High-precision Map Perception Container Design Method and Device, Storage Medium, and Terminal

Abstract

A design method, device, storage medium, and terminal for a high-precision map perception container for multi-vehicle joint perception. The method includes: obtaining perception target indication information, which is used to indicate the environmental state information that an autonomous vehicle needs to perceive; obtaining a high-precision map, establishing a map perception container to extract features of at least a part of the perception targets in the map to obtain sensorized information, so as to generate a signal with the same format as that of an ordinary sensor according to the original map data. In addition, considering that some information in the high-precision map data has a much higher confidence than on-vehicle sensors, it is directly superimposed on the perception result. Finally, the multi-vehicle joint perception information is fused and perceived under a unified map spatio-temporal reference to determine the perception result of the autonomous vehicle; wherein, the positioning error of the map is within a preset error range. The present invention can effectively eliminate perception blind spots and improve perception accuracy.

Description

Technical Field

[0001] The present invention relates to the field of autonomous vehicle perception, and particularly to a high-precision map perception container design method, device, storage medium, and terminal for multi-vehicle joint perception. Background Art

[0002] There are bottleneck problems in single-vehicle environment perception, such as difficulty in eliminating blind spots and low perception accuracy. Vehicle-to-Everything (V2X) vehicle networking can enable communication between vehicles, between vehicles and people, between vehicles and roadside units, and between vehicles and cloud platforms, realizing information sharing and enabling autonomous driving to bid farewell to the era of single-vehicle intelligence.

[0003] In recent years, with the development of vehicle networking technology, the range of information that a single vehicle can obtain has increased significantly. By fusing the perception information of other vehicles in the vehicle network, vehicle perception occlusion and ultra-field-of-view blind spots can be effectively eliminated, and the perception ability of a single vehicle can be improved. In addition, prior information about the road environment can be obtained through an electronic map, which provides a technical basis for further improving the perception ability of a single vehicle by fusing external perception resources.

[0004] However, in existing autonomous vehicle perception technologies based on vehicle networking technology, the local coordinate system of the host vehicle is mostly used as the information fusion benchmark, which depends on the degree of overlap of the fields of view between multiple vehicles and fails to form an effective constraint on the vehicle pose. There are blind spots in single-vehicle perception, and it is difficult to eliminate them through the vision data of other connected vehicles.

[0005] In the prior art, there is research on multi-vehicle joint perception based on the Global Navigation Satellite System (GNSS) benchmark. However, in complex scenarios, especially in small-scale joint perception scenarios, the GNSS coordinate system is unstable and often has an overall drift in a certain direction, resulting in a decrease in the accuracy of networked fusion perception.

[0006] There is an urgent need for a high-precision map perception container design method for multi-vehicle joint perception, which can provide a unified benchmark for the information fusion of multi-vehicle joint perception. This benchmark does not drift with the degradation of GNSS signals and can provide a stable spatial reference benchmark for the results of multi-vehicle joint perception, effectively eliminating perception blind spots and improving perception accuracy. Summary of the Invention

[0007] The technical problem solved by the present invention is to provide a high-precision map perception container design method, device, storage medium, and terminal for multi-vehicle joint perception, which can effectively eliminate perception blind spots and improve perception accuracy.

[0008] To solve the above technical problems, an embodiment of the present invention provides a high-precision map perception container design method for multi-vehicle joint perception, including the following steps: obtaining perception target indication information, which is used to indicate the environmental state information that the autonomous vehicle needs to perceive; obtaining a high-precision map, establishing a map perception container to extract features of at least a part of the perception targets in the map to obtain sensorized information; extracting non-sensorized information in the high-precision map data other than the above sensorized information, superimposing it with ordinary sensor perception information and map sensor perception information, and performing fusion perception under a unified map spatio-temporal benchmark, and obtaining the environmental perception result of the autonomous vehicle by solving the state estimation problem; wherein, the high-precision map is a map with a positioning error within a preset error range.

[0009] Optionally, at least a part of the perception targets includes connected vehicles and other obstacles except the connected vehicles; wherein, the sensorized information includes connected vehicle detection information for detecting connected vehicles and non-connected vehicle detection information for detecting other obstacles except the connected vehicles.

[0010] Optionally, the non-connected vehicle detection information includes one or more of the following: dynamic obstacle information for detecting dynamic obstacles except the connected vehicles; map detection information for detecting static obstacles on the map.

[0011] Optionally, using a map sensor to extract features of at least a part of the perception targets in the map to obtain sensorized information includes: determining the position information of the perception targets to be perceived according to the perception target indication information; calculating the covariance of the position information of at least a part of the perception targets according to the accuracy of the map, and using it as the sensorized information.

[0012] Optionally, the step of extracting non-sensorized information in the high-precision map data other than the above sensorized information, superimposing it with ordinary sensor perception information and map sensor perception information, performing fusion perception under a unified map spatio-temporal benchmark, and obtaining the environmental perception result of the autonomous vehicle by solving the state estimation problem includes: obtaining sensor information detected by one or more sensors for at least a part of the perception targets; fusing the sensor information of at least a part of the perception targets and the sensorized information to obtain fusion information of each perception target in at least a part of the perception targets; determining an observation set of each perception target in at least a part of the perception targets according to the fusion information; using the maximum likelihood estimation algorithm to determine the state quantity estimation of each perception target in at least a part of the perception targets according to the observation set of each perception target in at least a part of the perception targets.

[0013] Optionally, the non-sensorized information in the extracted high-precision map data other than the above-mentioned sensorized information is superimposed with the ordinary sensor perception information and the map sensor perception information, and fused perception is performed under the unified map spatio-temporal reference. The environmental perception result of the autonomous vehicle obtained by solving the problem through state estimation further includes: superimposing the state quantity estimation of each perception target in at least a part of the perception targets with the non-sensorized information to obtain the perception result of the autonomous vehicle.

[0014] Optionally, the following formula is used to determine the state quantity estimation of each perception target in at least a part of the perception targets according to the set of observed quantities of each perception target in at least a part of the perception targets:

[0015]

[0016] where is used to represent the estimation result of the driving environment space state parameters in the sense of maximum likelihood, X D is used to represent the set of observed quantities within a certain time depth in the driving environment space, Z p is used to represent the set of measurements in the asynchronous heterogeneous perception space that can be obtained by the intelligent vehicle, P(Z p |X D ) is the conditional probability of the set of observed quantities Z D under the given state parameter X p .

[0017] To solve the above technical problems, an embodiment of the present invention provides a high-precision map perception container design device for multi-vehicle joint perception, including: an information acquisition module, configured to acquire perception target indication information, where the perception target indication information is used to indicate the environmental state information that the autonomous vehicle needs to perceive; a sensorized information determination module, configured to acquire a high-precision map, establish a map perception container to perform feature extraction on at least a part of the perception targets in the map to obtain sensorized information; a perception result solving module, configured to extract non-sensorized information in the high-precision map data other than the above-mentioned sensorized information, superimpose it with ordinary sensor perception information and map sensor perception information, perform fused perception under the unified map spatio-temporal reference, and obtain the environmental perception result of the autonomous vehicle by solving the problem through state estimation; where the high-precision map is a map with a positioning error within a preset error range.

[0018] To solve the above technical problems, an embodiment of the present invention provides a storage medium, on which a computer program is stored, and when the computer program is run by a processor, it executes the steps of the above-mentioned high-precision map perception container design method for multi-vehicle joint perception.

[0019] To solve the above technical problems, an embodiment of the present invention provides a terminal, including a memory and a processor, where a computer program capable of running on the processor is stored on the memory, and it is characterized in that when the processor runs the computer program, it executes the steps of the above-mentioned high-precision map perception container design method for multi-vehicle joint perception.

[0020] Compared with the prior art, the technical solution of the embodiment of the present invention has the following beneficial effects:

[0021] In the embodiment of the present invention, by obtaining perception target indication information, the perception target indication information is used to indicate the environmental state information that the autonomous vehicle needs to perceive; obtaining a high-precision map, establishing a map perception container to extract features of at least a part of the perception targets in the map to obtain sensorized information, so as to generate a signal with the same format as that of a general sensor according to the original map data. In addition, considering that some information in the high-precision map data has a much higher confidence than that of in-vehicle sensors, it is directly superimposed on the perception result. Finally, the multi-vehicle joint perception information (including general sensor perception information, map sensor perception information, and other non-sensorized data in the high-precision map) is fused and perceived under a unified map spatio-temporal reference to determine the perception result of the autonomous vehicle; wherein, the positioning error of the map is within a preset error range. The present invention can effectively eliminate perception blind spots and improve perception accuracy.

[0022] Furthermore, obtaining sensor information detected by one or more sensors for the at least a part of the perception targets, and then obtaining the fusion information and the set of observation quantities of each perception target in the at least a part of the perception targets, and then determining the state quantity estimation of each perception target in the at least a part of the perception targets, can realize determining the perception result based on the information fusion of the map sensor and the sensor, and further improve the perception accuracy.

[0023] Furthermore, non-sensorized information other than the sensorized information can also be extracted from the map, and then the state quantity estimation of each perception target in the at least a part of the perception targets is superimposed with the non-sensorized information to obtain the perception result of the autonomous vehicle, so that on the basis of the detection of multiple sensors, other information of the high-precision map can be further added, and the overall effect of perception can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flowchart of a high-precision map perception container design method for multi-vehicle joint perception in an embodiment of the present invention;

[0025] Figure 2 is Figure 1 a flowchart of a specific implementation manner of step S13 in

[0026] Figure 3 It is a schematic structural diagram of a high-precision map perception container design device for multi-vehicle joint perception in an embodiment of the present invention;

[0027] Figure 4 It is a schematic structural diagram of a perception system of a high-precision map perception container design method for multi-vehicle joint perception in an embodiment of the present invention. Detailed implementation manners

[0028] As described above, in the prior art, by fusing the perception information of other vehicles in the vehicle network, it is possible to effectively eliminate vehicle perception occlusion and ultra-field-of-view blind spots and improve the perception ability of a single vehicle. However, in the existing autonomous vehicle perception technology based on vehicle network technology, the local coordinate system of the host vehicle is mostly used as the information fusion benchmark, depending on the degree of overlap of the fields of view between multiple vehicles, and an effective constraint on the vehicle pose cannot be formed, resulting in the difficulty of meeting the requirements of intelligent connected vehicles for the joint perception accuracy based on mass-produced vehicle sensors, and the lack of consistency in the fusion results between multiple sensors.

[0029] The inventors of the present invention have found through research that in the existing scenarios of multi-vehicle joint perception based on the GNSS benchmark, due to the instability of the GNSS coordinate system, there is often an overall drift in a certain direction, resulting in a decrease in the accuracy of networked fusion perception. And the high-precision map data can provide high-precision absolute position constraints for vehicles. At the same time, the high-precision map data contains accurate road attributes and topologies and other information that is difficult for sensors to perceive.

[0030] In an embodiment of the present invention, perception target indication information is obtained, and the perception target indication information is used to indicate the environmental state information that the autonomous vehicle needs to perceive; a high-precision map is obtained, and a map perception container is established to perform feature extraction on at least a part of the perception targets in the map to obtain sensorized information, so as to generate a signal in the same format as that of a general sensor according to the original map data. In addition, considering that some information in the high-precision map data has a much higher confidence than in-vehicle sensors, it is directly superimposed on the perception result. Finally, the multi-vehicle joint perception information (including general sensor perception information, map sensor perception information, and other non-sensorized data in the high-precision map) is fused and perceived under a unified map spatio-temporal benchmark to determine the perception result of the autonomous vehicle; wherein, the positioning error of the map is within a preset error range. The present invention can effectively eliminate perception blind spots and improve perception accuracy.

[0031] To make the above objects, features, and beneficial effects of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings.

[0032] Refer to Figure 1 , Figure 1It is a flowchart of a design method for a high-precision map perception container for multi-vehicle collaborative perception in an embodiment of the present invention. The design method for the high-precision map perception container for multi-vehicle collaborative perception may include steps S11 to S13:

[0033] Step S11: Obtain perception target indication information, where the perception target indication information is used to indicate the environmental state information that the autonomous vehicle needs to perceive;

[0034] Step S12: Obtain a high-precision map, establish a map perception container to extract features of at least part of the perception targets in the map to obtain sensorized information;

[0035] Step S13: Extract non-sensorized information in the high-precision map data other than the above-mentioned sensorized information, superimpose it with ordinary sensor perception information and map sensor perception information, perform fusion perception under the unified map spatio-temporal reference, and obtain the environmental perception result of the autonomous vehicle by solving the state estimation problem.

[0036] Among them, the high-precision map is a map with a positioning error within a preset error range.

[0037] In the specific implementation of step S11, the perception target indication information can be determined by a pre-set method, that is, the preset perception requirement is determined, or the user input method can be adopted, so that the user's perception requirement can be determined.

[0038] Among them, the perception target indication information is used to indicate the environmental state information that the autonomous vehicle needs to perceive. The environmental state information may include the spatial range that the autonomous vehicle needs to perceive, and may also belong to one or more types of perception targets that the autonomous vehicle needs to perceive, such as buildings, vehicles, people, etc. that the autonomous vehicle needs to perceive.

[0039] In the specific implementation of step S12, a high-precision map can be set to be used.

[0040] Specifically, the positioning error of the map is within a preset error range. It can be understood that the smaller the error value in the preset error range, the higher the accuracy requirement for the map.

[0041] As a non-limiting example, the upper limit value of the error value in the preset error range can be selected from: 10 cm to 50 cm. For example, the upper limit value of the error value in the preset error range can be set to 20 cm, that is, the positioning error of the map is within 20 cm.

[0042] In the specific implementation, a map perception container can also be established.

[0043] Specifically, the map perception container is a theoretical model based on a high-precision map to solve the problem of joint perception of intelligent vehicles. Based on the information of different in-vehicle sensors and high-precision map data in the vehicle networking environment, it constructs a driving environment space under the unified benchmark of the high-precision map, and performs state estimation based on redundant information from multiple sources and asynchronous information. Further superimposing with the non-sensorized information of the high-precision map sensor, finally completes the estimation of the state quantity in the driving environment space.

[0044] Further, at least a part of the perception targets may include connected vehicles and other obstacles other than the connected vehicles; wherein, the sensorized information may include connected vehicle detection information for detecting connected vehicles and non-connected vehicle detection information for detecting other obstacles other than the connected vehicles.

[0045] Furthermore, the non-connected vehicle detection information may include one or more of the following: dynamic obstacle information for detecting dynamic obstacles other than the connected vehicles; map detection information for detecting static obstacles on the map.

[0046] Wherein, the connected vehicle may be a vehicle in the same vehicle network as the autonomous vehicle; the dynamic obstacle may be an obstacle with a movable state, such as a truck, an animal, a person, etc. that is not connected to the vehicle network; the static obstacle may be an immovable obstacle, such as a building, a traffic light, etc.

[0047] Specifically, in the joint perception system, the original map data can be selectively loaded according to the perception requirements and vehicle state, and the map data is given a precision description to form an output consistent with the common sensor data form. This process is called sensorization of map data.

[0048] Specifically, the prior information about other connected vehicles in the map is regarded as the observation data of the connected vehicles, denoted as the observation formed by the data of connected vehicle i in the map on the state of the connected vehicle, called the map-connected vehicle observation. is the time set corresponding to this observation. The high-precision map sensor connected vehicle observation set can be defined as:

[0049]

[0050] Wherein, the connected vehicle state is used to represent the state information of the connected vehicles that the autonomous vehicle needs to perceive, such as the pose of the connected vehicle (6-degree-of-freedom model, including the position and angle of the origin of the connected vehicle coordinate system in the geographic projection coordinate system), size, etc.

[0051] The prior information about the dynamic obstacles of non-connected vehicles in the map is regarded as the observed data of the dynamic obstacles, denoted as the observation formed by the data pair of the connected vehicle o in the map on the state of the dynamic obstacle, which is called the map-dynamic obstacle observation. Let be the time set corresponding to this observation. Then the high-precision map sensor dynamic obstacle observation set is defined as:

[0052]

[0053] wherein, the state of the dynamic obstacle is used to represent the state information of the dynamic obstacle that the autonomous vehicle needs to perceive, such as the position, size, category, etc. of the dynamic obstacle in the geographic projection coordinate system.

[0054] The prior information about the positioning features in the map is regarded as the observed data of the static obstacles, denoted as the observation formed by the data pair of the static obstacle f in the map on the state of the feature, which is called the map-static obstacle observation. The high-precision map sensor static obstacle observation set is defined as:

[0055]

[0056] wherein, the state of the static obstacle is used to represent the state information of the static obstacle that the autonomous vehicle needs to perceive, such as the position, size, category, etc. of the static obstacle.

[0057] In summary, the high-precision map sensor observation is modeled as a set of map-static obstacle features, map-connected vehicles, and map-dynamic obstacle observations:

[0058] Z M ={Z M-F ,Z M-V ,Z M-o}

[0059] Furthermore, the steps of using the map sensor to extract features of at least a part of the perception targets in the map to obtain sensorized information may include: determining the position information of the perception targets to be perceived according to the perception target indication information; calculating the covariance of the position information of the at least a part of the perception targets according to the accuracy of the map, and using it as the sensorized information.

[0060] Specifically, the perception targets to be sensed are determined according to the perception target indication information, and the perception targets to be sensed can also be referred to as the set of road logical breakpoints. Specifically, the set of road logical breakpoints can be the range of perception requirements input by the map sensor. Based on the positioning information and dynamic and static obstacles, the corresponding positions on the high-precision map are determined (reference points divided by the smallest units of road structures such as road network positions, road segments in the road network, and road intersections).

[0061] Furthermore, the covariance is used to represent sensor noise or, so to speak, the uncertainty of perception. The covariance can be determined by using conventional and appropriate calculation methods, and the embodiments of the present application do not limit this.

[0062] In other words, the map data can be all the information within the global scope of the map, and the map sensor can generate sensor signals related to the driving task according to the range of the vehicle's perception requirements. Therefore, it is necessary to effectively organize the original map data.

[0063] From the perspective of the reliability of map data, according to the perception requirements, the accurate and reliable hierarchical vector high-precision map data is distinguished into non-sensorized information and sensorized information. The accurate and reliable real-time map data (road-level and lane-level road network information, other dynamic traffic flow information, traffic event information, and decision-making assistance information) is directly output as non-sensorized information for use in state estimation in the driving environment space.

[0064] Other data is then sensorized and output in the form of map sensor signals to be fused with other sensor data in the multi-vehicle joint perception system. This part of the data determines the set of road logical breakpoints related to the driving task through the range of perception requirements input by the map sensor. When generating the map sensor signal, first, in the positioning feature layer and the dynamic obstacle layer, all relevant dynamic and static target positions are indexed according to the set of logical breakpoints, and the covariance of the data is determined according to the accuracy of the map data as the output signal of the map sensor.

[0065] In the specific implementation of step S13, the non-sensorized information in the high-precision map data except the above-mentioned sensorized information is extracted, superimposed with the ordinary sensor perception information and the map sensor perception information, and fused and perceived under the unified map spatio-temporal benchmark. The environmental perception result of the autonomous vehicle is obtained by solving the problem of state estimation.

[0066] Refer to Figure 2 , Figure 2 is Figure 1Flowchart of a specific implementation of step S13. The step of determining the perception result of the autonomous vehicle according to the sensorized information may include steps S21 to S24, and may also include steps S21 to S26. Each step is described below.

[0067] In step S21, sensor information obtained by one or more sensors detecting at least a part of the perception targets is acquired.

[0068] Among them, the sensors may include on-vehicle sensors and off-vehicle sensors. That is, the sensor information may include self-detection information obtained by the on-vehicle sensors detecting the autonomous vehicle itself and external detection information obtained by each off-vehicle sensor detecting other objects (including the current autonomous vehicle) other than its own vehicle.

[0069] Specifically, the sensor information may include combined navigation observations of its own state and external sensor observations of other targets. The objects of the external observations may be other intelligent connected vehicles, or dynamic obstacles, or targets included or to be updated in the map data.

[0070] Furthermore, in the step of detecting the autonomous vehicle itself to obtain self-detection information, the observation data generated by the combined positioning system of the i-th connected vehicle at time k i for the state of its own vehicle can be described as The set of combined positioning data of the i-th connected vehicle at time k is Then the combined positioning observations directly observing the state of the vehicle itself in the perception space can be modeled as:

[0071]

[0072] Among them, the perception space is used to represent the environmental space where the sensors can detect targets.

[0073] Furthermore, the external detection information may include one or more of the following: connected vehicle detection information for detecting other connected vehicles in the vehicle network; dynamic obstacle information for detecting dynamic obstacles other than the connected vehicles; map detection information for detecting static obstacles and road information on the map; new static obstacle detection information for detecting static obstacles not recorded on the map; new road detection information for detecting road information not recorded on the map.

[0074] Specifically, data from external target sensors of other vehicles is obtained through the vehicle networking to make up for the deficiencies in the perception ability of a single vehicle. Considering the actual scenario, in addition to the observations of dynamic obstacles and static non-map targets in the on-vehicle external target sensor data, it also includes the observations of other networked vehicles, positioning features, road information, etc.

[0075] Furthermore, in the step of detecting other networked vehicles in the vehicle networking to obtain networked vehicle detection information, the observation of other vehicles by the networked vehicle can be referred to as vehicle-to-vehicle observation, denoted as the index of the static non-map target, for the i-th networked vehicle at time k i the observation data of another networked vehicle and the corresponding set of times for this observation is The corresponding set is defined as:

[0076]

[0077] In the step of detecting dynamic obstacles other than the networked vehicles to obtain dynamic obstacle information, it can be assumed that the sensor of a certain networked vehicle i makes an observation of a dynamic obstacle o at time k i and the data description of this observation is and the corresponding set of times for this observation is Then the vehicle-obstacle observation set is:

[0078]

[0079] In the step of detecting static obstacles and road information on the map to obtain map detection information, the observation of the positioning feature by the networked vehicle can be referred to as vehicle-feature observation, denoted as f for the index of the static non-map target, for the i-th networked vehicle at time k i the observation data of the positioning feature f, and the corresponding set of times for this observation is The corresponding set is defined as:

[0080]

[0081] In the step of detecting static obstacles not recorded on the map to obtain new static obstacle detection information, the observation of the static non-map target can be referred to as vehicle-static non-map target observation, denoted as s for the index of the static non-map target, for the i-th networked vehicle at time k i the observation data of the static non-map target s, and the corresponding set of times for this observation is Then the corresponding observation set is defined as:

[0082]

[0083] In the step of detecting road information not recorded in the map to obtain new road detection information, the observation of road information by connected vehicles is called vehicle-road information observation. Let r be the index of road information. For the i-th connected vehicle at time k i The observation data of road information r, and the set of corresponding times for this observation is The corresponding set is defined as:

[0084]

[0085] Based on the above information, the observation of external target sensors in the perception space can be modeled as:

[0086] Z V ={Z V-O ,Z V-S ,Z V-V ,Z V-F ,Z V-R}

[0087] In step S22, the sensor information of at least some of the perception targets and the sensorized information are fused to obtain the fusion information of each perception target among at least some of the perception targets.

[0088] It should be noted that the technology of fusing the sensor information of at least some of the perception targets and the sensorized information can be implemented using appropriate conventional technologies, and the embodiments of the present application do not limit this.

[0089] In step S23, according to the fusion information, the set of observable quantities of each perception target among at least some of the perception targets is determined.

[0090] Among them, according to the perception method in the embodiments of the present application, the perception information of dynamic and static targets and road traffic information can be determined according to the perception requirements of intelligent vehicles. Among them, dynamic targets can include connected vehicles and other dynamic obstacles For dynamic targets, a state set X H is constructed, and the dynamic target state set can be expressed as:

[0091] X H ={X V ,X O}

[0092] Static targets such as traffic signs, lane lines, and cones in traffic scenarios provide reference information for vehicle autonomous positioning. Additionally, static targets within the road edge need to be considered in the decision-making system. Static targets in the environment include positioning features already included in the map as well as static non-map targets which are also the targets to be updated in the map. For static targets, construct the state set X Z , and the static target state set can be expressed as:

[0093] X Z ={X F ,X S}

[0094] The road information state set includes road network information lane network information dynamic traffic information and decision-making assistance information For road information, construct the state set X R , and the road information state set can be expressed as:

[0095] X R ={X WL ,X w ,X B ,X P}

[0096] In summary, the targets in the driving environment space can be composed of dynamic targets, static targets, and road information, and their mathematical definition is:

[0097]

[0098] The corresponding perception result of the driving vehicle, that is, the set of observed quantities, can be expressed as:

[0099] X D ={X H ,X Z ,X R}

[0100] In the specific implementation of step S24, the maximum likelihood estimation algorithm can be used to determine the state quantity estimation of each perception target in the at least part of the perception targets according to the set of observed quantities of each perception target in the at least part of the perception targets.

[0101] Specifically, among various types of perception inputs of the joint perception system, there are generally observations of the same target from different sources, and there are contradictions among them. In order to obtain a consistent perception result, maximum likelihood estimation is performed. Since the joint perception problem needs to describe the state under a unified map spatio-temporal reference, it can be reduced to maximizing the likelihood probability of all system observations with respect to the state quantity. According to the Maximum Likelihood Estimation (MLE) theory, the driving environment space state estimator maximizes the probability of the observations conditional on the set of observations.

[0102] In the embodiment of the present invention, by obtaining sensor information of one or more sensors detecting at least a part of the perception targets, and then obtaining the fusion information and the set of observations of each perception target in at least a part of the perception targets, and then determining the state quantity estimation of each perception target in at least a part of the perception targets, it is possible to determine the perception result based on the information fusion of the map sensor and the sensor, and further improve the perception accuracy.

[0103] Further, the following formula can be used to determine the state quantity estimation of each perception target in at least a part of the perception targets according to the set of observations of each perception target in at least a part of the perception targets:

[0104]

[0105] Wherein, is used to represent the estimation result of the driving environment space state parameter in the sense of maximum likelihood, X D is used to represent the set of observations within a certain time depth in the driving environment space, Z p is used to represent the set of measurements in the asynchronous heterogeneous perception space that can be obtained by the intelligent vehicle, P(Z p |X D ) is the conditional probability of the set of observations Z D given the state parameter X p .

[0106] Assuming that the likelihood probabilities of different sensors at different times are independent of each other, then the conditional probability can be further written as the joint product of the likelihood probabilities of all observations within a period of time.

[0107]

[0108]

[0109] Assuming that the noise follows a Gaussian distribution, the problem of maximizing the likelihood can be transformed into the following least-squares optimization problem. In this equation, the residual is the difference between the actual observation and the prediction of the perception result by the system state. In the state estimator of the perception container, by optimizing the values of the state variables, the weighted sum of various observation residuals is minimized.

[0110]

[0111] Observation residual:

[0112]

[0113] The above equation gives the mathematical method for the state estimator in the map perception container to estimate the state X in the driving environment space. D It should be noted that this equation adjusts the weights of sensor data with different precisions in the optimization problem through the inverse matrix (Covariance Matrix) of the covariance matrix of each sensor noise, thereby organically integrating the data of each vehicle-mounted sensor and map sensor in the entire joint perception system, and finally forming a consistent perception result.

[0114] In the embodiment of the present invention, the maximum likelihood estimation algorithm is adopted to estimate the perception result of the autonomous vehicle, which can effectively improve the accuracy of the estimation result.

[0115] In the specific implementation of step S25, the state quantity estimations of each perception target in at least a part of the perception targets are superimposed with the non-sensor information to obtain the perception result of the autonomous vehicle.

[0116] Further, the non-sensor information may include one or more of the following: road information, new traffic flow information, traffic event information, and decision-making assistance information.

[0117] Specifically, the non-sensor information may include road network information at the road level and lane level, dynamic traffic flow information, traffic event information, decision-making assistance information, etc.

[0118] It should be pointed out that in the step of superimposing the state quantity estimations of each perception target in at least a part of the perception targets with the non-sensor information, the information from the map non-sensor information (road network, traffic information, etc.) and sensors (including actual sensors and map virtual sensors) is uniformly expressed in the global coordinate system of the high-precision map, achieving consistent spatial expression and can be directly superimposed.

[0119] In an embodiment of the present invention, non-sensorized information other than the sensorized information can also be extracted from the map, and the estimated state quantity of each perception target in the at least part of the perception targets can be superimposed with the non-sensorized information to obtain the perception result of the autonomous driving vehicle. In this way, other information of the high-precision map can be further added on the basis of the detection of multiple sensors to further improve the overall effect of perception.

[0120] In an embodiment of the present invention, by acquiring perception target indication information, the perception target indication information is used to indicate the environmental status information that the autonomous driving car needs to perceive; obtaining a high-precision map, establishing a map perception container, and performing feature extraction on at least a portion of the perception targets in the map to obtain sensorized information, thereby generating a signal in the same format as an ordinary sensor based on the original map data. In addition, considering that some information in the high-precision map data has a confidence level far exceeding that of the on-board sensor, it is directly superimposed on the perception result. Finally, the joint perception information of multiple vehicles (including ordinary sensor perception information, map sensor perception information, and other non-sensorized data in the high-precision map) is fused and perceived under a unified map spatiotemporal reference to determine the perception result of the autonomous driving car; wherein the positioning error of the map is within a preset error range. The present invention can effectively eliminate perception blind spots and improve perception accuracy.

[0121] Reference Figure 3 , Figure 3 This is a schematic diagram of the structure of a high-precision map perception container design device for multi-vehicle joint perception in an embodiment of the present invention. The high-precision map perception container design device for multi-vehicle joint perception may include:

[0122] An information acquisition module 31 is used to acquire perception target indication information, where the perception target indication information is used to indicate environmental state information that the autonomous driving vehicle needs to perceive;

[0123] a sensorized information determination module 32 for acquiring a high-precision map, establishing a map perception container, and performing feature extraction on at least a portion of the perception targets in the map to obtain sensorized information;

[0124] The perception result solving module 33 is used to extract non-sensorized information other than the above-mentioned sensorized information from the high-precision map data, superimpose it with the ordinary sensor perception information and the map sensor perception information, perform fusion perception under a unified map time and space reference, and obtain the environmental perception results of the autonomous driving vehicle through state estimation solution.

[0125] The high-precision map is a map with a positioning error within a preset error range.

[0126] For the principle, specific implementation, and beneficial effects of the high-precision map perception container design device for multi-vehicle joint perception, please refer to the relevant descriptions of the high-precision map perception container design method for multi-vehicle joint perception described above, and will not be elaborated here.

[0127] Refer to Figure 4 , Figure 4 is a schematic structural diagram of a perception system based on a high-precision map perception container design method in an embodiment of the present invention.

[0128] Such as Figure 4 In the solid line box, it is used to represent the entity module, and in the dashed line box, it is used to represent the determined information.

[0129] First, collect the sensor information determined by multiple sensors S1, S2, S3 to S4.

[0130] Use the map sensor M S to determine the map information M, and determine the perception target indication information P r , and output the sensorized information Z M .

[0131] Then use the driving environment space builder D C to fuse the sensor information of at least a part of the perception targets and the sensorized information, so as to obtain the fusion information of each perception target in at least a part of the perception targets, and determine the set of observables of each perception target in at least a part of the perception targets according to the fusion information.

[0132] In a specific implementation, D C can also allocate the state quantities in the driving environment space according to the expression results of multi-source sensors in the map coordinate system.

[0133] More specifically, Dc performs data association according to the expression results of multi-source sensors in the geographic projection coordinate system, determines the set of observables to be estimated, allocates the state quantities in the driving environment space, and completes the construction of the driving environment space.

[0134] Then use the driving environment space state estimator M E Based on the maximum likelihood estimation algorithm, according to the set of observables of each perception target in at least a part of the perception targets, determine the state quantity estimation of each perception target in at least a part of the perception targets

[0135] Specifically, M E is a driving environment space state estimator, which realizes the estimation of the state quantities in the driving environment space based on multi-source heterogeneous and asynchronous redundant data.

[0136] Further, a map sensor M is adopted S Extract non-sensorized information X0 other than the sensorized information from the map.

[0137] Estimate the state quantity of each of the at least part of the sensed objects Superimpose it with the non-sensorized information X0 to obtain the sensing result of the autonomous vehicle

[0138] Regarding Figure 4 For more details of the sensing system shown, reference may be made to the foregoing and will not be elaborated here.

[0139] In an embodiment of the present invention, a storage medium is further provided, on which a computer program is stored. When the computer program is run by a processor, the steps of the above method are executed. The storage medium may be a computer-readable storage medium, for example, it may include a non-volatile memory or a non-transitory memory, and may also include an optical disc, a mechanical hard disk, a solid-state drive, etc.

[0140] Specifically, in an embodiment of the present invention, the processor may be a central processing unit (CPU for short), and the processor may also be other general-purpose processors, a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0141] It should also be understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0142] In an embodiment of the present invention, a terminal is further provided, including a memory and a processor. A computer program capable of running on the processor is stored on the memory. When the processor runs the computer program, it executes the steps of the above method. The terminal includes, but is not limited to, automobiles, in-vehicle control devices, terminal devices externally connected to or integrated into the automobiles, mobile phones, computers, tablet computers, and other terminal devices.

[0143] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be subject to the scope defined by the claims.

Claims

1. A high-precision map perception container design method for multi-vehicle joint perception, characterized in that Including the following steps: Obtain perception target indication information, which is used to indicate the environmental state information that the autonomous vehicle needs to perceive; Obtain a high-precision map, establish a map perception container to extract features of at least a part of the perception targets in the map to obtain sensorized information; Extract non-sensorized information in the high-precision map data other than the above-mentioned sensorized information, superimpose it with ordinary sensor perception information and map sensor perception information, perform fusion perception under a unified map spatio-temporal reference, and obtain the environmental perception result of the autonomous vehicle by solving the state estimation problem; The extraction of non-sensorized information in the high-precision map data other than the above-mentioned sensorized information, superimposing it with ordinary sensor perception information and map sensor perception information, and performing fusion perception under a unified map spatio-temporal reference includes: Obtain sensor information detected by one or more sensors for at least a part of the perception targets; fuse the sensor information of at least a part of the perception targets and the sensorized information to obtain the fusion information of each perception target in at least a part of the perception targets; According to the fusion information, determine the set of observation quantities of each perception target in at least a part of the perception targets, and the set of observation quantities can be expressed as: X D = {X H , X Z , X R}, where X D represents the set of the observed quantity, X H represents the set of the dynamic target construction states, X Z represents the set of the static target construction states, X R represents the set of the road information construction states; using the maximum likelihood estimation algorithm, according to the set of the observed quantity of each sensing target in the at least part of the sensing targets, determining the state quantity estimation of each sensing target in the at least part of the sensing targets; superimposing the state quantity estimation of each sensing target in the at least part of the sensing targets with the non-sensorized information to obtain the sensing result of the autonomous vehicle; Wherein, the high-precision map is a map with a positioning error within a preset error range.

2. The high-precision map perception container design method for multi-vehicle joint perception according to claim 1, wherein At least a part of the perception targets include connected vehicles and other obstacles other than the connected vehicles; Wherein, the sensorized information includes connected vehicle detection information for detecting connected vehicles and non-connected vehicle detection information for detecting other obstacles other than the connected vehicles.

3. The high-precision map perception container design method for multi-vehicle joint perception according to claim 2, wherein The non-connected vehicle detection information includes one or more of the following: Dynamic obstacle information for detecting dynamic obstacles other than the connected vehicles; Map detection information for detecting static obstacles on the map.

4. The high-precision map perception container design method for multi-vehicle joint perception according to claim 1, wherein Using a map sensor to extract features of at least a part of the perception targets in the map to obtain sensorized information includes: Determine the position information of the perception targets that need to be perceived according to the perception target indication information; According to the accuracy of the map, calculate the covariance of the position information of at least a part of the perception targets and use it as the sensorized information.

5. The high-precision map perception container design method for multi-vehicle joint perception according to claim 1, characterized in that, Using the following formula, determine the state quantity estimation of each perception target in at least a part of the perception targets according to the set of observation quantities of each perception target in at least a part of the perception targets: Among them, is used to represent the estimation result of the driving environment space state parameters in the sense of maximum likelihood, X D is used to represent the set of observations within a certain time depth in the driving environment space, Z p is used to represent the set of measurements in the asynchronous heterogeneous perception space that can be obtained by the intelligent vehicle, P(Z p |X D ) is the conditional probability of the set of observations Z D under the given state parameter X p .

6. A high-precision map perception container design device for multi-vehicle joint perception, characterized in that, Including: An information acquisition module, configured to obtain perception target indication information, which is used to indicate the environmental state information that the autonomous vehicle needs to perceive; A sensorized information determination module, configured to obtain a high-precision map, establish a map perception container to extract features of at least a part of the perception targets in the map to obtain sensorized information; The perception result solving module is used to extract the non-sensorized information in the high-precision map data except the above-mentioned sensorized information, superimpose it with the ordinary sensor perception information and the map sensor perception information, perform fusion perception under the unified map spatio-temporal reference, and obtain the environmental perception result of the autonomous vehicle by solving the state estimation problem; The perception result solving module is further used for: Obtaining sensor information detected by one or more sensors for at least a part of the perception targets; fusing the sensor information of at least a part of the perception targets and the sensorized information to obtain the fusion information of each perception target in at least a part of the perception targets; Determining an observable set of each perception target in at least a part of the perception targets according to the fusion information, and the observable set can be expressed as: X D ={X H , X Z , X R}, where X D represents the set of the observed quantity, X H represents the set of the dynamic target construction states, X Z represents the set of the static target construction states, X R represents the set of the road information construction states; adopting the maximum likelihood estimation algorithm, determining the state quantity estimation of each of the at least part of the sensed targets according to the set of the observed quantity of each of the at least part of the sensed targets; superposing the state quantity estimation of each of the at least part of the sensed targets with the non-sensorized information to obtain the sensing result of the autonomous vehicle; wherein, the high-precision map is a map with a positioning error within a preset error range.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is run by a processor, it executes the steps of the method for designing a high-precision map perception container for multi-vehicle joint perception according to any one of claims 1 to 5.

8. A terminal, comprising a memory and a processor, wherein a computer program capable of running on the processor is stored on the memory, characterized in that When the processor runs the computer program, it executes the steps of the method for designing a high-precision map perception container for multi-vehicle joint perception according to any one of claims 1 to 5.

Images