Self-adaptive lane level map matching method and device
Through the adaptive lane-level map matching method, the hidden Markov chain model and Viterbi algorithm are used to optimize the calculation, and the real-time positioning accuracy and calculation efficiency problems in the existing technology are solved, and stable decimeter-level positioning information acquisition and high quality are achieved. lane-level map matching.
Patent Information
- Application Number
- CN202310536795.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-10
- Publication Date
- 2024-11-12
AI Technical Summary
The existing map matching method based on hidden Markov chains is insufficient in real-time and stable acquisition of decimeter-level positioning information, especially when processing large-scale road networks and fine map data, the calculation amount is large, the real-time performance is poor, and the positioning data is Quality affects the matching results and is prone to incorrect matching.
Adaptive lane-level map matching method is adopted to construct state space and observation space, and calculate the state transition probability matrix and the emission probability matrix based on the hidden Markov chain model, and use the Viterbi algorithm to output the most matching state sequence, optimize the calculation process and improve the Match accuracy.
It realizes stable and real-time acquisition of decimeter-level positioning information, reduces computing resources and time consumption, and improves the interpretability of map matching and positioning data quality.
Smart Images

Figure CN118936486A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of positioning technology, and in particular relates to an adaptive lane-level map matching method and device. Background Art
[0002] With the development of autonomous driving industry technology, single-vehicle intelligent perception and vehicle-road collaborative fusion perception technologies have developed rapidly. Among them, single-vehicle intelligent and vehicle-road collaborative fusion perception both require higher-precision positioning, especially real-time, lane-level positioning that adapts to various road conditions.
[0003] The core goal of map matching is to assist in correcting positioning information. Usually, due to the deviation of positioning information, the observed vehicle position is not on the correct road / lane. Obviously, if the road / lane is wrong, the positioning information must be wrong, and it cannot support the autonomous driving system. With the well-defined geographic information brought by high-precision maps, map matching technology has always been an important way to effectively improve positioning quality in addition to sensor data fusion.
[0004] At present, the decimeter-level positioning desired by autonomous driving technology cannot be obtained in real time and stably. Existing technologies such as map matching methods based on hidden Markov chains mainly involve: after completing data collection, the state space, observation space, generation probability, state transition probability, and the best matching path sequence are constructed for all point trajectories and traffic road network data, and the method is based on map matching of global data, with better matching accuracy and better calculation stability. However, there are also many defects as follows:
[0005] 1. The map matching based on the hidden Markov chain model often processes all historical trajectories. That is, the processing time is after the operation is completed, and all historical trajectories and related road traffic network map data are obtained for post-processing, and real-time map matching is not achieved. This makes the traditional hidden Markov chain model used for autonomous driving map matching, and the calculation model needs to be further optimized.
[0006] 2. In the process of map matching, when the road network space is large, such as the road network covered by the vehicle positioning trajectory is relatively wide, or the map scale is relatively fine, resulting in a large amount of road traffic map data, that is, the state space dimension in the map matching model is high, which will lead to a large amount of calculation for the overall model and be extremely time-consuming. This will also limit the real-time and availability of the map matching calculation method.
[0007] 3. The results of map matching are often affected by the quality of vehicle positioning data, such as overall deviation, maximum deviation, positioning frequency, presence or absence of bridge and tunnel scenarios, etc. Large local deviations, such as due to satellite signal obstruction, positioning interruption or large deviation amplitude, will lead to matching of wrong road routes and cause chained incorrect map matching.
[0008] In order to overcome the defects of the existing map matching method based on hidden Markov chain, this embodiment proposes a lane-level map matching technology to more accurately obtain the real-time trajectory of the vehicle. Summary of the invention
[0009] One of the purposes of the present invention is to provide an adaptive lane-level map matching method to provide stable and real-time decimeter-level positioning information.
[0010] To achieve the above object, the technical solution adopted by the present invention is:
[0011] An adaptive lane-level map matching method is applied to the positioning of a specified object. The adaptive lane-level map matching method is implemented based on a hidden Markov chain model and specifically includes the following steps:
[0012] Construct the current state space based on the latest positioning of the specified object and the historical lane sequence;
[0013] Construct the current observation space based on the latest positioning and historical trajectory of the specified object;
[0014] Based on the state space, construct a state transition probability matrix according to lane topology association;
[0015] Calculate the mapping probability between positioning in observation space and lanes in state space, and construct the emission probability matrix;
[0016] By calculating the maximum probability of the implicit state chain, the state sequence that best matches the observation space is output, that is, the predicted lane sequence;
[0017] According to the predicted lane sequence, a real-time lane sequence corresponding to the real-time trajectory is restored and output based on the lane-level road network topology relationship.
[0018] Several optional methods are also provided below, but they are not intended to be additional limitations on the above-mentioned overall solution, but are merely further supplements or preferences. Under the premise that there are no technical or logical contradictions, each optional method can be combined with the above-mentioned overall solution separately, and multiple optional methods can also be combined.
[0019] Preferably, the constructing the current state space according to the latest positioning of the designated object and the historical lane sequence includes:
[0020] Match the adjacent lane sets based on the latest positioning of the specified object;
[0021] Take the N lanes closest to the current moment in the historical lane sequence;
[0022] The set of matched adjacent lanes is merged with the taken N lanes as the state space.
[0023] Preferably, the matching of adjacent lane sets includes:
[0024] Get the lane group whose shortest distance to the latest positioning is less than the threshold;
[0025] Take the front and rear lane groups connected to the acquired lane group, and take the acquired lane group and the lanes included in the front and rear lane groups as a lane set.
[0026] Preferably, the constructing the current observation space according to the latest positioning and historical trajectory of the specified object includes:
[0027] Starting from the latest positioning, trace back the historical trajectory of the preset length;
[0028] The historical trajectories obtained by tracing are thinned out and combined with the latest positioning to obtain the observation space.
[0029] Preferably, the state transition probability matrix is constructed based on the state space and according to the lane topology association, including:
[0030] The scenarios where the object turns from the current lane to other lanes according to the lane topology association definition include: turning from the current lane to the lane of the same lane group and turning from the current lane to the lane of the downstream lane group;
[0031] Define the sum of the probabilities of turning from the current lane to the same lane group as σ, where the probability of staying in the current lane is σ*ρ, and the sum of the probabilities of turning to other lanes in the same lane group is σ*(1-ρ); define the sum of the probabilities of turning from the current lane to the lane in the downstream lane group as 1-σ;
[0032] Then the state transition probability matrix consists of lane q i Go to lane q j The probability calculation formula is as follows:
[0033]
[0034] In the formula, a ij For lane q i Go to lane q j The state transition probability of lane q i and Lane q j is the lane in the state space.
[0035] Preferably, the calculating the mapping probability between the positioning in the observation space and the lane in the state space and constructing the emission probability matrix comprises:
[0036] Take the tth observation value in the observation space and the i-th state value in the state space, and calculate the mapping probability as:
[0037]
[0038] In the formula, P(q i |o t ) is the state value q i Under these conditions, we get the observed value o t probability, called the mapping probability, the observed value o t is the tth observation value, state value q i is the i-th state value, Distancebetweenq i ando t is the state value q i With the observed value o t The shortest distance between them, m is the dimension of the current state space, that is, the total number of lanes in the state space.
[0039] Preferably, the maximum probability calculation of the implicit state chain to output the state sequence that best matches the observation space includes:
[0040] Construct the objective function of the maximum probability calculation of the implicit state chain;
[0041] The objective function is iteratively calculated through the Viterbi algorithm to obtain the maximum probability value of the implicit state chain, that is, the state sequence that best matches the observation space.
[0042] Preferably, the objective function is as follows:
[0043]
[0044] In the formula, P(q1···q i ···q j ···q m |o1···o i ···o j ···o m ) represents the observation sequence o1···o i ···o j ···o m The corresponding state sequence is q1···q i ···q j ···q m The probability of i is the i-th state value, o iis the i-th observation, q0 is the initial state, P(q0) is the probability distribution corresponding to the initial state, m is the dimension of the state space, P(q t |q t-1 ) is the value from the state q t-1 Transfer to state value q t The probability, P(q t |o t ) is in the state value q t Next, we get the observed value o t probability.
[0045] Preferably, the step of restoring and outputting a real-time lane sequence corresponding to the real-time trajectory based on the predicted lane sequence and the lane-level road network topology relationship comprises:
[0046] Acquire two adjacent but unconnected lanes in the predicted lane sequence as a lane pair to be completed;
[0047] According to the lane-level road network topology, the paths are radiated outward based on the lane pairs to be completed, and the lanes corresponding to the missing shortest paths between the lane pairs to be completed are completed, so as to obtain a spatially continuous lane sequence as the real-time lane sequence.
[0048] The present invention provides an adaptive lane-level map matching method, which uses the spatial and topological information of the road traffic network and the lane sequence of the vehicle to perform adaptive lane-level road network data filtering and determine the optimal state space of the map matching model, thereby greatly reducing the search space of the matching calculation, optimizing the calculation process, and reducing the requirements for computing resources and time consumption. Based on the topological association information of the road network and lanes, a probability matrix calculation method for state transition is constructed to accurately express the actual driving behavior and improve the interpretability of map matching.
[0049] A second object of the present invention is to provide an adaptive lane level map matching device, comprising a processor and a memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the adaptive lane level map matching method when executed by the processor. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 A flowchart of an adaptive lane-level map matching method of the present invention;
[0051] Figure 2 This is a schematic diagram of the front and rear lane groups of the present invention;
[0052] Figure 3 Schematic diagram of lane groups with and without a separation strip in the present invention;
[0053] Figure 4 This is a schematic diagram of the historical trajectory thinning process of the present invention;
[0054] Figure 5 It is a schematic diagram of the current lane group and the downstream lane group of the present invention. DETAILED DESCRIPTION
[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.
[0057] The operation of the autonomous driving system requires stable, real-time, decimeter-level positioning information. Real-time lane-level map matching is one of the methods to solve the high-precision positioning of autonomous driving. This embodiment is based on multi-sensor fusion positioning data, combined with the spatial and topological information of the road traffic network, to perform adaptive lane-level traffic network data filtering, determine the optimal state space of the map matching model, based on the dynamic hidden Markov chain model, and finally use the Vibert algorithm to perform map matching calculations to obtain the current lane of the vehicle, and update the lane sequence of the vehicle to complete the map matching.
[0058] An adaptive lane-level map matching method proposed in this embodiment is applied to the positioning of a specified object. The specified object here should be understood as any object that needs to be positioned and is traveling on a lane, such as a vehicle, a motorcycle, etc. For ease of description, this embodiment is described using a vehicle as an example.
[0059] Hidden Markov Model (HMM) is a statistical model used to describe a Markov process with hidden unknown parameters (Markov process is a type of random process with the following characteristics: given the current state, its future evolution does not depend on its past evolution). The difficulty of the hidden Markov model is to determine the implicit parameters of the process from the observable parameters, and then use these parameters for further analysis, such as pattern recognition. This embodiment uses the positioning point as the observation parameter and the lane as the implicit state parameter for lane-level map matching.
[0060] like Figure 1 The adaptive lane-level map matching method of this embodiment specifically includes the following steps:
[0061] Step 1: Construct the current state space based on the latest positioning of the specified object and the historical lane sequence.
[0062] This embodiment mainly follows the following principle when constructing the adaptive state space: the more adjacent nodes are, the more important they are, so the neighboring area has more points and is denser. However, too dense positioning points are likely to cause a large number of redundant calculations, so it is necessary to cut redundant trajectories. In this embodiment, whether it is positioning points or lanes, the impact on the results is small after exceeding a certain dimension, so this embodiment takes positioning points and lanes within a predetermined dimension, such as positioning points and lanes within a specified time length.
[0063] Step 1.1: Match the adjacent lane set according to the latest positioning of the specified object.
[0064] In this embodiment, adjacent lanes include lane groups with similar positioning and lanes included in the front and rear lane groups of the lane group. Therefore, when matching adjacent lane sets in this embodiment: first obtain a lane group with a shortest distance less than a threshold, the shortest distance is the shortest distance between the latest positioning and the lane group, then obtain the front and rear lane groups connected to the obtained lane group, and take the obtained lane group and the lanes included in the front and rear lane groups as the lane set.
[0065] The front and rear lane groups in this embodiment should be understood as lane groups connected upstream and downstream of the current lane group with the intersection as the boundary. Figure 2 As shown in the figure, there are 7 lane groups, the arrows indicate the driving direction of the lane group, and the star indicates the latest positioning of the specified object. When obtaining a lane group close to the positioning, the shortest line segment between the positioning and the 7 lane groups is calculated, and the length of the 7 shortest line segments is calculated. The one with the smallest length is taken as the lane group with similar positioning, that is, lane group 4 in the figure. Then, based on lane group 4, its front and rear lane groups are taken, that is, the rear lane group in this embodiment includes lane groups 5, 6 and 7, and the front lane group in this embodiment includes lane groups 1, 2 and 3, and the adjacent lane set is obtained.
[0066] like Figure 3 As shown in the figure, the upper part shows the situation where there is no separation strip in the middle of the lane, and the lower part shows the situation where there is a separation strip in the middle of the lane. The circle is the intersection, and the arrow indicates the driving direction of the lane in the lane group. The lane group in this embodiment is located between two adjacent intersections. If there is no separation strip between the forward and reverse lanes, then the lane group includes all the forward and reverse lanes, for example Figure 3 If there is no separation strip between the forward and reverse arrows (i.e., forward and reverse lanes), all forward and reverse lanes are considered as a lane group. If there is a separation strip between the forward and reverse lanes, the lane group is either all forward lanes or all reverse lanes. For example, Figure 3If there is a separation strip between the forward and reverse arrows (i.e., the forward and reverse lanes) in the lower part, two lane groups are formed on this road, one including all forward lanes and the other including all reverse lanes, and the two lane groups are separated by a separation strip.
[0067] In order to follow the principle that adjacent nodes are more important, in this embodiment, the threshold for lane group matching is set to 4m, that is, the threshold in the lane group with the shortest distance less than the threshold is first obtained and is set to 4m. In other embodiments, it can be adjusted according to actual needs.
[0068] Step 1.2: Take the N lanes in the historical lane sequence that are closest to the current moment.
[0069] The historical lane sequence is a set of lanes that the object to be located has passed through at historical moments. The set may include all lanes that the object to be located has passed through since its departure, or may be a portion of lanes that are retained after clipping. In order to reduce the calculation pressure, in this embodiment, the total number of historical lanes is not more than 4, and N is preferably 4, which can also be adjusted as needed.
[0070] Step 1.3: Combine the matched adjacent lane set and the taken N lanes as the state space. After obtaining the state space, update the initial state according to the state space.
[0071] This embodiment uses the spatial and topological information of the road traffic network and the lane sequence of vehicles to perform adaptive lane-level network data filtering to determine the optimal state space of the hidden Markov model, thereby greatly reducing the search space of the matching calculation, optimizing the calculation process, and reducing computing resources and time consumption.
[0072] Step 2: Construct the current observation space based on the latest positioning and historical trajectory of the specified object.
[0073] Based on the Bayesian network (the Bayesian network is a probabilistic network, a graphical network based on probabilistic reasoning, and the Bayesian formula is the basis of this probabilistic network), combined with the network characteristics of the road traffic map, it can be seen that the more adjacent nodes are, the more important their impact on the current node matching calculation process. Therefore, the more adjacent the area, the more points are taken, and the denser they are. At the same time, considering that the dimension of the observation space is an important factor affecting the calculation time and matching timeliness, this embodiment extracts historical trajectories. For the number of extracted trajectory points, the smaller the better without affecting the abnormality of the transmission probability calculation (such as non-convergence of the calculation).
[0074] Based on this, this embodiment constructs an adaptive observation space calculation method to extract more trajectory points in the adjacent area, and the overall extraction ratio of trajectory points is dynamically adjusted according to the characteristics of the trajectory sequence. Specifically, the following steps are included.
[0075] Step 2.1: Starting from the latest positioning, trace back the historical trajectory of the preset length.
[0076] In this embodiment, the preset length of tracing back is set to the length corresponding to the past four lanes from the current lane, that is, the three historical lanes closest to the current moment in the historical lane sequence are taken, and the earliest moment among the three historical lanes is determined. The time length between the current moment and the earliest moment is the time length of tracing back. The time length of tracing back is also dynamically changed according to the length of the tracing back lane.
[0077] like Figure 4 As shown, the current time is t0, and the time length of the retrospective is s, then the time t 0-s As the dividing line, take t within the dividing line 0-S The trajectory from time period to t0 is regarded as the valid trajectory area for matching calculation at the current moment. Correspondingly, the trajectory outside the dividing line is the trajectory area not considered for matching calculation at the current moment.
[0078] Step 2.2: Thin out the historical trajectories obtained by tracing back and combine them with the latest positioning to obtain the observation space.
[0079] The valid trajectory area obtained in step 2.1 is the historical trajectory obtained by tracing back. In the valid trajectory area, the trajectory data of the reaction state-observation probability model characteristics are obtained after thinning, expressed as gray dots. The gray dots in the valid trajectory area are the thinned trajectories. Combined with the latest positioning, it is the observation space after adaptive calculation. The asterisks in the figure are the time distribution after thinning.
[0080] When processing vectorized data, there are often a lot of duplicate data in the records, which brings a lot of inconvenience to further data processing. On the one hand, the redundant data wastes a lot of storage space, and on the other hand, it causes the graphics to be expressed to be not smooth or not up to standard. Therefore, it is necessary to use some rules to minimize the number of data points while ensuring that the shape of the vector curve remains unchanged. This process is called thinning. In this embodiment, conventional thinning processing is implemented, such as vertical distance limit method thinning, DP algorithm, etc.
[0081] Step 3: Based on the state space, construct the state transition probability matrix according to the lane topology association.
[0082] Topological association refers to the relationship between line objects such as roads, lanes, lane groups, and lane dividing lines in a road traffic network. The lane-level map matching calculation process mainly considers the topological association of lanes and lane groups. First, based on the geometric relationship of the lane group where the current lane is located, search for the front and rear lane groups, that is, the lane groups that have common endpoints with the current lane group. Then query all lanes within the lane through the geometric features of the lane.
[0083] Therefore, in this embodiment, the scenarios in which the specified object turns from the current lane to other lanes according to the lane topology association definition include: first, turning from the current lane to the lane of the same lane group; second, turning from the current lane to the lane of the downstream lane group. Figure 5 As shown, the current lane group and the downstream lane group are displayed.
[0084] This embodiment defines the sum of probabilities of turning from the current lane to the same lane group as σ (for example, σ=0.6, adjustable) based on the geometric features of the trajectory and lane (for example, the maximum distance between two adjacent positioning points in the historical trajectory, the average distance between positioning points in the historical trajectory, the shortest lane length, etc.), where the probability of staying in the current lane is σ*ρ (for example, ρ=0.7, adjustable), and the sum of probabilities of turning to other lanes in the same lane group is σ*(1-ρ); the sum of probabilities of turning from the current lane to the lane of the downstream lane group is defined as 1-σ.
[0085] Then the state transition probability matrix consists of lane q i Go to lane q j The probability calculation formula is as follows:
[0086]
[0087] In the formula, a ij For lane q i Go to lane q j The state transition probability of lane q i and Lane q j is the lane in the state space. When calculating the state transition probability matrix, lane q i and Lane q j It is understood that i and j are only used to distinguish any two lanes in the state space, and are not specifically limited to the i-th and j-th state values in the state space.
[0088] Step 4: Calculate the mapping probability between positioning in the observation space and lanes in the state space, and construct a launch probability matrix.
[0089] The emission probability mainly expresses the probability of mapping from a certain observation state to a specific state. In map matching, it is expressed as the probability of a certain trajectory point being mapped to a lane in the traffic network. Based on the first law of geography: the closer the distance, the closer the relationship. It can be seen that the main factor affecting the emission probability value in this embodiment is the distance from the trajectory to the lane. Taking into account the quality characteristics of GNSS positioning, the inverse distance calculation in this embodiment uses the result of the second square root, and the mapping probability calculation method is as follows:
[0090] Take the tth observation value in the observation space and the i-th state value in the state space, and calculate the mapping probability as:
[0091]
[0092] In the formula, P(q i |o t ) is the state value q i Under these conditions, we get the observed value o t probability, called the mapping probability, the observed value o t is the tth observation value, state value q i is the i-th state value, Distancebetweenq i ando t is the state value q i With the observed value o t The shortest distance between them, m is the dimension of the current state space, that is, the total number of lanes in the state space. In addition, the dimension of the observation space can be expressed as n, 1≤t≤n.
[0093] Step 5: By calculating the maximum probability of the implicit state chain, output the state sequence that best matches the observation space, that is, the predicted lane sequence.
[0094] In the stage of solving the implicit state sequence, we first construct the objective function of the maximum probability calculation of the implicit state chain, and then iterate the objective function through the Viterbi algorithm to obtain the maximum probability value of the implicit state chain, that is, the state sequence that best matches the observation space.
[0095] The Viterbi algorithm is a dynamic programming algorithm. It is used to find the Viterbi path that is most likely to produce an observation event sequence, that is, the implicit state sequence, that is, to solve the optimal one among all observation sequences. Through the constructed observation space, state space, initial state, state transition probability matrix, emission probability matrix and other model parameters, dynamic programming iterative calculation is performed to solve the maximum probability of the implicit state sequence chain. The objective function in the iterative calculation is as follows:
[0096]
[0097] In the formula, P(q1···q i ···q j ···q m |o1···o i ···o j ···o m ) represents the observation sequence o1···o i ···o j ···o m The corresponding state sequence is q1···q i ···q j ···q m The probability ofi is the i-th state value, o i is the i-th observation, q0 is the initial state, P(q0) is the probability distribution corresponding to the initial state, m is the dimension of the state space, P(q t |q t-1 ) is the value from the state q t-1 Transfer to state value q t The probability, P(q t |o t ) is in the state value q t Next, we get the observed value o t probability.
[0098] Here P(q t |q t-1 ) should be understood as the state transition probability from the lane at time t-1 to the lane at the next time t. The state transition probability matrix is calculated in step 3 mainly to express the lane topological association relationship, so a ij Here P(q t |q t-1 ) is calculated with a ij The calculation process is the same as that of t-1 As lane q i , the state q t As lane q j Substitute lane q i Go to lane q j It can be calculated in the probability calculation formula.
[0099] The above objective function is iteratively calculated through the Viterbi algorithm to obtain the maximum probability value of the implicit state sequence, that is, the predicted lane sequence.
[0100] Step 6: According to the predicted lane sequence, the real-time lane sequence corresponding to the real-time trajectory is restored and output based on the lane-level road network topology relationship.
[0101] After obtaining the maximum probability of the implicit state sequence, the best predicted lane sequence is matched. Through the topological relationship of the lane-level road network, the lane map is radiated outward, the missing lanes are completed, and a spatially continuous lane sequence is obtained.
[0102] When completing lanes, two adjacent but unconnected lanes in the predicted lane sequence are first determined as lane pairs to be completed. Then, based on the lane pair to be completed, the path is radiated outward according to the lane-level road network topology relationship, and the lanes corresponding to the missing shortest path between the lane pairs to be completed are completed, so as to obtain a spatially continuous lane sequence as the real-time lane sequence.
[0103] The path radiation in this embodiment should be understood as path search. For example, taking the upstream lane located upstream in the lane pair to be completed as an example, starting from the upstream lane, all downstream lanes of the upstream lane are obtained according to the lane-level road network topology, and then all downstream lanes obtained are used to continue to search for all downstream lanes of each downstream lane, and the search is continued until the downstream lane located downstream in the lane pair to be completed is found. If only one path is obtained after the search, the path is taken as the shortest path. If multiple paths are obtained after the search, the length of each path is calculated, and the lanes included in the shortest path are taken for completion.
[0104] The real-time lane sequence obtained by this map matching is a sequence that includes the historical lane sequence and the lane where the current positioning point is located. After a match, the historical trajectory with computational efficiency and the new initial position are obtained according to the real-time lane sequence obtained this time, and the initial state, state transition probability matrix, emission probability matrix, etc. in the model are updated to prepare for the matching calculation at the next moment.
[0105] The present invention uses lane-level map matching technology to achieve real-time, reliable, and highly fault-tolerant lane-level map matching for the perception and positioning results of autonomous driving and vehicle-road collaboration, thereby greatly improving the data quality of fusion positioning results. This provides a data basis for fully exploring the rich semantic information of high-precision maps and developing and expanding the use of positioning information.
[0106] In another embodiment, the present application also provides an adaptive lane-level map matching device, including a processor and a memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the adaptive lane-level map matching method when executed by the processor.
[0107] For the specific limitations of the adaptive lane-level map matching device, please refer to the limitations of the adaptive lane-level map matching method above, which will not be repeated here.
[0108] The memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines. The memory stores a computer program that can be run on the processor, and the processor implements the method in the embodiment of the present invention by running the computer program stored in the memory.
[0109] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory is used to store a program, and the processor executes the program after receiving an execution instruction.
[0110] The processor may be an integrated circuit chip with data processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0111] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0112] The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.
Claims
1. An adaptive lane-level map matching method, applied to the localization of a specified object, characterized in that: The adaptive lane-level map matching method is implemented based on a hidden Markov chain model and specifically includes the following steps: Construct the current state space based on the latest positioning of the specified object and the historical lane sequence; Construct the current observation space based on the latest positioning and historical trajectory of the specified object; Based on the state space, construct a state transition probability matrix according to lane topology association; Calculate the mapping probability between positioning in observation space and lanes in state space, and construct the emission probability matrix; By calculating the maximum probability of the implicit state chain, the state sequence that best matches the observation space is output, that is, the predicted lane sequence; According to the predicted lane sequence, a real-time lane sequence corresponding to the real-time trajectory is restored and output based on the lane-level road network topology relationship.
2. The adaptive lane-level map matching method according to claim 1, characterized in that: The current state space is constructed according to the latest positioning of the specified object and the historical lane sequence, including: Match the adjacent lane sets based on the latest positioning of the specified object; Take the N lanes closest to the current moment in the historical lane sequence; The set of matched adjacent lanes is merged with the taken N lanes as the state space.
3. The adaptive lane-level map matching method according to claim 2, characterized in that: The matching adjacent lane set includes: Get the lane group whose shortest distance to the latest positioning is less than the threshold; Take the front and rear lane groups connected to the acquired lane group, and take the acquired lane group and the lanes included in the front and rear lane groups as a lane set.
4. The adaptive lane-level map matching method according to claim 1, characterized in that: The current observation space is constructed according to the latest positioning and historical trajectory of the specified object, including: Starting from the latest positioning, trace back the historical trajectory of the preset length; The historical trajectories obtained by tracing are thinned out and combined with the latest positioning to obtain the observation space.
5. The adaptive lane-level map matching method according to claim 1, characterized in that: The constructing of a state transition probability matrix based on the state space and according to lane topology association includes: The scenarios where the object turns from the current lane to other lanes according to the lane topology association definition include: turning from the current lane to the lane of the same lane group and turning from the current lane to the lane of the downstream lane group; Define the sum of the probabilities of turning from the current lane to the same lane group as σ, where the probability of staying in the current lane is σ*ρ, and the sum of the probabilities of turning to other lanes in the same lane group is σ*(1-ρ); define the sum of the probabilities of turning from the current lane to the lane in the downstream lane group as 1-σ; Then the state transition probability matrix consists of lane q i Go to lane q j The probability calculation formula is as follows: In the formula, a ij For lane q i Go to lane q j The state transition probability of lane q i and Lane q j is the lane in the state space.
6. The adaptive lane-level map matching method according to claim 1, characterized in that: The calculation of the mapping probability between the positioning in the observation space and the lane in the state space and the construction of the emission probability matrix includes: Take the tth observation value in the observation space and the i-th state value in the state space, and calculate the mapping probability as: In the formula, P(q i |o t ) is the state value q i Under these conditions, we get the observed value o t probability, called the mapping probability, the observed value o t is the tth observation value, state value q i is the i-th state value, Distance between q i and o t is the state value q i With the observed value o t The shortest distance between them, m is the dimension of the current state space, that is, the total number of lanes in the state space.
7. The adaptive lane-level map matching method according to claim 1, characterized in that: The maximum probability calculation of the implicit state chain to output the state sequence that best matches the observation space includes: Construct the objective function of the maximum probability calculation of the implicit state chain; The objective function is iteratively calculated through the Viterbi algorithm to obtain the maximum probability value of the implicit state chain, that is, the state sequence that best matches the observation space.
8. The adaptive lane-level map matching method according to claim 7, characterized in that: The objective function is as follows: In the formula, P(q1…q i …q j …q m |o1…o i …o j …o m ) represents the observation sequence o1…o i …o j …o m The corresponding state sequence is q1…q i …q j …q m The probability of i is the i-th state value, o i is the i-th observation, q0 is the initial state, P(q0) is the probability distribution corresponding to the initial state, m is the dimension of the state space, P(q t |q t-1 ) is the value from the state q t-1 Transfer to state value q t The probability, P(q t |o t ) is in the state value q t Next, we get the observed value o t probability.
9. The adaptive lane-level map matching method according to claim 1, characterized in that: The method of restoring and outputting a real-time lane sequence corresponding to the real-time trajectory based on the predicted lane sequence and the lane-level road network topology relationship comprises: Acquire two adjacent but unconnected lanes in the predicted lane sequence as a lane pair to be completed; According to the lane-level road network topology, the paths are radiated outward based on the lane pairs to be completed, and the lanes corresponding to the missing shortest paths between the lane pairs to be completed are completed, so as to obtain a spatially continuous lane sequence as the real-time lane sequence.
10. An adaptive lane-level map matching device, comprising a processor and a memory storing a plurality of computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the adaptive lane level map matching method described in any one of claims 1 to 9 are implemented.
Citation Information
Cited By
Vehicle route matching method and device, processor and vehicle
CN119984318A
