WO2019047593A1

WO2019047593A1 - Method and device for processing automatic driving training data

Info

Publication number: WO2019047593A1
Application number: PCT/CN2018/093353
Authority: WO
Inventors: 姜雨; 郁浩; 闫泳杉; 郑超; 唐坤; 张云飞
Original assignee: 百度在线网络技术（北京）有限公司
Priority date: 2017-09-05
Filing date: 2018-06-28
Publication date: 2019-03-14
Also published as: CN107784709A

Abstract

The goal of the present invention is to provide a method and device for processing automatic driving training data. The method for processing automatic driving training data according to the present invention comprises: providing a method for processing automatic driving training data, the method comprising: obtaining information related to data filtration and automatic driving training data at a plurality of moments; according to the information related to data filtration, determining whether a driving behavior indicated by automatic driving training data at each moment is abnormal; filtering out automatic driving training data corresponding to an abnormal driving behavior. According to the technology of the present invention, data causing automatic driving abnormalities in automatic driving training data may be filtered out to ensure the stability and safety of automatic driving.

Description

Method and apparatus for processing automatic driving training data

This patent application claims the Chinese patent application filed on September 5, 2017, the application number is 201710792053.5, the applicant is Baidu Online Network Technology (Beijing) Co., Ltd., and the invention is entitled "Method and Device for Processing Autopilot Training Data". Priority is hereby incorporated by reference in its entirety in its entirety.

Technical field

The present invention relates to the field of automatic driving technology, and in particular, to a technology for processing automatic driving training data.

Background technique

Autonomous driving is a smart car that is driven by a computer system. Automated driving vehicles rely on artificial intelligence, visual computing, radar, surveillance devices and global positioning systems to work together to allow the computer to operate the vehicle automatically and safely without any human active operation.

Driving scenarios in a variety of environments are complex and unpredictable, which is a difficult problem with autonomous driving. Therefore, it is necessary to integrate sensing, positioning, decision making and planning by combining data from multiple sensors. Among them, "decision" and "planning" have always been difficult issues. The deep learning training of autonomous driving is suitable for solving the tasks of "decision making" and "planning" in autonomous driving. Training requires first collecting driving scenarios and making driving decisions, and establishing a database of training data. Some of the autopilot training data comes from the actual driving environment of the driver. Some of the collected training data are “dirty data”, which may cause autopilot instability. These “dirty data” are not abnormal data caused by the environment or equipment reliability, but the training data that will normally occur during the driver's driving process, such as the driver entering the rest stop and taking a break, the driver changing lanes at will. Generally, due to the high proportion of qualified data, in many cases these "dirty data" will not affect the autonomous driving effect. However, due to the existence of “dirty data”, some abnormal behaviors may occur occasionally in automatic driving. Autopilot has a very high safety requirement, so how to filter out the “dirty data” in the autopilot training data is a very worthwhile research topic.

Summary of the invention

In accordance with an embodiment of the present invention, it is desirable to provide a method and apparatus for processing autopilot training data such that data indicative of autopilot anomalies in autopilot training data can be filtered out to ensure stability and safety of autopilot.

According to an embodiment of the first aspect of the present invention, a method of processing autopilot training data is provided, the method comprising:

a. acquiring data filtering related information and automatic driving training data at multiple times;

b. judging whether the driving behavior represented by the automatic driving training data at each moment is abnormal according to the data filtering related information;

c. Filter the autopilot training data corresponding to the abnormal driving behavior.

Specifically, step b includes:

- establishing an observation state sequence based on the hidden behavior defined by the hidden Markov model;

- Establishing a hidden Markov model of the driving behavior state, the driving behavior states include: normal driving state and abnormal driving state;

- According to the data filtering related information and the hidden Markov model, it is judged whether the observation state sequence at each moment corresponds to an abnormal driving state.

Specifically, the steps of establishing a hidden Markov model of the driving behavior state include:

- establishing a sample library of hidden Markov models;

- determining an observation state sequence trained by the hidden Markov model at each moment according to the data filtering related information in the sample library;

- According to the data in the sample library and the observation state sequence trained by the hidden Markov model, the hidden Markov model is trained to determine the parameters of the hidden Markov model.

Specifically, step b further includes:

The parameters of the hidden Markov model are updated based on whether or not the observed state sequence at each time corresponds to the judgment result of the abnormal driving state.

Specifically, step b includes:

- obtaining traffic signal information;

- judging whether the driving behavior at the corresponding moment violates the traffic rules according to the data filtering related information and the traffic signal information;

- Determining driving behavior in violation of traffic rules is abnormal driving behavior.

Specifically, the step of acquiring traffic signal information further includes:

Receiving a wireless signal transmitted by the traffic signal indicator to obtain traffic signal information.

Specifically, step b includes:

- Determining the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.

According to an embodiment of the second aspect of the present invention, an apparatus for processing autopilot training data is provided, the apparatus comprising:

a data acquisition unit configured to acquire data filtering related information and automatic driving training data at a plurality of times;

An abnormality determining unit configured to determine whether the driving behavior indicated by the automatic driving training data at each moment is abnormal according to the data filtering related information;

- A data filtering unit configured to filter autopilot training data corresponding to abnormal driving behavior.

Specifically, the abnormality determining unit includes:

An observation state establishing module configured to establish an observation state sequence of driving behavior defined by a hidden Markov model;

a model building module configured to establish a hidden Markov model of driving behavior states, the driving behavior states including: a normal driving state and an abnormal driving state;

The state determination module is configured to determine whether the sequence of observation states at each time corresponds to an abnormal driving state based on the data filtering related information and the hidden Markov model.

Specifically, the model building module further includes:

- a sample library creation sub-module configured to build a sample library of hidden Markov models;

a first state establishing submodule configured to determine an observation state sequence trained by the hidden Markov model at each moment according to the data filtering related information in the sample library;

The model training sub-module is configured to filter the related information according to the data in the sample library and the observation state sequence trained by the hidden Markov model, and train the hidden Markov model to determine the parameters of the hidden Markov model.

Specifically, the abnormality determining unit further includes:

The parameter update module is configured to update the parameters of the hidden Markov model according to whether the sequence of observation states at each time corresponds to the judgment result of the abnormal driving state.

Specifically, the abnormality determining unit includes:

a traffic signal acquisition module configured to obtain traffic signal information;

a traffic rule judging module configured to determine, according to the data filtering related information and the traffic signal information, whether the driving behavior at the corresponding moment violates the traffic rule;

- The first abnormality determining module configured to determine that the driving behavior when the traffic rule is violated is an abnormal driving behavior.

Specifically, the traffic signal acquisition module further includes:

a signal receiving submodule configured to receive a wireless signal transmitted by the traffic signal indicator to obtain traffic signal information.

Specifically, the abnormality determining unit includes:

a second abnormality determining module configured to determine that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.

According to an embodiment of the third aspect of the present invention, there is provided a computer apparatus comprising: one or more processors; a memory for storing one or more programs when one or more programs are one or more The processor, when executed, causes one or more processors to perform the method of processing the autonomous driving training data as previously described.

According to an embodiment of the fourth aspect of the present invention, there is provided a computer readable storage medium having stored thereon a computer program, wherein the computer program is executed by a processor to implement the aforementioned method of processing automatic driving training data .

According to an embodiment of the fifth aspect of the present invention, there is provided a computer program product for implementing the aforementioned method of processing autopilot training data when the computer program product is executed by a computer device.

The autopilot training data filtered as above may be output by any suitable output device, including but not limited to a computer display, projector, printer, for control of an automated driving process or other suitable operation.

Compared with the prior art, the embodiment of the present invention has the advantage of improving the safety and stability of the automatic driving by filtering out the data causing the autopilot instability in the autopilot training data. The method for filtering out unstable data in the automatic driving training data using the hidden Markov model of the embodiment of the present invention creatively applies the hidden Markov model to the filtering technical solution of the automatic driving training data, not only You can make full use of the existing training data to get more accurate filtering results, and iteratively update according to the latest training data to adapt to more complex autonomous driving scenarios and environments.

DRAWINGS

Other features, objects, and advantages of the present invention will become apparent from the Detailed Description of Description

1 shows a block diagram of an exemplary computer system/server suitable for implementing embodiments of the present invention;

2 is a flow chart showing a method of processing automatic driving training data according to an embodiment of the present invention;

3 is a flow chart showing whether a driving behavior is abnormal by a hidden Markov model based on an embodiment of the present invention;

4 is a flow chart showing a hidden Markov model for establishing a driving behavior state based on an embodiment of the present invention;

FIG. 5 is a flow chart showing whether the driving behavior is abnormal by the traffic signal information based on the embodiment of the present invention.

6 is a schematic diagram of an apparatus for processing automatic driving training data based on a preferred embodiment of the present invention;

7 is a schematic diagram of an abnormality determining unit that determines whether a driving behavior is abnormal by a hidden Markov model, according to an embodiment of the present invention;

8 is a schematic diagram of a model building module based on an embodiment of the present invention;

Fig. 9 is a schematic diagram of an abnormality determining unit that judges whether or not the driving behavior is abnormal by the traffic signal information, based on the embodiment of the present invention.

The same or similar reference numerals in the drawings denote the same or similar components.

Detailed ways

Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as a process or method depicted as a flowchart. Although the flowcharts describe various operations as a sequential process, many of the operations can be implemented in parallel, concurrently or concurrently. In addition, the order of operations can be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The processing may correspond to methods, functions, procedures, subroutines, subroutines, and the like.

By "computer device", also referred to as "computer" in the context, is meant an intelligent electronic device that can perform predetermined processing, such as numerical calculations and/or logical calculations, by running a predetermined program or instruction, which can include a processor and The memory is executed by the processor to execute a predetermined process pre-stored in the memory to execute a predetermined process, or is executed by hardware such as an ASIC, an FPGA, a DSP, or the like, or a combination of the two. Computer devices include, but are not limited to, servers, personal computers, notebook computers, tablets, smart phones, and the like.

The computer device includes a user device and a network device. The user equipment includes, but is not limited to, a computer, a smart phone, a PDA, etc.; the network device includes but is not limited to a single network server, a server group composed of multiple network servers, or a cloud computing based computer Or a cloud composed of a network server, wherein cloud computing is a type of distributed computing, a super virtual computer composed of a group of loosely coupled computers. Wherein, the computer device can be operated separately to implement the present invention, and can also access the network and implement the present invention by interacting with other computer devices in the network. The network in which the computer device is located includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, and the like.

It should be noted that the user equipment, the network equipment, the network, and the like are merely examples, and other existing or future possible computer equipment or networks, such as those applicable to the present invention, are also included in the scope of the present invention. It is included here by reference.

The methods discussed below, some of which are illustrated by flowcharts, can be implemented in hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware or microcode, the program code or code segments to carry out the necessary tasks can be stored in a machine or computer readable medium, such as a storage medium. The processor(s) can perform the necessary tasks.

The specific structural and functional details disclosed are merely representative and are for the purpose of describing exemplary embodiments of the invention. The present invention may, however, be embodied in many alternative forms and should not be construed as being limited only to the embodiments set forth herein.

It will be understood that when a unit is referred to as "connected" or "coupled" to another unit, it can be directly connected or coupled to the other unit, or an intermediate unit can be present. In contrast, when a unit is referred to as being "directly connected" or "directly coupled" to another unit, there is no intermediate unit. Other words used to describe the relationship between the units should be interpreted in a similar manner (eg "between" and "directly between" and "adjacent to" Than "directly adjacent to", etc.).

It should be understood that although the terms "first," "second," etc. may be used herein to describe the various elements, these elements should not be limited by these terms. These terms are used only to distinguish one unit from another. For example, a first unit could be termed a second unit, and similarly a second unit could be termed a first unit, without departing from the scope of the exemplary embodiments. The term "and/or" used herein includes any combination of one or more of the associated listed items.

The terminology used herein is for the purpose of describing the particular embodiments, The singular forms "a", "an", It is also to be understood that the terms "comprising" and """ Other features, integers, steps, operations, units, components, and/or combinations thereof.

It should also be noted that in some alternative implementations, the functions/acts noted may occur in a different order than that illustrated in the drawings. For example, two figures shown in succession may in fact be executed substantially concurrently or sometimes in the reverse order, depending on the function/acts involved.

The invention is further described in detail below with reference to the accompanying drawings.

FIG. 1 illustrates a block diagram of an exemplary computer system/server suitable for use in implementing embodiments of the present invention. The computer system/server 12 shown in FIG. 1 is merely an example and should not impose any limitation on the function and scope of use of the embodiments of the present invention.

As shown in FIG. 1, computer system/server 12 is embodied in the form of a general purpose computing device. The components of computer system/server 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and bus 18 that connects different system components, including system memory 28 and processing unit 16.

Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of a variety of bus structures. For example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA Bus, a Video Electronics Standards Association (VESA) local bus, and peripheral component interconnects ( PCI) bus.

Computer system/server 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer system/server 12, including both volatile and non-volatile media, removable and non-removable media.

Memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32. Computer system/server 12 may further include other removable/non-removable, volatile/non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (not shown in FIG. 1, commonly referred to as "hard disk drives"). Although not shown in FIG. 1, a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk"), and a removable non-volatile disk (such as a CD-ROM, DVD-ROM) may be provided. Or other optical media) read and write optical drive. In these cases, each drive can be coupled to bus 18 via one or more data medium interfaces. Memory 28 can include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

A program/utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28, such program modules 42 including, but not limited to, an operating system, one or more applications, other programs Modules and program data, each of these examples or some combination may include an implementation of a network environment. Program module 42 typically performs the functions and/or methods of the described embodiments of the present invention.

Computer system/server 12 may also be in communication with one or more external devices 14 (e.g., a keyboard, pointing device, display 24, etc.), and may also be in communication with one or more devices that enable a user to interact with the computer system/server 12. And/or in communication with any device (e.g., network card, modem, etc.) that enables the computer system/server 12 to communicate with one or more other computing devices. This communication can take place via an input/output (I/O) interface 22. Also, computer system/server 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through network adapter 20. As shown, network adapter 20 communicates with other modules of computer system/server 12 via bus 18. It should be understood that although not shown in FIG. 1, other hardware and/or software modules may be utilized in conjunction with computer system/server 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems. , tape drives, and data backup storage systems.

Processing unit 16 executes various functional applications and data processing by running programs stored in memory 28.

For example, the memory 28 stores therein a computer program for performing the functions and processes of the present invention, and when the processing unit 16 executes the corresponding computer program, the identification of the incoming call intention at the network side by the present invention is implemented.

The specific functions/steps of the present invention for determining autopilot training data are described in detail below.

2 is a flow chart showing a method of processing automatic driving training data according to an embodiment of the present invention. The method of an embodiment of the present invention is for processing autopilot training data, which can be implemented by an electronic device. Electronic devices include, but are not limited to, computer devices and automotive electronics. A computer device refers to an intelligent electronic device that can perform a predetermined process such as numerical calculation and/or logic calculation by running a predetermined program or instruction, which can include a processor and a memory, which are executed by the processor to execute a pre-stored survival instruction in the memory. The predetermined processing is performed by a hardware such as an ASIC, an FPGA, a DSP, or the like, or a combination of the two. Computer devices include, but are not limited to, servers, personal computers, notebook computers, tablets, smart phones, and the like. The server includes, but is not limited to, a single server, a plurality of servers, or a cloud-based cloud composed of a large number of computers or servers, wherein the cloud computing is a type of distributed computing, a super virtual composed of a group of loosely coupled computers. computer. Automotive electronics are electronic devices used in or related to automobiles.

As shown in FIG. 2, the method of processing the automatic driving training data according to the present embodiment includes step S1, step S2, and step S3.

First, in step S1, data filtering related information and automatic driving training data at a plurality of times are acquired. Autopilot training data refers to data used for autonomous driving training or learning. The data may be collected by various sensors, monitors or radars on the vehicle during driving of the vehicle, combined with navigation maps, etc., and may be data generated by various sensors, monitors or radars in a simulated driving environment. It can also be data generated in software or in a database. The data filtering related information refers to information required for filtering training data that causes autopilot instability in the autopilot training data, including the following information: GPS data, map information, and vehicle control parameter information. GPS data refers to data information acquired by satellite communication with a satellite positioning navigation system, including time and location information of the vehicle. In the present invention, the GPS data is not limited to data information acquired from a GPS (Global Position System), and other similar navigation systems from the Beidou navigation satellite system, the Galileo satellite positioning system, or currently existing or future appearances. Information such as time and location obtained in the present invention is applicable to the present invention and is also included in the scope of the present invention and is incorporated herein by reference. The map information refers to the data information obtained from the navigation map, and includes at least information such as the location of the road on which the vehicle is driven, the shape of the road, the attribute of the road, the condition of the road, the intersection of the road, the direction of the passage, and the like. The vehicle control parameter information refers to data of control parameters generated when the driver drives the vehicle, such as the speed of the vehicle, the steering wheel angle of the vehicle, the vehicle turn signal, and the like, which are generated when the vehicle is in a control operation such as constant speed running, acceleration, braking, steering, and the like. The GPS data and map information in the data filtering related information can be obtained from a map navigation device with a GPS receiving device, and the vehicle control parameter information can be acquired by a sensor installed on the vehicle, or through the vehicle trajectory and map at each moment. The information is obtained together after the calculation. For example, knowing the position of the vehicle at each moment, according to the map information, it can be known that the vehicles between the two moments are traveling straight and knowing the distance between the two time points of the vehicle, so that the speed of the vehicle can be obtained.

The data filtering related information and the automatic driving training data have multiple time point information, and each time point corresponds to a set of data filtering related information and automatic driving training data. The time point information may be GPS time or time point information from the GPS time, or may be time point information of other time sources. These time points may be fixed at a time interval, such as 1/40 s, or may be an interval that is not fixed or scattered. For example, in a driving trajectory, the frequency of the starting or stopping segment is 25 times per second, in the middle. The frequency of the uniform driving section is 5 times per second.

The autopilot training data and the data filtering related information may include driving data of a driving trajectory, and may also include driving data of a plurality of driving trajectories. The autopilot training data contains a lot of data information. If the autopilot training data contains data that filters out relevant information, the data filtering related information may also come from the autopilot training data.

In step S2, based on the data filtering related information, it is judged whether or not the driving behavior indicated by the automatic driving training data at each time is abnormal. Unstable training data for autonomous driving training does not refer to abnormal data with abnormal data or irrational data. For example, in a training data with a constant vehicle speed, a sudden or large vehicle speed suddenly appears, because the speed of the vehicle cannot be changed so drastically. Therefore, such data is erroneous data or unreasonable abnormal data appearing in the data acquisition process, and can be eliminated by filtering the data. However, such data is not training data that is desired to be filtered out in embodiments of the present invention. What is filtered out in the embodiment of the present invention is autopilot training data that makes autopilot unstable. For example, when the autopilot training data is the data of the actual driving of the vehicle directly collected, the driver refuels during the driving process or wants to stop and stop in the middle of the driving. Therefore, the training data of the vehicle deviating from the route may appear in the training data; if the driver and others When you call or chat, you will lose your concentration, so that the driving data will show the training data of the emergency brake. If these data are used for deep learning or training of autonomous driving, it may cause sudden braking or off-route during the automatic driving when the automatic driving function is activated. Therefore, it is necessary to find a way to filter out to improve the stability of autonomous driving.

Specifically, in step S2, the relevant information is filtered based on the data of each time, and the driving behavior of the driver corresponding to the time is judged to determine whether the driving behavior is an abnormal driving behavior of the automatic driving training. The abnormal driving behavior of the automatic driving training refers to the driving behavior that affects the automatic driving and makes the automatic driving unstable. Abnormal driving behaviors of autonomous driving training include: abnormal acceleration, abnormal braking, abnormal steering, and deviation from the route. Here, an abnormality in abnormal driving behavior such as "abnormal acceleration", "abnormal brake", or "abnormal steering" refers to an abnormal driving behavior that is not suitable for automatic driving training. The way to judge the driver's behavior includes: the data filtering related information can be input into the graphic simulation software of the data, and the amplitude of each data of the vehicle control parameter information is graphically displayed, and the amplitude changes more than two times before and after each data The threshold value, at this time, if the automatic driving training data shows that the environment of the vehicle is not abnormal, the corresponding driver behavior is determined to be abnormal driving behavior. In addition, by manually combining the contents of the automatic driving training data, the abnormal points whose amplitude changes of the respective data in the vehicle control parameter information exceed the threshold value are manually marked, and the driving behavior at the corresponding time is marked as abnormal driving behavior.

3 is a flow chart showing whether a driving behavior is abnormal by a hidden Markov model according to an embodiment of the present invention. As shown in FIG. 3, in a preferred embodiment, step S2 includes step S21, step S22, and step S23.

In step S21, an observation state sequence of the driving behavior defined based on the hidden Markov model is established. The observation state sequence of the driving behavior refers to a state that can be observed in the hidden Markov model, and in the present invention, the driving behavior is regarded as an observable state of the hidden Markov model. The sequence of observation states of driving behavior includes the following observable states: normal driving state, rapid acceleration state, sudden braking state, sharp turning state, and deviation from the route state. The data filtering related information is sampled at the time, and the vehicle control information and the map information corresponding to each time can be obtained. The observable state of the driving behavior at each moment can be obtained based on the vehicle control information and the map information. For example, according to the vehicle control information, the speed v1 of the vehicle at time t1 and the steering angle k1 of the vehicle can be known, and then the acceleration of the speed of the vehicle is increased according to the speed v0 of the vehicle at the previous moment a1=(v1-v0)/(t1-t0) If the value of a1 exceeds the preset acceleration threshold, the observation state at the current time is the rapid acceleration state; the acceleration of the vehicle speed decreases b1=(v0-v1)/(t1-t0), if the value of b1 exceeds The preset deceleration threshold, the current state of observation is the sudden braking state; according to the steering wheel angle k0 of the vehicle at the previous moment, the angle of change of the steering angle of the vehicle c1=(k1-k0)/(t1-t0) If the value of c1 exceeds the preset corner threshold, the observation state of the current moment is a sharp turn state; according to the map information, the position of the vehicle travel trajectory can be known, if the current vehicle trajectory and the vehicle information of the detachment on the map information If the distance between the positions of the trajectories exceeds the position threshold, the observation state at the current time is the deviation from the route state; if the vehicle is not in the rapid acceleration state, the sudden braking state, the sharp turning state, and the deviation route state, I think the current state of the observation time of the vehicle normal driving conditions. The driving state of each vehicle at each of the above points can be formed into an observable state chain. For example, the observation state of the driving behavior at time 1 is the normal driving state of the vehicle, the observation state at time 2 is the sharp turning state of the vehicle, and the observation state at time 3 is the vehicle deviation from the route state, then the normal driving state of the vehicle -> the vehicle sharp turning state - > The vehicle deviates from the route state to form an observable state chain of time 1 -> time 3. The above observable states can be encoded for easy implementation and understanding. It should be noted that the above content regarding the observable state of obtaining driving behavior is only an example, and other existing or future possible observable state of obtaining driving behavior may be applicable to the present invention and should also be included in the present invention. It is within the scope of the invention and is hereby incorporated by reference.

In step S22, a hidden Markov model of the driving behavior state is established, and the driving behavior state includes a normal driving state and an abnormal driving state. The normal driving state and the abnormal driving state are two hidden states in the hidden Markov model. The process of establishing a hidden Markov model of driving behavior state is mainly to determine the parameters of the hidden Markov model, namely the initial probability vector, the state transition probability matrix and the observation probability matrix. In step S22, the hidden Markov model parameters may be determined by manually labeling the data. This method of manually labeling data requires a large amount of manual labeling of training data.

4 is a flow chart showing a hidden Markov model for establishing a driving behavior state in accordance with a preferred embodiment of the present invention. As shown in Figure 4, in the preferred embodiment, a hidden Markov model is built by means of model learning or model training. The step of establishing a hidden Markov model of the driving behavior state includes step S31, step S32 and step S33.

In step S31, a sample library of the hidden Markov model is established. The sample library of the hidden Markov model is a sample database required for hidden Markov model learning, and includes data filtering information related to time information of one or more driving track segments.

In step S32, the observation state sequence trained by the hidden Markov model at each moment is determined according to the data filtering related information in the sample library. Because the sample library of the model includes multi-segment data filtering related information, the data filtering related information of each driving track segment is processed to obtain a sequence of observation states corresponding to a plurality of time points. The sequence of observation states of a plurality of time points of a driving track segment is connected into an observation state sequence chain. The length of the observed state chain may be the number of all time points contained in a driving trajectory, or may be a value determined according to the capabilities of the device or system. The length of the observable state chain is not limited in embodiments of the invention.

In step S33, the hidden Markov model is trained according to the data filtering related information in the sample library and the observation state sequence trained by the hidden Markov model, and the parameters of the hidden Markov model are determined. Determining the parameters of the hidden Markov model is also determining the initial probability vector, state transition probability matrix and observation probability matrix of the model. The initial probability vector includes the probability that the initial moment is in the normal driving state and the initial moment is in the abnormal driving state. The state transition probability matrix contains the probability of mutual transfer between the normal driving state and the abnormal driving state. The observation probability matrix contains the probability of generating various observation states under normal driving conditions and abnormal driving conditions. The hidden Markov model is trained by filtering relevant data of a driving trajectory in the sample library to: in the driving trajectory segment, first obtaining the value of the observed state sequence at each moment according to step S32, for example, according to the vehicle The control parameter determines that time 1 is the normal running state, time 2 is the acceleration state, time 3 is the normal running state, and time 4 is the braking state... and then estimating the parameters of the hidden Markov model based on the data filtering related information (initial probability vector, The state transition probability matrix and the observation probability matrix) make the probability of the observed sequence obtained under the model parameters the largest. The algorithm for estimating the parameters of the hidden Markov model can use the maximum likelihood method, or the Baum-Welch algorithm or other algorithms can be used for training learning. The specific algorithm for estimating the parameters of the hidden Markov model is not limited in the present invention. The hidden Markov model of the driving behavior state is trained by filtering the relevant information of the data of the plurality of driving trajectories in the sample library by the above method. The parameters of the hidden Markov model acquired for each training are processed to determine the parameter values of the final hidden Markov model. For example, according to the multi-segment model training data, the transition probabilities P ₁ , P ₂ , P ₃ , ..., P _I (I represents the total number of segments of the model training data) of the normal driving state of each piece of data to the normal driving state are obtained, and finally The transition probability P of the normal driving state to the normal driving state in the state transition probability matrix is the average value of P ₁ , P ₂ , P ₃ , ..., P _I .

In step S23, based on the data filtering related information and the hidden Markov model, it is judged whether or not the observation state sequence at each time corresponds to the abnormal driving state. Specifically, according to steps S21 and S22, the parameters of the hidden Markov model and the values of the observed state chain corresponding to the currently required autopilot training data have been obtained, and thus can be based on the obtained observed state chain. The value and hidden Markov model predict whether the driving behavior state corresponding to the observation state at each moment is a normal driving state or an abnormal driving state. The prediction algorithm can use an approximation algorithm or a Viterbi algorithm. Which prediction algorithm is used is not limited in the embodiment of the present invention.

In another preferred embodiment, the step S2 of determining whether the driving behavior is abnormal includes the step S24 (not shown): updating the hidden Markov according to whether the observation state sequence of each time corresponds to the judgment result of the abnormal driving state. The parameters of the model. After the results of the driving behavior states corresponding to the observation state sequence at the respective times are acquired according to steps S21, S22, and S23, the results and the values of the observed state sequences at the respective times are applied to the hidden behavior of the driving behavior state. In the model, the value of the hidden Markov model parameters is obtained, and the new hidden Markov model parameters are processed together with the hidden Markov model parameters, such as averaging, and the processed values are used instead of the original Hidden Markov model parameters.

The electronic device of the embodiment of the present invention uses a hidden Markov model to filter out unstable data in the automatic driving training data. This method is an innovative application of Hidden Markov Models for the anomaly data filtering in autonomous driving training data. The method creatively applies the hidden Markov model to the filtering technical solution of the automatic driving training data, and can not only make full use of the existing training data, but also obtain more accurate filtering results, and can be based on the latest training data. Iteratively updated to accommodate more complex autonomous driving scenarios and environments.

Fig. 5 is a flow chart showing the determination of whether the driving behavior is abnormal by the traffic signal information in the preferred embodiment of the present invention. As shown in FIG. 5, in the preferred embodiment, step S2 includes step S41, step S42, and step S43. In step S41, traffic signal information is acquired. The method of obtaining traffic signal information can be various. The traffic signal information can be determined from the acquired image information by image recognition technology by image recognition technology. The data of the traffic signal information can be obtained with the traffic dispatch center. In a preferred embodiment, step S41 includes receiving a wireless signal transmitted by the traffic signal indicator to obtain traffic signal information. Specifically, the wireless transmitting device is installed on the traffic signal indicating sign, and the result of the traffic signal or the change rule and the information such as the time point and the location of the traffic signal are transmitted through the wireless signal. Here, the wireless transmitting device is not limited to which wireless transceiver technology and wireless communication protocol or message format are used. The electronic device of the embodiment of the present invention may install a wireless receiving device matched with the transmitting portion, and directly receive traffic signal information sent by the traffic signal indicating flag. Or the other electronic device firstly receives the information sent by the traffic signal indicator, and then the electronic device in the embodiment of the present invention communicates with other electronic devices through wireless or wired manner to finally obtain the traffic signal information.

In step S42, it is determined whether the driving behavior at the corresponding time violates the traffic rule based on the data filtering related information and the traffic signal information. Specifically, according to the map information, the GPS information, and the traffic signal information in the data filtering related information, it is possible to know the location of the vehicle and the traffic sign near the area where the vehicle is located, such as the traffic light at the intersection where the vehicle is located or the traffic at the intersection where the vehicle is located. According to the requirements of speed limit or traffic direction, it is possible to judge whether the vehicle violates the illegal traffic rules such as traffic light signs, speeding or driving on an incorrect road according to the obtained traffic sign information and data filtering related information.

In step S43, it is determined that the driving behavior when the traffic rule is violated is an abnormal driving behavior. According to the judgment result of the step S42, if the driving behavior has no illegal traffic rules, the automatic driving training data does not need to be processed. If the driving behavior violates the traffic rules, the driving behavior at this time is determined as the abnormal driving behavior of the automatic driving training.

In another preferred embodiment of the present invention, step S2 includes determining that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior. When the vehicle is waiting for a red light, or when the vehicle stops before the "stop" traffic sign, or when the vehicle is stopped after avoiding pedestrians or obstacles, the vehicle is at a standstill; after the vehicle is still waiting for the condition to disappear, the vehicle has a slave The process of stationary to startup. Since the driver's own reaction time is different, the time between the start of the vehicle from the conditional permission and the actual start of the vehicle is different. Therefore, it is reflected in the automatic driving training data, and the length of the corresponding training data is different when the vehicle is stationary. The data of the vehicle when it is stationary is useless data for the automatic driving training, because the automatic driving equipment itself has its own processing reaction time, and the corresponding automatic driving training data when the vehicle before the vehicle starts in the training data is Abnormal data that needs to be filtered out, so it is determined that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.

In step S3, the automatic driving training data corresponding to the abnormal driving behavior is filtered. Specifically, the abnormal driving behavior of the automatic driving training has been determined in step S2, and the time corresponding to the abnormal driving behavior has also been clarified, and thus the automatic driving training data corresponding to these moments is the unnecessary automatic driving training data. . The filtering method is to directly delete the training data of the time corresponding to the abnormal driving behavior from the original training data, or replace the data without affecting the training.

FIG. 6 is a schematic flow chart of an apparatus for processing automatic driving training data according to an embodiment of the present invention. The apparatus of an embodiment of the present invention is for processing autopilot training data, which can be applied to an electronic device. Electronic devices include, but are not limited to, computer devices and automotive electronics. A computer device refers to an intelligent electronic device that can perform a predetermined process such as numerical calculation and/or logic calculation by running a predetermined program or instruction, which can include a processor and a memory, which are executed by the processor to execute a pre-stored survival instruction in the memory. The predetermined processing is performed by a hardware such as an ASIC, an FPGA, a DSP, or the like, or a combination of the two. Computer devices include, but are not limited to, servers, personal computers, notebook computers, tablets, smart phones, and the like. The server includes, but is not limited to, a single server, a plurality of servers, or a cloud-based cloud composed of a large number of computers or servers, wherein the cloud computing is a type of distributed computing, a super virtual composed of a group of loosely coupled computers. computer. Automotive electronics are electronic devices used in or related to automobiles.

As shown in FIG. 6, the apparatus for processing automatic driving training data according to the present embodiment includes a data acquiring unit 51, an abnormality determining unit 52, and a data filtering unit 53.

First, the data acquisition unit 51 is configured to acquire data filtering related information and automatic driving training data at a plurality of times. Autopilot training data refers to data used for autonomous driving training or learning. The data may be collected by various sensors, monitors or radars on the vehicle during driving of the vehicle, combined with navigation maps, etc., and may be data generated by various sensors, monitors or radars in a simulated driving environment. It can also be data generated in software or in a database. The data filtering related information refers to information required for filtering training data that causes autopilot instability in the autopilot training data, including the following information: GPS data, map information, and vehicle control parameter information. GPS data refers to data information acquired by satellite communication with a satellite positioning navigation system, including time and location information of the vehicle. In the present invention, the GPS data is not limited to data information acquired from a GPS (Global Position System), and other similar navigation systems from the Beidou navigation satellite system, the Galileo satellite positioning system, or currently existing or future appearances. Information such as time and location obtained in the present invention is applicable to the present invention and is also included in the scope of the present invention and is incorporated herein by reference. The map information refers to the data information obtained from the navigation map, and includes at least information such as the location of the road on which the vehicle is driven, the shape of the road, the attribute of the road, the condition of the road, the intersection of the road, the direction of the passage, and the like. The vehicle control parameter information refers to data of control parameters generated when the driver drives the vehicle, such as the speed of the vehicle, the steering wheel angle of the vehicle, the vehicle steering light, and the like, which are generated when the vehicle is in a control operation such as constant speed running, acceleration, braking, steering, and the like. The GPS data and map information in the data filtering related information can be obtained from a map navigation device with a GPS receiving device, and the vehicle control parameter information can be acquired by a sensor installed on the vehicle, or through the vehicle trajectory and map at each moment. The information is obtained together after the calculation. For example, knowing the position of the vehicle at each moment, according to the map information, it can be known that the vehicles between the two moments are traveling straight and knowing the distance between the two time points of the vehicle, so that the speed of the vehicle can be obtained.

The data filtering related information and the automatic driving training data have multiple time point information, and each time point corresponds to a set of data filtering related information and automatic driving training data. These time point information may be GPS time or time point information from GPS time, or may be time point information of other time sources. These time points may be fixed at a time interval, such as 1/40 s, or may be an interval that is not fixed or scattered. For example, in a driving trajectory, the frequency of the starting or stopping segment is 25 times per second, in the middle. The frequency of the uniform driving section is 5 times per second.

The abnormality determining unit 52 is configured to determine whether the driving behavior indicated by the automatic driving training data at each time is abnormal based on the data filtering related information. Unstable training data for autonomous driving training does not refer to abnormal data with abnormal data or irrational data. For example, in a training data with a constant vehicle speed, a sudden or large vehicle speed suddenly appears, because the speed of the vehicle cannot be changed so drastically. Therefore, such data is erroneous data or unreasonable abnormal data appearing in the data acquisition process, and can be eliminated by filtering the data. However, such data is not training data that is desired to be filtered out in embodiments of the present invention. What is filtered out in the embodiment of the present invention is autopilot training data that makes autopilot unstable. For example, when the autopilot training data is the data of the actual driving of the vehicle directly collected, the driver refuels during the driving process or wants to stop and stop in the middle of the driving. Therefore, the training data of the vehicle deviating from the route may appear in the training data; if the driver and others When you call or chat, you will lose your concentration, so that the driving data will show the training data of the emergency brake. If these data are used for deep learning or training of autonomous driving, it may cause sudden braking or off-route during the automatic driving when the automatic driving function is activated. Therefore, it is necessary to find a way to filter out to improve the stability of autonomous driving.

Specifically, the abnormality determining unit 52 determines the driving behavior of the driver corresponding to the time based on the data filtering related information at each time, and determines whether the driving behavior is an abnormal driving behavior of the automatic driving training. The abnormal driving behavior of the automatic driving training refers to the driving behavior that affects the automatic driving and makes the automatic driving unstable. Abnormal driving behaviors of autopilot training include: abnormal acceleration, abnormal braking, abnormal steering, and deviation from the route. Here, an abnormality in abnormal driving behavior such as "abnormal acceleration", "abnormal brake", or "abnormal steering" refers to an abnormal driving behavior that is not suitable for automatic driving training. The way to judge the driver's behavior includes: the data filtering related information can be input into the graphic simulation software of the data, and the amplitude of each data of the vehicle control parameter information is graphically displayed, and the amplitude changes more than two times before and after each data The threshold value, at this time, if the automatic driving training data shows that the environment of the vehicle is not abnormal, the corresponding driver behavior is determined to be abnormal driving behavior. In addition, by manually combining the contents of the automatic driving training data, the abnormal points whose amplitude changes of the respective data in the vehicle control parameter information exceed the threshold value are manually marked, and the driving behavior at the corresponding time is marked as abnormal driving behavior.

7 is a schematic diagram of an abnormality determining unit that determines whether a driving behavior is abnormal by a hidden Markov model, according to an embodiment of the present invention. In a preferred embodiment, the abnormality determining unit 52 includes an observation state establishing module 61, a model establishing module 62, and a state judging module 63, as shown in FIG.

The observation state establishing module 61 is configured to establish an observation state sequence based on the hidden behavior defined by the hidden Markov model. The observation state sequence of the driving behavior refers to a state that can be observed in the hidden Markov model, and in the present invention, the driving behavior is regarded as an observable state of the hidden Markov model. The sequence of observation states of driving behavior includes the following observable states: normal driving state, rapid acceleration state, sudden braking state, sharp turning state, and deviation from the route state. The data filtering related information is sampled at the time, and the vehicle control information and the map information corresponding to each time can be obtained. The observable state of the driving behavior at each moment can be obtained based on the vehicle control information and the map information. For example, according to the vehicle control information, the speed v1 of the vehicle at time t1 and the steering angle k1 of the vehicle can be known, and then the acceleration of the speed of the vehicle is increased according to the speed v0 of the vehicle at the previous moment a1=(v1-v0)/(t1-t0) If the value of a1 exceeds the preset acceleration threshold, the observation state at the current time is the rapid acceleration state; the acceleration of the vehicle speed decreases b1=(v0-v1)/(t1-t0), if the value of b1 exceeds The preset deceleration threshold, the current state of observation is the sudden braking state; according to the steering wheel angle k0 of the vehicle at the previous moment, the angle of change of the steering angle of the vehicle c1=(k1-k0)/(t1-t0) If the value of c1 exceeds the preset corner threshold, the observation state at the current moment is a sharp turn state; according to the map information, the position of the vehicle travel trajectory can be known, if the current vehicle trajectory and the vehicle information of the detachment on the map information If the distance between the positions of the trajectories exceeds the position threshold, the observation state at the current time is the deviation from the route state; if the vehicle is not in the rapid acceleration state, the sudden braking state, the sharp turning state, and the deviation route state, I think the current state of the observation time of the vehicle normal driving conditions. The driving state of each vehicle at each of the above points can be formed into an observable state chain. For example, the observation state of the driving behavior at time 1 is the normal driving state of the vehicle, the observation state at time 2 is the sharp turning state of the vehicle, and the observation state at time 3 is the vehicle deviation from the route state, then the normal driving state of the vehicle -> the vehicle sharp turning state - > The vehicle deviates from the route state to form an observable state chain of time 1 -> time 3. The above observable states can be encoded for easy implementation and understanding. It should be noted that the above observation state establishing module for obtaining an observable state of driving behavior is only an example, and other existing or future observation state establishing modules for obtaining an observable state of driving behavior may be applied to the present invention. It is also intended to be included within the scope of the invention and is hereby incorporated by reference.

The model building module 62 is configured to establish a hidden Markov model of the driving behavior state, and the driving behavior states include: a normal driving state and an abnormal driving state. The normal driving state and the abnormal driving state are two hidden states in the hidden Markov model. The process of establishing a hidden Markov model of driving behavior state is mainly to determine the parameters of the hidden Markov model, namely the initial probability vector, the state transition probability matrix and the observation probability matrix. The model building module 62 can determine the hidden Markov model parameters by manually annotating the data. This method of manually labeling data requires a large amount of manual labeling of training data.

8 is a schematic diagram of a model building block of a hidden Markov model for establishing a driving behavior state in accordance with a preferred embodiment of the present invention. In the preferred embodiment, the hidden Markov model is built by means of model learning or model training. The model building module 62 includes a sample library building sub-module 71, a first state building sub-module 72, and a model training sub-module 73, as shown in FIG.

The sample library creation sub-module 71 is configured to build a sample library of hidden Markov models. The sample library of the hidden Markov model is a sample database required for hidden Markov model learning, and includes data filtering information related to time information of one or more driving track segments.

The first state establishing sub-module 72 is configured to determine an observed state sequence trained by the hidden Markov model at each moment based on the data filtering related information in the sample library. Because the sample library of the model includes multi-segment data filtering related information, the data filtering related information of each driving track segment is processed to obtain a sequence of observation states corresponding to a plurality of time points. The sequence of observation states of a plurality of time points of a driving track segment is connected into an observation state sequence chain. The length of the observed state chain may be the number of all time points contained in a driving trajectory, or may be a value determined according to the capabilities of the device or system. The length of the observable state chain is not limited in embodiments of the invention.

The model training sub-module configuration 73 is to train the hidden Markov model according to the data filtering related information in the sample library and the observation state sequence trained by the hidden Markov model to determine the parameters of the hidden Markov model. Determining the parameters of the hidden Markov model is also determining the initial probability vector, state transition probability matrix and observation probability matrix of the model. The initial probability vector includes the probability that the initial moment is in the normal driving state and the initial moment is in the abnormal driving state. The state transition probability matrix contains the probability of mutual transfer between the normal driving state and the abnormal driving state. The observation probability matrix contains the probability of generating various observation states under normal driving conditions and abnormal driving conditions. The model training sub-module 73 filters the data of a driving trajectory in the sample library to filter the hidden Markov model: in the driving trajectory segment, the first state establishing sub-module 72 first acquires the observation at each moment. The value of the state sequence, for example, is determined according to the vehicle control parameter, time 1 is the normal driving state, time 2 is the acceleration state, time 3 is the normal driving state, and time 4 is the braking state...; then the model training sub-module 73 filters the data according to the data. The information estimates the parameters of the hidden Markov model (initial probability vector, state transition probability matrix and observation probability matrix) such that the probability of the observed sequence obtained under the model parameters is the largest. The algorithm for estimating the parameters of the hidden Markov model can use the maximum likelihood method, or the Baum-Welch algorithm or other algorithms can be used for training learning. The specific algorithm for estimating the parameters of the hidden Markov model is not limited in the present invention. The model training sub-module 73 trains the data of the plurality of driving trajectories in the sample library to filter the hidden Markov model of the driving behavior state by the above method. The parameters of the hidden Markov model acquired for each training are processed to determine the parameter values of the final hidden Markov model. For example, according to the multi-segment model training data, the transition probabilities P ₁ , P ₂ , P ₃ , ..., P _I (I represents the total number of segments of the model training data) of the normal driving state of each piece of data to the normal driving state are obtained, and finally The transition probability P of the normal driving state to the normal driving state in the state transition probability matrix is the average value of P ₁ , P ₂ , P ₃ , ..., P _I .

The state determination module 63 is configured to determine whether the observed state sequence at each time corresponds to an abnormal driving state based on the data filtering related information and the hidden Markov model. Specifically, according to the observation state establishing module 61 and the model establishing module 62, the parameters of the hidden Markov model and the values of the observation state chain corresponding to the currently required automatic driving training data are obtained, and thus the state judging module 63 According to the obtained value of the observation state chain and the hidden Markov model, it is predicted whether the driving behavior state corresponding to the observation state at each moment is a normal driving state or an abnormal driving state. The prediction algorithm can use an approximation algorithm or a Viterbi algorithm. Which prediction algorithm is used is not limited in the embodiment of the present invention.

In another preferred embodiment, the abnormality determining unit 52 further includes a parameter updating module 64 (not shown) configured to update the hidden Markov according to whether the observed state sequence of each time corresponds to the judgment result of the abnormal driving state. The parameters of the model. After the observation state establishing module 61, the model establishing module 62, and the state judging module 63 acquire the results of the driving behavior states corresponding to the observation state sequences at the respective moments, the parameter updating module 64 together the results and the values of the observed state sequences at the respective moments together. Applying to the hidden Markov model of the driving behavior state, obtaining the values of the parameters of the hidden Markov model, and processing the new hidden Markov model parameters together with the hidden Markov model parameters, for example, The average value is substituted for the hidden Markov model parameters with the processed values.

The apparatus of an embodiment of the present invention uses a hidden Markov model to filter out unstable data in the autonomous driving training data. This device is an innovative application of Hidden Markov Models for anomalous data filtering in autonomous driving training data. The device creatively applies the hidden Markov model to the filtering scheme of the automatic driving training data, and can not only make full use of the existing training data, but also obtain more accurate filtering results, and can be based on the latest training data. Iteratively updated to accommodate more complex autonomous driving scenarios and environments.

Fig. 9 is a view showing an abnormality determining unit for judging whether or not the driving behavior is abnormal by traffic signal information according to a preferred embodiment of the present invention. As shown in FIG. 9, in the preferred embodiment, the abnormality determining unit 52 includes a traffic signal acquiring module 81, a traffic rule determining module 82, and a first abnormality determining module 8. The traffic signal acquisition module 81 is configured to acquire traffic signal information. The method of obtaining traffic signal information can be various. The traffic signal acquisition module 81 may determine the traffic signal information according to the image recognition technology from the acquired image information by using an image recognition technology; and may acquire the data of the traffic signal information with the traffic dispatch center. In a preferred embodiment, the traffic signal acquisition module 81 further includes a signal receiving sub-module 84 (not shown) configured to receive the wireless signal transmitted by the traffic signal indicating flag to obtain traffic signal information. Specifically, the wireless transmitting device is installed on the traffic signal indicating sign, and the result of the traffic signal or the change rule and the information such as the time point and the location of the traffic signal are transmitted through the wireless signal. Here, the wireless transmitting device is not limited to which wireless transceiver technology and wireless communication protocol or message format are used. The traffic signal acquisition module 81 can install a wireless receiving device that matches the transmitting portion and directly receives the traffic signal information sent by the traffic signal indicator. Or the other electronic device first wirelessly receives the information sent by the traffic signal indicator, and then the traffic signal acquisition module 81 communicates with other electronic devices through wireless or wired manner to finally obtain the traffic signal information.

The traffic rule determination module 82 is configured to determine whether the driving behavior at the corresponding time violates the traffic rule based on the data filtering related information and the traffic signal information. Specifically, the traffic rule determination module 82 can know the location of the vehicle and the traffic sign near the area where the vehicle is located, such as the map information, the GPS information, and the traffic signal information in the data filtering related information, such as the traffic light at the intersection location of the vehicle or The traffic speed limit or the traffic direction of the intersection where the vehicle is located, so it is possible to judge whether the vehicle violates the traffic light sign, speeding or driving on an incorrect road according to the obtained traffic sign information and data filtering related information. The behavior of the rules.

The first abnormality determining module 83 is configured to determine that the driving behavior when the traffic rule is violated is an abnormal driving behavior. According to the judgment result of the traffic rule judging module 82, if the driving behavior has no illegal traffic rules, the autopilot training data does not need to be processed. If the driving behavior violates the traffic rules, the first abnormality determining module 83 determines the driving behavior at this time as the abnormal driving behavior of the automatic driving training.

In another preferred embodiment of the present invention, the abnormality determining unit 52 is configured to determine that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior. When the vehicle is waiting for a red light, or when the vehicle stops before the "stop" traffic sign, or when the vehicle is stopped after avoiding pedestrians or obstacles, the vehicle is at a standstill; after the vehicle is still waiting for the condition to disappear, the vehicle has a slave The process of stationary to startup. Since the driver's own reaction time is different, the time between the start of the vehicle from the conditional permission and the actual start of the vehicle is different. Therefore, it is reflected in the automatic driving training data, and the length of the corresponding training data is different when the vehicle is stationary. The data of the vehicle when it is stationary is useless data for the automatic driving training, because the automatic driving equipment itself has its own processing reaction time, and the corresponding automatic driving training data when the vehicle before the vehicle starts in the training data is The abnormality data that needs to be filtered out is determined, so the abnormality judging unit 52 determines that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.

The data filtering unit 53 is configured to filter the automatic driving training data corresponding to the abnormal driving behavior. Specifically, the abnormality determining unit 52 has determined the abnormal driving behavior of the automatic driving training, and the time corresponding to the abnormal driving behavior has also been clarified, and thus the automatic driving training data corresponding to these moments is the unnecessary automatic driving training data. . The data filtering unit 53 filters the data by directly deleting the training data of the time corresponding to the abnormal driving behavior from the original training data, or replacing the data without affecting the training.

It should be noted that the present invention can be implemented in software and/or a combination of software and hardware. For example, the various devices of the present invention can be implemented using an application specific integrated circuit (ASIC) or any other similar hardware device. In one embodiment, the software program of the present invention may be executed by a processor to implement the steps or functions described above. As such, the software programs (including related data structures) of the present invention can be stored in a computer readable medium. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive lists) of computer readable storage media include: electrical connections having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing. In this document, a computer readable storage medium can be any tangible medium that can contain or store a program, which can be used by or in connection with an instruction execution system, apparatus or device.

A computer readable signal medium may include a data signal that is propagated in the baseband or as part of a carrier, carrying computer readable program code. Such propagated data signals can take a variety of forms including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer readable signal medium can also be any computer readable medium other than a computer readable storage medium, which can transmit, propagate, or transport a program for use by or in connection with the instruction execution system, apparatus, or device. .

Program code embodied on a computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including an object oriented programming language such as Java, Smalltalk, C++, and conventional A procedural programming language - such as the "C" language or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer, partly on the remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (eg, using an Internet service provider) Internet connection).

Additionally, some of the steps or functions of the present invention may be implemented in hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the invention is defined by the appended claims instead All changes in the meaning and scope of equivalent elements are included in the present invention. Any reference signs in the claims should not be construed as limiting the claim. In addition, it is to be understood that the word "comprising" does not exclude other elements or steps. A plurality of units or devices recited in the system claims can also be implemented by a unit or device by software or hardware. The first, second, etc. words are used to denote names and do not denote any particular order.

Claims

A method of processing autonomous driving training data, comprising:

a. acquiring data filtering related information and the automatic driving training data at multiple moments;

b. determining, according to the data filtering related information, whether the driving behavior represented by the automatic driving training data at each moment is abnormal;

c. Filtering the autopilot training data corresponding to the abnormal driving behavior.
The method of claim 1 wherein said step b comprises:

- establishing an observation state sequence based on the hidden behavior defined by the hidden Markov model;

a hidden Markov model for establishing a driving behavior state, the driving behavior state comprising: a normal driving state and an abnormal driving state;

Determining, according to the data filtering related information and the hidden Markov model, whether the observed state sequence at each moment corresponds to the abnormal driving state.
The method of claim 2 wherein said step of establishing a hidden Markov model of the driving behavior state comprises:

- establishing a sample library of the hidden Markov model;

Determining an observation state sequence trained by the hidden Markov model at each moment according to the data filtering related information in the sample library;

- training the hidden Markov model according to the data filtering related information in the sample library and the observation state sequence trained by the hidden Markov model to determine parameters of the hidden Markov model.
The method of claim 3 wherein said step b further comprises:

- updating the parameters of the hidden Markov model according to whether the sequence of observation states at each time corresponds to the judgment result of the abnormal driving state.
The method according to any one of claims 2 to 4, wherein the sequence of observed states comprises: a normal driving state, a sudden acceleration state, a sudden braking state, a sharp turning state, and a deviation route state.
The method of claim 1 wherein said step b comprises:

- obtaining traffic signal information;

Determining, according to the data filtering related information and the traffic signal information, whether the driving behavior at the corresponding moment violates the traffic rule;

- Determining driving behavior in violation of traffic rules is abnormal driving behavior.
The method of claim 6, wherein the step of acquiring traffic signal information further comprises:

Receiving a wireless signal transmitted by the traffic signal indicator to obtain the traffic signal information.
The method of claim 1 wherein said step b comprises:

- Determining the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.
An apparatus for processing automatic driving training data, wherein the apparatus comprises:

a data acquisition unit configured to acquire data filtering related information and the automatic driving training data at a plurality of times;

An abnormality determining unit configured to determine whether the driving behavior represented by the automatic driving training data at each time is abnormal according to the data filtering related information;

a data filtering unit configured to filter the autopilot training data corresponding to the abnormal driving behavior.
The apparatus according to claim 9, wherein said abnormality determining unit comprises:

An observation state establishing module configured to establish an observation state sequence of driving behavior defined by a hidden Markov model;

a model building module configured to establish a hidden Markov model of a driving behavior state, the driving behavior state comprising: a normal driving state and an abnormal driving state;

a state determination module configured to determine whether the sequence of observed states at each time corresponds to the abnormal driving state based on the data filtering related information and the hidden Markov model.
The apparatus of claim 10, wherein the model building module further comprises:

a sample library creation submodule configured to create a sample library of the hidden Markov model;

a first state establishing submodule configured to determine an observed state sequence trained by the hidden Markov model at each moment according to the data filtering related information in the sample library;

a model training sub-module configured to train the hidden Markov model according to the data filtering related information in the sample library and the observation state sequence trained by the hidden Markov model to determine hidden Markov The parameters of the model.
The apparatus according to claim 11, wherein the abnormality determining unit further comprises:

a parameter update module configured to update parameters of the hidden Markov model according to whether the sequence of observed states at each time corresponds to a determination result of the abnormal driving state.
The apparatus according to any one of claims 10 to 12, wherein the sequence of observation states comprises: a normal running state, a sudden acceleration state, a sudden braking state, a sharp turning state, and a deviation course state.
The apparatus according to claim 9, wherein said abnormality determining unit comprises:

a traffic signal acquisition module configured to obtain traffic signal information;

a traffic rule judging module configured to determine, according to the data filtering related information and the traffic signal information, whether a driving behavior at a corresponding moment violates a traffic rule;

- The first abnormality determining module configured to determine that the driving behavior when the traffic rule is violated is an abnormal driving behavior.
The device of claim 14, wherein the traffic signal acquisition module further comprises:

a signal receiving submodule configured to receive a wireless signal transmitted by the traffic signal indicator to obtain the traffic signal information.
The apparatus according to claim 9, wherein said abnormality determining unit comprises:

a second abnormality determining module configured to determine that the driving behavior when the vehicle is stationary before the vehicle starts is an abnormal driving behavior.
A computer device, the computer device comprising:

- one or more processors;

- a memory for storing one or more programs,

- causing the one or more processors to perform the method of any of claims 1-8 when the one or more programs are executed by the one or more processors.
A computer readable storage medium storing computer code, the method of any one of claims 1 to 8 being executed when the computer code is executed.
A computer program product, when the computer program product is executed by a computer device, the method of any one of claims 1 to 8 being performed.