,i-1 To V s,i The directed edge of represents the event V s,i In the event V s,i-1 Afterwards passed; Step 1.3: Encode each event encountered in this journey described in step 1.2 into K block vectors. Assume that each block represents the factor characteristics that affect the event, combine the factors and construct a set: C={c1,...c7} (3) The set hypothesis includes seven types of factors: buildings, vegetation, animals, basic transportation facilities, pedestrians, roads, and other vehicles on the road; Step 1.4, the set S' of step 1.2 = {V s,1 ,V s,2 ,...V s,n } for each event V s,i After being mapped into a vector, it is represented as the embedding vector e s,i , so that the set S' is represented by E s ={e s,1 ,e s,2 ,...e a,n }; Each factor that affects the event is encoded and embedded, and the formula for embedding the kth factor that affects event i on this trip set S' is: where c i,k ——Represents the embedding learning of the kth factor affecting event i; ——are the weight matrix and parameter matrix of the kth factor respectively; e s,i ——represents the embedding of event i in trip S'; Step 1.5: Use the factors learned by encoding embedding in step 1.4 to solve the similarity of adjacent events: in ——It represents the similarity between adjacent nodes (events) i and j; ——represents the embedding of the kth factor of nodes (events) i and j; Step 1.6: After calculating the similarity between each node and its neighboring nodes in step 1.5, normalize the similarity obtained for each node: The similarity normalization formula is: in ——It represents the normalized value of the similarity between adjacent nodes i and j; ——refers to the set of outgoing neighbors, and the incoming ones can be expressed as ——refers to one of the outgoing neighbor sets; Step 1.7: After calculating and normalizing the similarity of all nodes, we get the out-degree similarity matrix and the in-degree similarity matrix, which are Out-degree refers to the directed edge pointing outward, and in-degree refers to the directed edge pointing inward; Step 2, input the factors that affect each event after embedding in Step 1 into the first GGNN layer of the first DIEL layer, use distance correlation to decompose similar factors independently, and then propagate and update learning through multiple GGNN layers, and then connect and fuse the decomposed factors through the attention mechanism; Step 2.1: Input the initialization embedding of step 1 into the GGNN layer of the first DIEL layer, and use the distance correlation method to decompose the different factors of each event so that the factors are independent of each other. The distance correlation method formula can be expressed as: in ——It means to separate the two factors and make them independent of each other; ——represents the initial embedding representation of the kth factor of each event in the trip set S' of the first GGNN layer of the first DIEL layer; ——It means and Initial embedding representation of neighboring factors; dCor(·) is a function of distance correlation, and its formula is: Where dCov(·)——is the distance covariance between the two matrices; dVar(·)——find the distance covariance of itself; Step 2.2: After decomposing the different factors of each event in step 2.1, the characteristics of each factor are independent of each other. Then, multiple GGNN layers are used to propagate and update the factors one by one for the different factors of each event. The propagation update rules of factor embedding are as follows: in ——is the embedding propagation of the kth factor of event i in the gth GGNN layer; ——respectively, the incoming and outgoing weight matrices to be learned; ——is the k-th factor embedding representation of all events in the itinerary set S' of the g-1-layer GGNN layer; b in 、b out ——represent the incoming and outgoing parameter matrices of the i-th factor respectively; ——They are respectively the in-degree similarity matrices and out-degree similarity matrix The i-th row of in ——They are update gate and reset gate respectively; W G , Q G , W R , Q R ——are their respective weight matrices; ——represents the embedding of the i-th node after the g-1-th GGNN layer; in ——refers to the embedding of event i at the g-th GGNN layer; W c , Q c ——represents the weight matrix; ——represents the embedding of the i-th node after the g-th GGNN layer; when You can get the embedding of all nodes after the g-layer GGNN layer; Step 2.3: After updating and learning each factor in step 2.2, the attention mechanism is used to fuse and connect the decomposed and updated factors of each event to complete the learning of the update event based on the update factor; the attention mechanism formula can be expressed as: where α L ——refers to the attention weight of the factor to be retained; ——refers to the final event embedding output by the previous (L-1th) DIEL layer of event i in the trip set S'; — Concatenate the learned embeddings of all factors to the embedding of event i obtained from all GGNN layers; ——refers to the final event embedding output by the Lth DIEL layer for event i in the trip set S'; W f ∈R d ——is the weight vector; W p , W q ∈R d×d ——is the weight matrix; Step 2.4: The factor embedding of the update event completed in step 2.3 is used as the output of this DIEL layer, and is input to the next DIEL layer. Then, the process returns to step 2.1 and the update is propagated again. The number of DIEL layers is set to two. Step 3: After all events are embedded after updating factor by factor in step 2, the embeddings of all events are aggregated to complete the embedding of the trip set. During the aggregation process, attention weights are assigned to each factor of each event to evaluate the influence of the event factor on the event. Step 3.1: To aggregate all the events experienced during this journey, we must first assign attention weights to each event factor. The formula is: α i,k =Q T tanh(W1c n,k +W2c i,k +b) (16) where α i,k ——is the attention weight of the kth factor of event i; Q——weight vector; b——is the parameter to be learned by the attention neural network; ——represents the embedding of the kth factor of node (event) i; ——represents the embedding of the kth factor of the last node (event) n; Step 3.2: After getting the attention weights of the factors in step 3.1, we can generate the embedding of this trip by aggregating events based on the attention weights of the factors: S g,k ——is the trip embedding that aggregates all events based on the kth factor attention weight; S g ′=[S g,1 ,S g,2 ,...,S g,K ] is the final itinerary embedding set; Step 4: After all events are aggregated and embedded into the itinerary in step 3, the itinerary S g,K and each event factor c i,k The dot product of is used to obtain the probability of each event, and the maximum probability among the same type of events is taken as the representative event of this type of event, and the factor of this representative event is highlighted, including: the interface module presents the factor of this representative event as a scenario element affecting the current event through the central control screen, visualizes it to the driver, and then returns to step 1 to repeat the above steps to ensure real-time updating of environmental data, thereby realizing real-time targeted and focused presentation; wherein: the same type of event refers to the same type of event that is repeatedly experienced in a journey divided into 100 meters.

2. The method for improving driving confidence based on optimizing the presentation of interactive interface scenario elements according to claim 1, characterized in that: Step 4 is achieved by: Step 4.1: According to the itinerary S g,K and the factor c for each event on this itinerary i,k The dot product of the event is used to obtain the probability of each event. The probability of an event is calculated by using the dot product of the event factor and the trip: Step 4.2: Take the maximum value among the probabilities of similar events. That is, there are two left-turn events, which are called similar events. Take the one with the largest probability value as its representative event: AND' i =softmax(X′ i ) (19) Step 4.3: After obtaining the representative event of each event in step 4.2, the factors of the representative event are presented: AND' i =[y′1,y′2,...y′ M ] (20) y′ m (1≤m≤M)——is the mth influencing factor of the representative event to be presented; Step 4.4: After highlighting the influencing factors representing the events, real-time updates are required, and the process returns to step 1, and continues in sequence to achieve continuous dissemination and updates, and to present the influencing factors (scenario-aware elements) in a targeted manner, thereby optimizing the complex and unfocused scenario elements in the interactive interface.

Images