Obstacle screening method, device, medium and program product based on binning algorithm

Abstract

This application provides an obstacle screening method, device, medium, and program product based on a binning algorithm. The method includes: acquiring a set of candidate obstacles, wherein each candidate obstacle includes at least a first position parameter and a second position parameter; dividing the target space into multiple continuous intervals based on a preset binning interval parameter, and creating a left candidate binning container and a right candidate binning container based on the continuous intervals; traversing the set of candidate obstacles, determining the lateral affiliation and the continuous interval to which a candidate obstacle meets preset candidate conditions based on its first and second position parameters, and mapping it to the bins in the corresponding candidate binning container to establish a binning index; traversing the left and right candidate binning containers according to preset hierarchical screening rules to determine the target obstacles to form a target obstacle set and screening according to a preset number. This application reduces computational load and decision latency, and improves the screening reliability in complex dynamic environments.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to an obstacle screening method, device, medium and program product based on binning algorithm. Background Technology

[0002] Unmanned vehicles (such as unmanned land vehicles and unmanned water vessels) typically rely on environmental perception and obstacle avoidance decision-making systems to achieve safe and autonomous navigation. Existing unmanned driving platforms can acquire information about the surrounding environment through lidar, millimeter-wave radar, and visual sensors, and output a set of target obstacles for further processing by trajectory prediction, collision risk assessment, and path planning modules. In actual driving or navigation environments, dynamic obstacles that cross or merge at large angles (such as vehicles or pedestrians crossing intersections, vehicles cutting into the main road from the side, and vessels crossing on water) are typical high-risk objects. The movement direction of such obstacles usually has a large angle with the heading of the vehicle / vehicle, and they can easily enter the collision hazard zone in a short time, thus posing a significant threat to the safe operation of the unmanned platform.

[0003] To reduce collision risk, systems typically need to consider candidate obstacles as comprehensively as possible and perform trajectory prediction and interactive decision-making for potential risk targets. In high-density dynamic scenarios (such as urban intersections, congested road sections, or busy waterways), the number of candidate targets in the perception output may increase significantly. If intensive prediction and decision calculations are performed on all incoming targets, it can easily lead to increased computational load and processing latency, thereby affecting the real-time performance and reliability of obstacle avoidance decisions.

[0004] In existing technologies, obstacle screening methods based on distance thresholds or preset rules are commonly used to reduce computational load. For example, obstacles within a preset distance range in front of the vehicle / boat are given priority processing. However, such screening methods often rely on a single spatial distance indicator or fixed rule conditions, making it difficult to fully characterize the movement trends, interaction relationships, and potential collision risks of obstacles. This can lead to insufficient screening coverage or unstable screening results in complex dynamic scenarios. On the other hand, some solutions attempt to introduce deep learning models to improve the accuracy of obstacle behavior recognition or trajectory prediction. However, due to the large computational load of model inference, it is usually difficult to achieve stable real-time operation on embedded platforms with limited computing resources.

[0005] Therefore, how to quickly select key targets that have a significant impact on obstacle avoidance decisions from a set of multi-target obstacles under the conditions of limited computing power and real-time constraints, and how to balance the effectiveness and spatial coverage of the selection results in complex interactive scenarios, remains a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides an obstacle screening method, device, medium, and program product based on binning algorithms. This addresses the problem that existing technologies, due to limitations in computing power and real-time performance, cannot accurately predict and make obstacle avoidance decisions for all obstacles in high-density dynamic scenarios, which can easily lead to missed screening or insufficient screening coverage of critical high-risk obstacles.

[0007] To achieve the above objectives and other advantages, some embodiments of this application provide the following aspects:

[0008] In a first aspect, some embodiments of this application provide an obstacle screening method based on a binning algorithm, including:

[0009] Obtain a set of candidate obstacles, each candidate obstacle including at least a first position parameter and a second position parameter, the first position parameter and the second position parameter being used to determine the relative positional relationship of the candidate obstacle in the local coordinate system of the controlled intelligent agent;

[0010] Based on the preset binning interval parameter, the target space is divided into multiple continuous intervals along the path reference direction of the controlled intelligent agent, and left candidate binning containers and right candidate binning containers are created based on the continuous intervals respectively.

[0011] Traverse the set of candidate obstacles, and for candidate obstacles that meet the preset candidate conditions, map the candidate obstacles to the corresponding bins in the left candidate bin container or the right candidate bin container according to the first position parameter and the second position parameter of the candidate obstacles, and create bin indexes;

[0012] The candidate obstacles in each compartment are sorted according to the second position parameter, and the left candidate compartment container and the right candidate compartment container are traversed according to the preset hierarchical filtering rules. At least one target obstacle is determined from each compartment to form a target obstacle set.

[0013] The target obstacle set is sorted according to the first position parameter, and a preset number of key obstacles are output as the filtering results.

[0014] Secondly, some embodiments of this application also provide an electronic device, the electronic device comprising:

[0015] One or more processors; and a memory storing computer program instructions that, when executed, cause the processors to perform the obstacle screening method based on the binning algorithm as described above.

[0016] Thirdly, some embodiments of this application also provide a computer-readable storage medium having a computer program and / or instructions stored thereon, which, when executed by a processor, implement the obstacle screening method based on the binning algorithm as described above.

[0017] Fourthly, some embodiments of this application also provide a computer program product, including a computer program and / or instructions, which, when executed by a processor, implement the obstacle screening method based on the binning algorithm as described above.

[0018] Compared with existing technologies, the solution provided in this application obtains the relative position parameters of candidate obstacles in the local coordinate system of the controlled intelligent agent, divides the target space into multiple continuous intervals along the path reference direction to construct left and right candidate bin containers, and maps candidate obstacles to corresponding bins according to their spatial position relationships; sorts candidate obstacles within each bin based on a second position parameter, and determines at least one target obstacle from each bin using a hierarchical screening rule, and then outputs a preset number of key obstacles according to a first position parameter. Therefore, this application can utilize a structured processing mechanism combining spatial binning, bin-based sorting, and hierarchical screening to quickly identify key targets with high threat to the controlled intelligent agent from a large number of candidate obstacles, avoiding intensive calculations on all detected objects one by one, significantly reducing computational load and decision latency, thus ensuring real-time performance and being suitable for operation on embedded platforms with limited computing power. Meanwhile, this application uses spatial distribution as the basis for screening, rather than relying on a single distance threshold rule. This can maintain the spatial coverage and balance of the screening results in complex dynamic environments, enhance the anti-interference ability against environmental disturbances such as wind and fluid or measurement noise, reduce the risk of missing key obstacles, improve the safety and reliability of obstacle avoidance decisions, and have universal adaptability to different traffic scenarios. It can be adapted to the screening of key targets in scenarios such as land vehicles merging at urban intersections and water vessels crossing waterways. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other implementation methods can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an obstacle screening method based on a binning algorithm provided in an embodiment of this application;

[0021] Figure 2This is a schematic diagram of the candidate obstacle bin distribution and key obstacle screening results provided in the embodiments of this application;

[0022] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Some embodiments of this application relate to an obstacle screening method based on a binning algorithm, see below. Figure 1 As shown, the method may include the following steps:

[0025] Step S1: Obtain a set of candidate obstacles. Each candidate obstacle includes at least a first position parameter and a second position parameter. The first position parameter and the second position parameter are used to determine the relative positional relationship of the candidate obstacle in the local coordinate system of the controlled intelligent agent.

[0026] This obstacle selection method based on binning algorithm is applied to the dynamic obstacle avoidance decision-making scenario of unmanned vehicles. Unmanned vehicles refer to the vehicles themselves whose path planning and motion control are performed by autonomous driving or autonomous navigation systems, and can be land vehicles or water vessels. For ease of unified description, unmanned vehicles are regarded as the controlled objects of the algorithm, and are collectively referred to as controlled intelligent agents (also simply referred to as autonomous vehicles or autonomous vessels).

[0027] The candidate obstacle set is generated by the environmental perception module. The environmental perception module may include one or more sensors such as lidar, millimeter-wave radar, camera, AIS (Automatic Identification System for Ships), ultrasonic sensors, and inertial measurement units, to collect sensor data of the environment around the controlled intelligent agent and form target-level perception results through multi-sensor fusion algorithms.

[0028] Candidate obstacles include, but are not limited to, dynamic targets such as vehicles, pedestrians, and non-motorized vehicles crossing or merging at large angles at urban intersections, and ships traversing or intersecting in waterway intersection areas. The motion direction of these targets forms a large angle with the direction of travel or reference path of the controlled intelligent agent, thus constituting candidate decision objects with a high collision risk to the controlled intelligent agent. The path reference direction is the tangent direction of the reference path adopted by the controlled intelligent agent at its current position, i.e., the desired direction of travel, used as the first directional reference of the local coordinate system.

[0029] Each candidate obstacle includes at least a first position parameter and a second position parameter, used to determine the relative position of the candidate obstacle in the local coordinate system of the controlled agent. The local coordinate system of the controlled agent can be set with the agent's own centroid, a positioning reference point, or its geometric center as the origin, the agent's direction of travel or reference path as the first axis direction, and a lateral direction perpendicular to the first axis direction as the second axis direction. For example, in a vehicle scenario, the first axis direction can correspond to the vehicle's forward direction, and the second axis direction can correspond to the vehicle's right-hand direction; in a ship scenario, the first axis direction can correspond to the heading direction or planned trajectory direction, and the second axis direction can correspond to the starboard direction. The first position parameter can be the position coordinates of the candidate obstacle in the first axis direction (e.g., longitudinal relative distance, projected distance along the trajectory direction), and the second position parameter can be the position coordinates of the candidate obstacle in the second axis direction (e.g., lateral relative distance, lateral offset relative to the trajectory). By introducing the first and second position parameters mentioned above, the spatial relative position of candidate obstacles can be expressed under a unified coordinate reference, thereby providing a core positional basis for subsequent determination of bin sections along the path reference direction, selection of left and right bin containers, and sorting and filtering within bins.

[0030] Step S2: Based on the preset binning interval parameters, the target space is divided into multiple continuous intervals along the path reference direction of the controlled agent, and left candidate binning containers and right candidate binning containers are created based on the continuous intervals respectively.

[0031] As an optional embodiment, step S2 specifically includes:

[0032] Step S201: Take the path reference direction as the first direction and the direction perpendicular to the first direction as the second direction. The first direction and the second direction constitute the local coordinate system of the controlled intelligent agent. The origin of the controlled intelligent agent in the local coordinate system is taken as the reference point, and the axis passing through the reference point and extending along the first direction is taken as the reference line. The second position parameter corresponding to the reference line is zero, and the positive direction of the second direction is defined as the right direction and the negative direction is defined as the left direction.

[0033] The path reference direction of the controlled agent at the current moment is obtained. The path reference direction can be the tangent direction of the reference path output by the planning module at the current position of the controlled agent. The path reference direction is defined as the first direction, which is used to represent the forward and backward direction along the planned movement trend of the controlled agent in the target space. At the same time, the direction perpendicular to the first direction is defined as the second direction, which is used to represent the lateral offset direction of the target space relative to the path reference direction of the controlled agent, thus forming the local coordinate system of the controlled agent.

[0034] The local coordinate system uses the location reference point of the controlled agent at that moment as its origin, and the axis extending along the first direction through the origin as its baseline, such that the second position parameter corresponding to the location on the baseline is zero. Furthermore, to achieve a unified determination of the left and right lateral orientation of candidate obstacles, the sign of the second direction is defined: a positive second direction is defined as the right-side direction, and a negative second direction as the left-side direction. Therefore, when a candidate obstacle is located on the positive side of the second direction in the local coordinate system, its second position parameter is greater than zero; when the candidate obstacle is located on the negative side of the second direction, its second position parameter is less than zero. This allows for direct differentiation of whether a candidate obstacle is located to the left or right of the controlled agent based on the sign of the second position parameter.

[0035] In one specific embodiment, the first position parameter and the second position parameter of the candidate obstacle can be obtained by coordinate transformation from the position of the candidate obstacle in the global coordinate system. For example, the global coordinates of the candidate obstacle can be transformed to the local coordinate system based on the positioning and pose information (including position and heading angle) of the controlled intelligent body to obtain the longitudinal distance and lateral distance of the candidate obstacle relative to the controlled intelligent body.

[0036] For example, the position of the controlled agent in the global coordinate system is The heading angle is The position of the candidate obstacle in the global coordinate system is Then, the longitudinal and lateral relative positions of the candidate obstacle in the local coordinate system can be determined by rotation and translation transformation as follows:

[0037]

[0038]

[0039]

[0040] in, The first position parameter is used to characterize the relative position of the candidate obstacle in the path reference direction; The second position parameter characterizes the relative position of the candidate obstacle in the lateral direction and its left / right assignment. When When the candidate obstacle is located to the right of the controlled agent, The time characterization indicates that the candidate obstacle is located to the left of the controlled agent.

[0041] Step S202: Based on the preset bin spacing parameters, the position range of the target space in the path reference direction is divided into multiple continuous intervals.

[0042] Specifically, since the path reference direction has been defined as the first direction in step S201, the position range in the path reference direction in step S202 can be understood as the projection range of the target space in the first direction, that is, the longitudinal range defined by the starting and ending positions of the target space in the first direction under the local coordinate system of the controlled agent. The target space can be the effective decision area or obstacle avoidance decision focus area of ​​the controlled agent at the current moment, such as the space range that includes a certain distance in front of the controlled agent and can cover the approach to obstacles by crossing or intersecting them. The boundary of the first direction range of the target space can be preset according to the path planning and obstacle avoidance requirements, for example, setting the near boundary in the first direction as The far boundary is ,in, It can be set to zero or to the minimum attention distance in front of the controlled agent. This can be taken as the maximum perception distance or the maximum decision distance in front of the controlled agent. Using the above method, the position range of the target space along the path reference direction is obtained. .

[0043] Obtain the preset binning interval parameter ds. This parameter characterizes the length of each interval when dividing the space along the path reference direction. ds can be set based on candidate obstacle density, perception resolution, decision cycle, and computational resource constraints. For example, in urban intersections or waterway junctions with high obstacle density, a smaller ds can be used to improve spatial resolution; in scenarios with low obstacle density or limited computational resources, a larger ds can be used to reduce the number of intervals, thereby reducing subsequent traversal costs. Calculate the number of consecutive intervals N based on the location range and ds. For example, divide the location range length by ds and round up to obtain the number of intervals N. To ensure that the subsequent binning selection process meets real-time requirements, a constraint of a maximum number of segments d can be set on the number of intervals N. When the calculated number of intervals N is greater than the maximum number of segments d, the number of intervals is limited to no more than d to avoid dividing the target space into too many intervals, resulting in excessive computation and storage overhead.

[0044] Furthermore, a validity check is performed on the number of intervals N. If ds does not meet the preset validity conditions (e.g., ds is less than or equal to zero), or if the location range is invalid, resulting in an invalid number of intervals N, the system terminates the subsequent binning process. When the number of intervals N passes the validity check, the location range is divided into N consecutive intervals according to ds, so that each consecutive interval continuously covers the target space in the path reference direction and does not overlap with each other, thereby completing the continuous interval division of the target space. This provides the interval division basis for subsequent steps to establish bins based on continuous intervals and map candidate obstacles to the corresponding bins.

[0045] Step S203: Based on multiple continuous intervals, create a right candidate bin container and a left candidate bin container in the right direction and the left direction, respectively. The right candidate bin container is used to manage the bins of candidate obstacles whose second position parameter is greater than zero, and the left candidate bin container is used to manage the bins of candidate obstacles whose second position parameter is less than zero.

[0046] Specifically, in step S201, the positive direction of the second direction is defined as the right direction, and the negative direction as the left direction, and the corresponding second position parameter is set to zero. Therefore, in step S203, the lateral assignment of the candidate obstacle is determined based on the sign of its second position parameter: when the second position parameter of the candidate obstacle is greater than zero, it indicates that the candidate obstacle is located to the right of the controlled intelligent agent, and it is included in the right candidate bin container for management; when the second position parameter of the candidate obstacle is less than zero, it indicates that the candidate obstacle is located to the left of the controlled intelligent agent, and it is included in the left candidate bin container for management. Through the above left-right division method, candidate obstacles can be diverted to different candidate bin containers according to their lateral assignment, thereby avoiding the complexity of the screening logic caused by the mixing of left and right candidate data.

[0047] In one specific embodiment, the right-side candidate bin container and the left-side candidate bin container are implemented using a two-dimensional vector data structure. The first dimension is used to store the index of the divided interval, and the second dimension is used to store the set of candidate obstacles falling into the corresponding interval. Furthermore, the capacity of the two-dimensional vector data structure can be initialized or dynamically adjusted based on the number of segments determined in step S202. For example, the two-dimensional vector data structure can be expanded or pruned according to the number of segments, so that the length of the first dimension of the container matches the number of consecutive intervals, thereby reducing container out-of-bounds access and lowering additional storage overhead.

[0048] Through steps S201-S203, the target space's position range along the path reference direction is divided into multiple continuous intervals based on preset binning interval parameters. The number of intervals can be controlled by limiting the maximum number of segments, achieving a balance between spatial discretization granularity and real-time computational overhead, avoiding ineffective computation and resource waste caused by excessively dense intervals. This enables structured organization and left-right separation management of candidate obstacles in the path and lateral directions, providing an efficient data organization form for subsequent binning sorting, hierarchical filtering, and key obstacle output.

[0049] Step S3: Traverse the candidate obstacle set. For candidate obstacles that meet the preset candidate conditions, map the candidate obstacles to the corresponding bins in the left candidate bin container or the right candidate bin container according to the first position parameter and the second position parameter of the candidate obstacles, and create bin indexes.

[0050] As an optional embodiment, step S3 specifically includes:

[0051] Step S301: Obtain candidate obstacles and determine whether the candidate obstacles are within the preset position range along the path reference direction in the target space based on the first position parameter.

[0052] The set of candidate obstacles is traversed, and each candidate obstacle is sequentially selected as a processing object. For each candidate obstacle, its first position parameter is obtained, and based on the first position parameter, it is determined whether the candidate obstacle is within the preset position range. For example, starting from the first direction projection position corresponding to the current position of the controlled agent, the range is extended forward to the position corresponding to the maximum attention distance, thus obtaining the longitudinal range boundary. The first position parameter of the candidate obstacle is compared with the boundary of the preset position range: when the first position parameter falls within the preset position range, the candidate obstacle is determined to be within the effective attention range of the target space, and it is allowed to enter the subsequent cross-approach condition determination and binning mapping process; when the first position parameter is not within the preset position range, the candidate obstacle is determined to be outside the longitudinal attention range of the target space, and the subsequent processing of the candidate obstacle is skipped.

[0053] Step S302: When the candidate obstacle is within the preset position range, determine whether the candidate obstacle meets the preset intersection approach condition, wherein the preset intersection approach condition includes: the angle between the movement direction of the candidate obstacle and the path reference direction is greater than the preset angle threshold.

[0054] To prioritize the identification of lateral intrusion candidate obstacles that pose a higher collision risk to the controlled intelligent agent before candidate obstacles enter the binning mapping and hierarchical screening, a preset intersection approach condition judgment is performed on the candidate obstacles. Specifically, when a candidate obstacle passes the preset position range verification in step S301, that is, when its first position parameter is within the preset position range along the path reference direction in the target space, the motion direction parameter of the candidate obstacle is obtained, and the angle θ between the motion direction parameter and the path reference direction is calculated.

[0055] When the calculated included angle θ is greater than the preset angle threshold, the candidate obstacle is determined to meet the preset intersection approach condition, that is, the candidate obstacle shows a clear intersection intrusion trend relative to the path reference direction of the controlled intelligent agent, and can be regarded as crossing or merging into the target at a large angle; when the included angle θ is less than or equal to the preset angle threshold, the candidate obstacle is determined not to meet the preset intersection approach condition, and the subsequent binning mapping and screening process of the candidate obstacle is skipped.

[0056] Large-angle merging is used to characterize a situation where there is a significant deflection angle between the movement direction of a candidate obstacle and the path reference direction of the controlled agent, and the trajectory of the candidate obstacle tends to intrude into or intersect with the path region of the controlled agent in space. Large-angle merging can occur in scenarios where the candidate obstacle and the overall movement direction of the controlled agent are the same, or in scenarios where they are opposite, like a merging from opposite directions.

[0057] For example, in a scenario of merging at a large angle in the same direction, the controlled agent moves along the main road or channel. A candidate obstacle cuts into the controlled agent's path area from a side road, ramp, or adjacent direction at a large merging angle. Although both travel in the same overall direction, the large merging angle and significant lateral intrusion mean the candidate obstacle will still enter the controlled agent's potential collision zone within a short time. In a scenario of merging at a large angle in opposite directions, the controlled agent moves along the path reference direction. A candidate obstacle approaches from the opposite or diagonally opposite direction and performs a left turn, U-turn, or cross-entry operation, causing its movement direction to form a large angle with the path reference direction and intersecting with the controlled agent's trajectory in a local area, thus posing a significant collision risk. By using an angle threshold-based approach determination, high-risk targets such as those crossing the path or merging at large angles in the same or opposite directions can be effectively identified with low computational overhead, and these targets can then be included in the subsequent bin sorting and screening process.

[0058] Step S303: When a candidate obstacle meets the preset intersection approach condition, the interval to which the candidate obstacle belongs in multiple consecutive intervals is determined based on the first position parameter.

[0059] Once a candidate obstacle meets the preset intersection approach condition in step S302, it is necessary to determine its interval affiliation within the multiple continuous intervals divided in step S202, so that it can be used for bin positioning and management in subsequent steps based on interval affiliation. This involves obtaining the first position parameter S of the current candidate obstacle and determining the starting boundary position S along the path reference direction of the target space. min And a preset bin spacing parameter ds is used to perform interval mapping on candidate obstacles. Specifically, the path direction distance ΔS of the candidate obstacle relative to the starting boundary position is determined as ΔS = S - S min The ΔS is discretized into interval indices i based on the bin spacing parameter ds, ensuring that the first position parameter of a candidate obstacle falls within the position range corresponding to the i-th consecutive interval. The interval index i can be determined as follows: i = floor(ΔS / ds) + 1, where the floor function represents the floor function. Thus, the candidate obstacle is mapped to the consecutive interval with interval index i, thereby completing the determination of its assigned interval. Furthermore, boundary constraints can be applied to the interval index i to ensure it remains within the valid index range. For example, when i is less than 1, i is set to 1; when i is greater than the number of consecutive intervals, i is set to the number of consecutive intervals, thus preventing candidate obstacles from being mapped out of bounds and affecting the stability of subsequent bin management.

[0060] Step S304: Determine the lateral assignment of the candidate obstacle based on the second position parameter, so as to determine whether to map the candidate obstacle to the left candidate bin container or the right candidate bin container.

[0061] The lateral assignment of candidate obstacles is determined based on the sign of the second position parameter L: when L is greater than zero, the candidate obstacle is determined to be located to the right of the baseline, its lateral assignment is determined to be right, and it is mapped to the right candidate bin container; when L is less than zero, the candidate obstacle is determined to be located to the left of the baseline, its lateral assignment is determined to be left, and it is mapped to the left candidate bin container. In an optional implementation, when L is equal to zero or |L| is less than a preset lateral zero interval threshold, the candidate obstacle can be assigned to a candidate bin container of a preset default side, or it can be consistently classified according to the historical lateral assignment of the candidate obstacle to improve the mapping stability under measurement noise conditions near zero.

[0062] Step S305: Based on the interval and lateral affiliation, query the bins in the corresponding candidate bin containers. If the bin has not yet been created, create the bin, load the candidate obstacle into the bin, and assign a bin index to the bin when the candidate obstacle is loaded into the bin for the first time.

[0063] Step S303 has determined the interval index to which the candidate obstacle belongs, and step S304 has determined whether its lateral assignment is left or right. In step S305, the corresponding candidate bin container is first selected based on the lateral assignment: the left candidate bin container is selected when the lateral assignment is left, and the right candidate bin container is selected when the lateral assignment is right; then, the bin is queried in the candidate bin container according to the interval index. The bins are used to store the set of candidate obstacles falling within the same interval, so that the candidate obstacles can be organized and managed according to the spatial interval of the path reference direction.

[0064] To avoid inefficient storage and traversal overhead caused by allocating empty bins to intervals without candidate obstacles, this embodiment adopts a sparse bin creation method: the first dimension of the two-dimensional vector is used to provide interval indexing and positioning capabilities, but in the specific bin creation, a bin is created and the candidate obstacle is written at the corresponding interval index position only when a candidate obstacle that meets the preset candidate conditions exists in a certain interval; when there is no candidate obstacle in the interval, the corresponding interval index position is kept empty to avoid empty bins occupying storage resources and reduce the number of subsequent traversals.

[0065] Candidate obstacles are written into the candidate list of the target bin, enabling the target bin to aggregate and manage candidate obstacles belonging to the same interval and lateral affiliation. When a candidate obstacle is first loaded into a target bin, a bin index is assigned. This bin index identifies the bin and supports subsequent target obstacle aggregation and ordered reading. The bin index can include a lateral affiliation identifier and interval index information; for example, it can be divided into a left bin index and a right bin index, allowing left and right bins with the same interval index to be distinguished. For a bin with interval index i, a left bin index L can be assigned to the corresponding bin in the left candidate bin container. i Assign the right bin index R to the corresponding bin in the right candidate bin container. i This is to ensure that the spatial distribution characteristics of the left and right sides are maintained based on the binning index when summarizing the output later.

[0066] Through steps S301-S304, in complex dynamic environments, the obstacle set is rapidly converged into a candidate set that poses a real risk of interaction with the controlled intelligent agent. Within this candidate set, a well-defined spatial structure is formed, elevating the key obstacle selection from a "simple distance threshold rule" to a "structured selection combining intersection and approach risks with spatial distribution characteristics." This significantly reduces redundant computation on numerous low-risk obstacles, concentrating limited computing power on high-risk targets such as those crossing or merging at large angles, improving the real-time performance and stability of online operation. Furthermore, the spatial distribution constraints formed by the path direction interval and lateral separation ensure that candidate targets possess good spatial order before selection, preventing bias in selection results caused by single-point targets or local noise disturbances, thereby enhancing robustness against external disturbances such as wind and fluid flow, as well as sensing and measurement noise.

[0067] Step S4: Sort the candidate obstacles in each sub-box according to the second position parameter, and traverse the left candidate sub-box container and the right candidate sub-box container respectively according to the preset hierarchical filtering rules, and determine at least one target obstacle from each sub-box to form a target obstacle set.

[0068] As an optional embodiment, step S4 specifically includes:

[0069] Step S401: Sort the candidate obstacles in each bin of the left candidate bin container and the right candidate bin container according to the second position parameter to obtain the candidate sequence corresponding to each bin.

[0070] To facilitate the subsequent selection of representative target obstacles from each sub-box according to the preset hierarchical screening rules, the candidate obstacles in each sub-box of the left and right candidate sub-box containers are sorted to form a candidate sequence that can be accessed by serial number.

[0071] To ensure that candidate obstacles with higher sequence numbers are closer to the baseline and more likely to pose a lateral intrusion risk to the controlled agent, different sorting methods can be used based on left and right lateral attribution: For bins in the right candidate bin container, since the second position parameter of the candidate obstacles is positive, they can be sorted in ascending order based on the second position parameter, making the candidate obstacle with sequence number 1 the closest candidate obstacle to the baseline on the right; for bins in the left candidate bin container, since the second position parameter of the candidate obstacles is negative, they can be sorted in descending order based on the second position parameter, making the candidate obstacle with sequence number 1 the closest candidate obstacle to the baseline on the left. Through this sorting method, the candidate obstacle with sequence number 1 in both left and right bins corresponds to the candidate target closest to the baseline under its respective lateral attribution.

[0072] After sorting, the candidate obstacles in each bin are written into the candidate sequence according to the sorting results, and the candidate obstacles in the candidate sequence are assigned a sequence number in sequence. The candidate sequence is used to support subsequent hierarchical filtering by accessing the candidate obstacle with sequence number k using layer number k, so that the system can perform aligned traversal between different bins using the same layer number, thereby realizing the hierarchical filtering logic of progressively selecting target obstacles from each bin according to level.

[0073] Step S402: Determine the left-side stratified screening layer m based on the maximum length of the candidate sequence corresponding to each bin in the left-side candidate bin container, and determine the right-side stratified screening layer n based on the maximum length of the candidate sequence corresponding to each bin in the right-side candidate bin container.

[0074] After the sorting process in step S401, each bin in both the left and right candidate bin containers corresponds to a candidate sequence. The length of the candidate sequence represents the number of candidate obstacles within that bin. Since the dynamic obstacle density may differ across different continuous intervals, the lengths of the candidate sequences corresponding to different bins within the same candidate bin container vary. To ensure that the tiered screening process covers the bin with the most candidate obstacles, the lengths of the candidate sequences corresponding to each bin in the left candidate bin container are obtained, and the maximum value is used as the left-side tiered screening layer number m. Similarly, the lengths of the candidate sequences corresponding to each bin in the right candidate bin container are obtained, and the maximum value is used as the right-side tiered screening layer number n.

[0075] Therefore, when performing layer-by-layer filtering of candidate bins on the left side with layer numbers k=1 to m, bins with candidate sequence lengths less than k can skip the access process of that layer directly; the same logic is used when performing layer-by-layer filtering of candidate bins on the right side with layer numbers k=1 to n, so as to ensure the coverage of the maximum candidate depth while avoiding unnecessary traversal overhead for sparse bins.

[0076] Step S403: Perform stratified filtering on each sub-box in the left candidate sub-box container based on the left stratified filtering layer number m, and perform stratified filtering on each sub-box in the right candidate sub-box container based on the right stratified filtering layer number n, so as to identify at least one target obstacle from each sub-box.

[0077] As an optional embodiment, in step S403, a hierarchical filtering process is performed on each bin in the left candidate bin container based on the left hierarchical filtering layer number m, specifically including:

[0078] Step SA4031: Obtain the first candidate sequence corresponding to each bin in the left candidate bin container;

[0079] Step SA4032: In the first layer of screening, for each bin in the left candidate bin container, select the candidate obstacle with the sequence number 1 from the corresponding first candidate sequence as the left first layer target obstacle, and determine the second position parameter of the left first layer target obstacle as the left reference position parameter of the bin in each subsequent layer of screening.

[0080] Since the second position parameters of the candidate obstacles in the left candidate bin container are all less than zero, and the larger the value of the second position parameter, the closer it is to the baseline, this embodiment sorts the candidate obstacles in each bin in the left candidate bin container in descending order according to the second position parameter to form the first candidate sequence corresponding to each bin, so that the candidate obstacle with the sequence number 1 is the candidate object closest to the baseline in that bin.

[0081] In the first-level screening, for each bin in the left-side candidate bin container, the candidate obstacle with sequence number 1 is selected from the corresponding first candidate sequence as the left-side first-level target obstacle. Since the first candidate sequence is obtained by sorting by the second position parameter, the candidate obstacle with sequence number 1 has determinism and representativeness in lateral position and can serve as the benchmark reference target for that bin within the current continuous interval. The second position parameter of the left-side first-level target obstacle is determined as the left-side benchmark position parameter corresponding to that bin, ensuring that the subsequent second to m-th layers of screening all use this left-side benchmark position parameter as a unified reference for lateral offset constraints. This allows subsequent layers of screening to supplement the candidate sequence with candidate obstacles that meet the conditions without changing the representative reference of the bin, thereby avoiding screening reference drift and improving the stability and repeatability of the layered screening results. At the same time, the first-level screening ensures that each bin outputs at least one representative target obstacle, thus guaranteeing the coverage of the screening results in spatial distribution.

[0082] Step SA4033: In the filtering of layers 2 to m, the layers k from 2 to m are traversed layer by layer. At the beginning of each layer filtering, the current position parameter on the left is initialized to the preset left boundary parameter. The preset left boundary parameter is the second position parameter corresponding to the maximum lateral boundary of the preset position range of the target space along the path reference direction on the left.

[0083] Layer number k is traversed layer by layer from 2 to m to supplement the candidate sequences of each bin with representative candidate obstacles at different levels. To ensure that each layer of the screening process has a unified initial constraint boundary and to avoid coupling interference between the screening results of the previous layer and the screening results of the next layer, the left current position parameter used for the intra-layer iteration is initialized at the beginning of each layer of screening.

[0084] In one specific embodiment, the left current position parameter is initialized to a preset left boundary parameter. The preset left boundary parameter is determined by the maximum lateral boundary of the preset position range of the target space along the path reference direction in the left direction, that is, the second position parameter corresponding to the left outer boundary position of the target space in the local coordinate system of the controlled agent is used as the initial value. Thus, at the beginning of this layer of screening, the left current position parameter is first set as the lateral extreme boundary characterizing the allowable range of the target space. Subsequently, during the screening process of this layer, as each bin of candidate obstacles is traversed sequentially, this parameter is gradually updated to the lateral position of the candidate obstacle that better meets the current layer of screening constraints, thereby realizing an iterative screening process that converges from the spatial boundary to the optimal lateral position inside the candidate set.

[0085] By initializing with a preset left boundary parameter at the beginning of each layer of screening, it is possible to ensure that the screening of that layer has a consistent comparison benchmark when traversing different bins, avoiding the failure of intra-layer updates due to initial values ​​that are too small or too large. On the other hand, since this initialization is re-executed between different layers, the screening of each layer is independent of each other, thereby enabling the supplementary screening of candidate obstacles to be completed layer by layer while keeping the bin benchmark position parameters unchanged, thus improving the stability and repeatability of the screening process.

[0086] Step SA4034: In each layer of filtering, for each bin in the left candidate bin container, the following steps are performed sequentially: When the length of the first candidate sequence corresponding to the bin is not less than the layer number k, the candidate obstacle with the sequence number k in its first candidate sequence is obtained as the current candidate obstacle. When the absolute value of the difference between the second position parameter of the current candidate obstacle and the left reference position parameter corresponding to the bin is less than or equal to the preset left distance threshold, and the second position parameter of the current candidate obstacle is greater than the current position parameter on the left, then the current candidate obstacle is determined as the target obstacle, and the current position parameter on the left is updated to the second position parameter of the target obstacle.

[0087] Specifically, during the selection process at the k-th level (where k is from 2 to m), each bin in the left-hand candidate bin container is traversed sequentially according to its bin index to supplement representative obstacles that satisfy lateral constraints from the candidate sequences of different bins at the same level. For any bin to be processed, it is first determined whether the length of the first candidate sequence corresponding to the bin is not less than the level number k; when the length of the first candidate sequence meets the requirement, the candidate obstacle with the index k is obtained from the first candidate sequence as the current candidate obstacle, so that different bins use the k-th candidate obstacle as the comparison object in the same level selection, ensuring the consistency of the candidate supplementation process of each bin at the level.

[0088] Subsequently, two levels of constraint judgment are performed on the current candidate obstacle.

[0089] First, lateral proximity constraint: Determine whether the absolute value of the difference between the second position parameter of the current candidate obstacle and the left reference position parameter of the bin is less than or equal to a preset left distance threshold. When the absolute value of the difference is less than or equal to the preset left distance threshold, it is determined that the current candidate obstacle meets the lateral proximity constraint with the reference target of the bin, thereby avoiding the candidate obstacle deviating too much from the representative lateral position of the bin and ensuring that the spatial distribution characteristics of the candidate selections in each layer within the same bin remain consistent.

[0090] Second, the minimum lateral convergence constraint across bins: Given that the current candidate obstacle satisfies the lateral proximity constraint, it is further determined whether the second position parameter of the current candidate obstacle is greater than the current position parameter on the left. When this condition is met, it indicates that the current candidate obstacle is closer to the baseline in the left direction than the selected target (the second position parameter is negative; a larger value indicates a smaller lateral offset, meaning the current candidate obstacle's absolute lateral distance to the baseline is smaller), thus possessing higher representativeness of interactive risk. At this point, the current candidate obstacle is determined as the target obstacle for this layer's screening, and the current position parameter on the left is updated to the target obstacle's second position parameter. This ensures that subsequent bins continue to use the position closer to the baseline as the comparison threshold during the same layer's screening process, thereby gradually converging to obtain the most representative key target of this layer during cross-bin traversal.

[0091] As an optional embodiment, in step S403, a hierarchical filtering process is performed on each bin in the candidate bin container on the right based on the number of hierarchical filtering layers n on the right, specifically including:

[0092] Step SB4031: Obtain the second candidate sequence corresponding to each bin in the right candidate bin container;

[0093] Step SB4032: In the first layer of screening, for each bin in the candidate bin container on the right, select the candidate obstacle with the sequence number 1 from the corresponding second candidate sequence as the target obstacle of the first layer on the right, and determine the second position parameter of the target obstacle of the first layer on the right as the right reference position parameter of the bin in each subsequent layer of screening.

[0094] Step SB4033: In the filtering of layers 2 to n, the layers k from 2 to n are traversed layer by layer. At the beginning of each layer filtering, the current position parameter on the right is initialized to the preset right boundary parameter. The preset right boundary parameter is the second position parameter corresponding to the maximum lateral boundary of the preset position range of the target space along the path reference direction on the right.

[0095] Step SB4034: In each layer of filtering, for each bin in the right candidate bin container, the following steps are performed sequentially: When the length of the second candidate sequence corresponding to the bin is not less than the layer number k, the candidate obstacle with the sequence number k in its second candidate sequence is obtained as the current candidate obstacle. When the absolute value of the difference between the second position parameter of the current candidate obstacle and the right reference position parameter corresponding to the bin is less than or equal to the preset right distance threshold, and the second position parameter of the current candidate obstacle is less than the right current position parameter, then the current candidate obstacle is determined as the target obstacle, and the right current position parameter is updated to the second position parameter of the target obstacle.

[0096] Specifically, after completing the spatial binning of candidate obstacles, multiple bins are formed in the right-side candidate bin container. Each bin stores candidate obstacles associated with its corresponding interval and located to the right. To support hierarchical filtering, the list of candidate obstacles stored in each right-side bin is retrieved from the right-side candidate bin container, and a corresponding second candidate sequence is constructed for each right-side bin. The second candidate sequence represents the ordered arrangement of right-side candidate obstacles within each bin, allowing for the sequential extraction and comparison of candidate obstacles in subsequent hierarchical filtering.

[0097] The second candidate sequence is sorted in ascending order according to the second position parameter of the candidate obstacles, so that the lateral offset corresponding to the candidate obstacle with the sequence number 1 is minimized, that is, it is closer to the baseline in the right direction. This allows the first layer of screening to prioritize candidate targets with higher potential interaction risks to the controlled agent. To ensure a consistent reference for the lateral offset constraints of the same bin in subsequent layers of screening, the second position parameter of the first-layer target obstacle on the right is determined as the right-side reference position parameter for that bin. The right-side reference position parameter is used to characterize the reference position for the lateral screening of that bin in subsequent layers 2 to n, so that subsequent layers of screening can supplement the candidate sequence without changing the bin's representative reference, thereby maintaining the lateral consistency of target obstacle selection within the same bin.

[0098] The current position parameter on the right is initialized to a preset right boundary parameter. The preset right boundary parameter can be determined by the second position parameter corresponding to the maximum lateral boundary of the preset position range along the path reference direction in the target space on the right. For example, in the local coordinate system of the controlled agent, the system can set the maximum lateral boundary on the right to +10m or +20m to characterize the maximum lateral range where candidate obstacles on the right may appear, thereby ensuring that the current position parameter on the right has a sufficiently large initial comparison threshold when the cross-bin traversal of the filtering layer begins.

[0099] For any layer number k (k is 2 to n), during the screening process of that layer, each bin in the candidate bin container on the right is traversed sequentially according to the bin index on the right, so as to supplement the selection of representative obstacles that satisfy the lateral constraints from the second candidate sequence of different bins at the same level.

[0100] When traversing to a certain right-side bin, first determine whether the length of the second candidate sequence corresponding to the right-side bin is not less than the layer number k; when the length of the second candidate sequence meets the requirement, obtain the candidate obstacle with the sequence number k from the second candidate sequence as the current candidate obstacle, so that different bins use the k-th candidate obstacle as the comparison object in the same layer screening, ensuring the consistency of the candidate supplementation process of each bin at the layer level.

[0101] Subsequently, a two-level constraint assessment is performed on the current candidate obstacle. This assessment, in addition to satisfying the lateral proximity constraint, further satisfies the minimum lateral convergence constraint across bins to determine whether to update the current candidate obstacle as the target obstacle. The processing mechanism of the two-level constraint assessment in the right-side hierarchical screening mechanism is similar to that in the left-side hierarchical screening mechanism, except that the second position parameter of the current candidate obstacle in the right direction is positive, and the smaller the value of this second position parameter, the closer the candidate obstacle is to the baseline in the lateral direction. Therefore, in the lateral convergence constraint, when the second position parameter of the current candidate obstacle is less than the current position parameter on the right, it is determined that the current candidate obstacle meets the lateral convergence requirement, and the current position parameter on the right is updated to the second position parameter of the current candidate obstacle. The remaining processing is consistent with the left-side hierarchical screening mechanism and will not be described further here.

[0102] In one implementation, the left-side hierarchical filtering process and the right-side hierarchical filtering process are independent of each other. The system can perform hierarchical filtering on the left-side candidate bin containers and the right-side candidate bin containers in a preset processing order. In another implementation, to improve real-time performance, the system can use a parallel processing method to execute the left-side hierarchical filtering process and the right-side hierarchical filtering process simultaneously. For example, the hierarchical filtering of the left-side candidate bin containers and the right-side candidate bin containers can be assigned to different threads, different processing cores or different asynchronous tasks, thereby shortening the overall filtering time without increasing the number of filtering layers.

[0103] Step S404: Combine the target obstacles identified in each compartment of the left candidate sub-container and the right candidate sub-container to form a target obstacle set.

[0104] After the candidate bins on the left and right sides have completed their stratified screening, the target obstacles identified by each bin are aggregated to form a target obstacle set for subsequent trajectory prediction, risk assessment, or collision avoidance decisions. Since the left and right candidate bins manage candidate obstacles in different lateral regions, simply merging them could easily lead to a bias in the lateral distribution of the output targets, reducing the representativeness of the screening results for spatial risk distribution. Therefore, this embodiment introduces a unified target container and an ordered reading mechanism based on bin indexes during the aggregation stage. This ensures that the target obstacle set maintains a balanced spatial distribution on both sides while advancing longitudinally, thereby improving the adaptability and stability of the key target set in complex traffic scenarios.

[0105] As an optional embodiment, step S404 specifically includes:

[0106] Step S4041: Construct a unified target container, which includes a mapping structure with bin indexes as keys.

[0107] To achieve structured aggregation and ordered output of the filtering results from both sides, a unified target container is constructed to manage the target obstacles output by the left and right candidate bin containers. The unified target container can be implemented using a mapping structure, such as a hash table, dictionary, or associative array, to support fast key-value location and write operations. The unified target container includes a mapping structure with bin indices as keys and target lists as values. The bin index uniquely identifies the bin corresponding to each output target obstacle. The bin index includes at least a range index and a lateral attribution identifier to distinguish the output targets from the left and right bins within the same range. For example, the range index indicates the consecutive range number where the target obstacle is located, and the lateral attribution identifier indicates whether the target obstacle belongs to the left or right candidate bin container, thus enabling the unified target container to maintain both left and right target lists under the same range index. The target list is used to store one or more target obstacles associated with the corresponding bin index. The target list includes at least the first target obstacle obtained from the bin filtering, and can be extended to store multiple target obstacles to support the target obstacles output in subsequent target filtering processes at different levels.

[0108] By constructing the unified target container described above, we can merge and manage the targets on both sides using the bin index as the primary key of organization. This provides a data structure foundation for subsequent ordered reading based on the incrementing of the index of the corresponding interval and the alternation of left and right sides, and ensures that the output targets maintain the consistency and reproducibility of spatial distribution characteristics.

[0109] Step S4042: Write the target obstacles determined by each bin in the left candidate bin container and the right candidate bin container into the target list associated with the corresponding bin index in the unified target container.

[0110] The target obstacles selected from each bin in the left and right candidate bin containers are written into a unified target container to aggregate the filtering results across containers. Specifically, for any bin in the left candidate bin container, the bin index corresponding to that bin is read, and the target list associated with that bin index is located in the unified target container. If the target list has not yet been created, it is created in the unified target container, and the target obstacles identified in that bin are written into that target list. The same writing operation is performed for the right candidate bin container. Thus, the target obstacles on both sides are written into the target lists in the unified target container that correspond one-to-one with their bin indices. This allows the subsequent reading stage to achieve structured output based on the bin index, belonging to intervals and lateral directions, rather than simply concatenating the results from the left and right sides. This ensures that the output order is controllable and reproducible from the underlying data structure level.

[0111] Step S4043: Read the target list in the unified target container in ascending order of the interval index corresponding to the bin index to form a target obstacle set. Under the same interval index, read the corresponding target list and output the target obstacles in an alternating manner of the left bin index and the right bin index to maintain the spatial distribution characteristics of the target obstacle set.

[0112] To form a set of target obstacles and maintain their spatial distribution characteristics, a unified target container is read in an ordered manner according to a preset reading strategy. Specifically, the keys of the unified target container are first sorted according to the ascending order of the interval indices corresponding to the binning indices, and then the target lists corresponding to each interval index are read sequentially.

[0113] When reading the target list under the same interval index, the system does not always read the left side first and then the right side. Instead, it uses an alternating reading method. For example, under the same interval index, the system first reads the target list associated with the left bin index and outputs its target obstacles, then reads the target list associated with the right bin index and outputs its target obstacles. Alternatively, the system can read the right side first and then the left side, or use a rotation strategy to perform alternating reading. Through this alternating reading mechanism, the system avoids the problem of output bias towards one side due to a larger number of candidate targets on one side. This ensures that the target obstacle set covers potential risk sources on both the left and right sides within the same vertical interval, thus maintaining the lateral spatial distribution characteristics of the target obstacle set.

[0114] For example, the target space can be divided into multiple consecutive intervals along the path reference direction: Interval 1 (10m~20m), Interval 2 (20m~30m), and Interval 3 (30m~40m). Each interval has a left bin and a right bin, and the target list corresponding to each bin stores the target obstacles. The system reading order can then be: left target of Interval 1 → right target of Interval 1 → left target of Interval 2 → right target of Interval 2 → left target of Interval 3 → right target of Interval 3. By outputting the target obstacles in a left-right staggered manner within each interval, the target obstacle set maintains a balanced coverage of risk distribution on both sides while advancing vertically, thereby avoiding insufficient spatial representativeness caused by a bias towards one side of the key target set.

[0115] Furthermore, during the process of reading the data incrementally according to the index of the corresponding interval, the output target obstacle set is ensured to advance from near to far along the path reference direction, thereby ensuring that the resulting target obstacle set possesses both vertical interval ordering and lateral distribution balance. This set can be directly used in subsequent key target prediction and decision-making modules to obtain spatially representative key obstacle inputs within a limited computational budget, thereby improving the stability and safety of obstacle avoidance decisions.

[0116] Through steps S401-S404, this embodiment of the application can sort and stratify candidate obstacles within the left and right bins, and summarize the target obstacles output by each bin, thereby quickly obtaining a set of key target obstacles under limited computing power constraints. Compared with rule-based filtering based on a single distance threshold, this scheme can maintain the orderly output of target obstacles along the path reference direction while taking into account the spatial distribution characteristics of the left and right sides. It can provide low-redundancy, high-information input for subsequent trajectory prediction and obstacle avoidance decisions, significantly reducing the system's computational load and decision latency, and improving the reliability and environmental adaptability of key target identification.

[0117] Step S5: Sort the target obstacle set according to the first position parameter, and output a preset number of key obstacles as the filtering results.

[0118] In this embodiment, after forming the target obstacle set, the target obstacle set is further sorted based on the first position parameter, and a preset number of key obstacles are output as the final screening result, so that the subsequent trajectory prediction and collision avoidance decision module can complete real-time processing under a fixed computational budget. Specifically, the corresponding first position parameter is extracted for each target obstacle in the target obstacle set, and the obstacles are sorted in ascending order of the first position parameter, so that the target obstacles ranked higher represent risk targets that are closer to the controlled intelligent agent and have a more urgent need for collision avoidance decision. The first position parameter can be selected as the longitudinal center position, longitudinal starting position, or longitudinal projection distance of the target obstacle, which can characterize the relative distance of the target obstacle in the path reference direction, thereby adapting to different traffic forms such as vehicles and ships and different expression methods of perceived targets.

[0119] After sorting, a set of key obstacles is extracted and output from the sorting results according to a preset number. For example, the preset number can be set to M (such as 8, 10 or 12) to match the number of targets that the subsequent prediction module can process in parallel; when the number of target obstacles is greater than the preset number, the top M target obstacles are output; when the number of target obstacles is less than or equal to the preset number, all target obstacles are output as the filtering result.

[0120] Through the above processing, it can be ensured that the output set of key obstacles has a clear longitudinal distance priority, which not only meets the requirement of prioritizing the response of obstacle avoidance decision to nearby risk targets, but also constrains the number of key targets within a preset range, thereby reducing the computational load of subsequent modules and improving real-time performance and stability.

[0121] Figure 2 This diagram illustrates the spatial binning distribution of candidate obstacles and the key obstacle screening results in this embodiment. In the diagram, blue rectangles represent the controlled intelligent agent, and numbered rectangles represent candidate obstacles; dashed lines represent multiple continuous intervals divided along the path reference direction of the controlled intelligent agent according to a preset binning interval parameter ds, thus forming multiple bins within the candidate binning containers on the left and right sides; h represents a preset left-side distance threshold, used to constrain the lateral offset range of candidate obstacles during the hierarchical screening process; red rectangles in the diagram represent the key obstacle output results selected from the candidate obstacles.

[0122] It should be noted that this embodiment uses a straight navigation route as an example for illustration, but this application is not limited to this. For curved paths or reference paths in the form of polylines, the path reference direction can also be determined based on the tangent direction of the reference path and the method of this application can be implemented.

[0123] In this embodiment, the environmental perception module outputs a set of candidate obstacles, which includes 10 candidate obstacles (IDs 1 to 10). Each candidate obstacle includes at least a first position parameter and a second position parameter. The first position parameter characterizes the relative position of the candidate obstacle along the path reference direction of the controlled agent (which can be understood as longitudinal distance or projected distance along the reference direction), and the second position parameter characterizes the lateral position of the candidate obstacle relative to the path reference direction (which can be understood as lateral distance or lateral offset). The location of the controlled agent is taken as the origin of the local coordinate system, and a baseline passing through the origin and extending along the path reference direction is used as the lateral zero-value reference, such that the second position parameter corresponding to this baseline is zero. A second position parameter greater than zero indicates the obstacle is located in the right direction, and a second position parameter less than zero indicates the obstacle is located in the left direction.

[0124] The preset binning interval parameter ds is set to 3m. Within a preset position range along the path reference direction, the target space is divided into multiple continuous intervals with ds as the binning interval. Each continuous interval corresponds to a segment along the path reference direction. Subsequently, left and right candidate binning containers are created to manage the binning of candidate obstacles with a second position parameter less than zero and candidate obstacles with a second position parameter greater than zero, respectively. The binning in the left and right candidate binning containers is sparsely created, meaning that a bin is created only if a candidate obstacle meeting the conditions exists in the corresponding continuous interval.

[0125] The candidate obstacle set is traversed. For each candidate obstacle, it is first determined whether it is within a preset position range along the path reference direction in the target space based on the first position parameter. If it is within the preset position range, it is further determined whether it meets the preset intersection approach condition (e.g., the angle between the candidate obstacle's movement direction and the path reference direction is greater than a preset angle threshold) so that candidate obstacles that cross or merge at large angles are used as valid candidate objects for bin sorting.

[0126] When a candidate obstacle meets the aforementioned preset candidate conditions, its assigned interval is determined based on its first position parameter, and its lateral affiliation is further determined based on its second position parameter, thus mapping it to the bin corresponding to its assigned interval in the left or right candidate bin container. For bins that have not yet been created, a corresponding bin is created and a bin index is assigned when a candidate obstacle is first written. Figure 2 Taking the scenario shown as an example, the mapping results are as follows: obstacles 1 and 2 are mapped to the left bin 1 of the left candidate bin container; obstacles 3 and 4 are mapped to the left bin 2; obstacles 6 and 7 are mapped to the left bin 3; obstacles 5 are mapped to the right bin 1; obstacles 8 are mapped to the right bin 2; obstacles 9 and 10 are mapped to the right bin 3.

[0127] After mapping, the left and right candidate bin containers are sorted and stratified for selection. First, within each bin, candidate obstacles are sorted according to the second position parameter to obtain a candidate sequence. This ensures that obstacles with smaller absolute values ​​of the second position parameter (i.e., closer to the path reference direction baseline) are placed at the beginning of the candidate sequence, prioritizing obstacles with greater invasiveness to the controlled agent's path as representative targets. Then, the number of stratified selection layers for the left and right candidate bin containers is determined based on the maximum length of the candidate sequence in each bin. In this embodiment, the maximum candidate sequence length for the left candidate bin container is 2, so the number of stratified selection layers for the left is m=2; the maximum candidate sequence length for the right candidate bin container is 2, so the number of stratified selection layers for the right is n=2. Furthermore, a preset lateral distance threshold h=20m is set to constrain the lateral offset range of target obstacles in subsequent stratified selection.

[0128] In the first layer of screening of the left candidate bin container, for left bin 1, left bin 2, and left bin 3, the candidate obstacle with the sequence number 1 is selected from the corresponding candidate sequence as the target obstacle of the first layer on the left, i.e., obstacle 1, obstacle 3, and obstacle 6 are selected. In the second layer of screening, the left current position parameter Lmin1 is initialized with a preset left boundary parameter, so that Lmin1 takes a sufficiently outer initial threshold at the beginning of this layer of screening. Subsequently, according to the bin index order in the left candidate bin container, the candidate obstacle with the sequence number 2 in the first candidate sequence of each bin is obtained as the current candidate obstacle, and a joint judgment of lateral proximity constraint and cross-bin minimum lateral convergence constraint is performed on it; wherein, the lateral proximity constraint is used to limit the lateral offset of the current candidate obstacle relative to the left reference position parameter of the bin to not exceed the preset lateral distance threshold h, and the cross-bin minimum lateral convergence constraint is used to limit the current candidate obstacle to have a better lateral position relative to the comparison threshold, i.e., the left current position parameter Lmin1, in this layer of screening.

[0129] by Figure 2Taking the left bin as an example, when traversing the left bin 1, if the lateral offset of obstacle 2 exceeds the preset lateral distance threshold h (for example, the lateral distance represented by its second position parameter satisfies |L2|>(|L1|+h)), then obstacle 2 will not be selected as the target obstacle of this layer, and the current position parameter Lmin1 on the left remains unchanged at its initial value. When traversing to the left bin 2, if obstacle 4 satisfies the lateral offset constraint (for example, |L4|≤(|L3|+h)), and its lateral position is better than the current threshold (L4>Lmin1, where Lmin1 is still the preset left boundary parameter), then obstacle 4 will be identified as the newly added target obstacle of this layer, and the current position parameter Lmin1 on the left will be updated to the second position parameter corresponding to obstacle 4 (Lmin1=L4). This ensures that when traversing other bins in this layer, the comparison threshold will be updated and iterated to a position closer to the baseline, thereby prompting the selection in the same layer to continuously converge towards a better lateral position during cross-bin traversal. Although obstacle 7 satisfies the lateral proximity constraint, it does not satisfy the minimum lateral convergence constraint across the bins (i.e., |L7|≤(|L6|+h), but L7≤Lmin1), and is therefore not considered a target obstacle. The final target obstacles on the left include: obstacle 1, obstacle 3, obstacle 4, and obstacle 6.

[0130] The stratified screening process for the right candidate bin container is similar to that for the left candidate bin container. In the first layer of screening for the right candidate bin container, for right bin 1, right bin 2 and right bin 3, the candidate obstacle with the sequence number 1 is selected from the corresponding candidate sequence as the target obstacle for the first layer on the right, that is, obstacle 5, obstacle 8 and obstacle 10 are selected.

[0131] In the second layer of screening, the right-side current position parameter Lmin2 is initialized with a preset right-side boundary parameter, so that Lmin2 takes a sufficiently outer initial threshold at the start of this layer of screening. Subsequently, according to the order of the bin index in the right-side candidate bin container, the candidate obstacle with the index 2 in the second candidate sequence of each bin is obtained as the current candidate obstacle, and a joint judgment of lateral proximity constraint and cross-bin minimum lateral convergence constraint is performed on it. Among them, the lateral proximity constraint is used to limit the lateral offset of the current candidate obstacle relative to the right-side reference position parameter of the bin to not exceed the preset lateral distance threshold h, and the cross-bin minimum lateral convergence constraint is used to limit the current candidate obstacle to have a better lateral position relative to the comparison threshold, that is, the right-side current position parameter Lmin2, in this layer of screening.

[0132] by Figure 2Taking the right-side bin as an example, when traversing right-side bin 3, if the lateral offset of obstacle 9 exceeds the preset lateral distance threshold h (for example, the lateral distance represented by its second position parameter satisfies L9>(L10+h)), then obstacle 9 will not be selected as a target obstacle in this layer, and the current position parameter Lmin2 on the right side will remain unchanged at its initial value. Since no new target obstacles satisfying the joint constraints appear during this layer traversal, after the second layer of filtering, the target obstacles on the right side are still composed of the results of the first layer of filtering. Finally, the target obstacles on the right side include: obstacle 5, obstacle 8, and obstacle 10.

[0133] The candidate merging process is performed on the screening results on the left and right sides. The target obstacles are written into the target list associated with their bin index in the unified target container. The target list is read from near to far according to the index of the interval to which they belong. The target obstacles are output in a left-right staggered manner within the same interval to obtain the merged target obstacle set.

[0134] To ensure the output meets the processing order requirements of the subsequent obstacle avoidance decision-making module, the system sorts the target obstacle set according to the first position parameter, arranging the target obstacles from closest to furthest from the controlled agent. Finally, a preset number of key obstacles are output from the sorted target obstacle set as the selection results; for example, in this example, when the preset output quantity M is 5, the first 5 target obstacles are selected as key obstacles from the sorted target obstacle set and their identifiers are output. The final output key obstacles are obstacle 1, obstacle 3, obstacle 4, obstacle 5, and obstacle 6 (corresponding to...). Figure 2 (The red rectangular obstacle in the image).

[0135] Compared with existing technologies, the obstacle screening method based on binning algorithm provided in this application, through the collaborative mechanism of spatial binning, hierarchical screening and sorting output, can screen a limited number of representative key obstacles from the candidate obstacle set with low computational overhead in scenarios with a large number of candidate obstacles and complex dynamic interactions. This avoids performing dense prediction and collision avoidance calculations on all obstacles, balances real-time performance and safety, effectively optimizes computing resources and improves the decision reliability of the autonomous driving platform.

[0136] Furthermore, the screening process in this embodiment is based on the relative positional relationships and spatial distribution patterns of candidate obstacles in the local coordinate system of the controlled intelligent agent, without relying on prior knowledge of specific environments such as road structures, waterway structures, or fixed lane boundaries. Therefore, it is applicable to unstructured or weakly structured traffic environments. It can be universally deployed under different traffic modes. For example, it can be applied to obstacle screening for land vehicles in high-density dynamic scenarios such as urban intersections, and also to the identification and output of key obstacles in complex water environments such as cross-navigation of water vessels.

[0137] In summary, the obstacle screening method based on binning algorithm provided in this application obtains the relative position parameters of candidate obstacles in the local coordinate system of the controlled intelligent agent, divides the target space into multiple continuous intervals along the path reference direction to construct left and right candidate bin containers, and maps candidate obstacles to corresponding bins according to their spatial position relationships. Candidate obstacles within each bin are sorted based on a second position parameter, and at least one target obstacle is determined from each bin using a hierarchical screening rule. Finally, a preset number of key obstacles are output according to the first position parameter. Therefore, this application can utilize a structured processing mechanism combining spatial binning, bin-based sorting, and hierarchical screening to quickly identify key targets with high threat to the controlled intelligent agent from a large number of candidate obstacles, avoiding intensive calculations on all detected objects one by one, significantly reducing computational load and decision latency, thus ensuring real-time performance and suitability for operation on embedded platforms with limited computing power. Meanwhile, this application uses spatial distribution as the basis for screening, rather than relying on a single distance threshold rule. This can maintain the spatial coverage and balance of the screening results in complex dynamic environments, enhance the anti-interference ability against environmental disturbances such as wind and fluid or measurement noise, reduce the risk of missing key obstacles, improve the safety and reliability of obstacle avoidance decisions, and have universal adaptability to different traffic patterns.

[0138] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.

[0139] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, etc. The electronic device can also be various forms of mobile devices, such as cellular phones, smartphones, wearable devices, and other similar computing devices.

[0140] The electronic device includes: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform an obstacle screening method based on a binning algorithm as provided in any one or more of the above embodiments. Figure 3An exemplary structural diagram of the electronic device is disclosed. The electronic device includes one or more processors 1101, a memory 1102, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0141] The electronic device may further include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103, and output device 1104 may be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0142] Input device 1103 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display, a light-emitting diode display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0143] To provide interaction with the user, the electronic device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).

[0144] In this embodiment, a computer-readable medium stores a computer program / instructions, which, when executed by a processor, implement the obstacle screening method based on a binning algorithm provided in any one or more of the above embodiments. The computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more computer-readable instructions.

[0145] The memory 1102 can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.

[0146] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0147] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0148] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0149] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0150] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0151] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0152] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0153] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. Terms such as "first," "second," etc., are used only to distinguish descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. An obstacle screening method based on binning algorithm, characterized in that, include: Obtain a set of candidate obstacles, each candidate obstacle including at least a first position parameter and a second position parameter, the first position parameter and the second position parameter being used to determine the relative positional relationship of the candidate obstacle in the local coordinate system of the controlled intelligent agent; Based on the preset binning interval parameter, the target space is divided into multiple continuous intervals along the path reference direction of the controlled intelligent agent, and left candidate binning containers and right candidate binning containers are created based on the continuous intervals respectively. Traverse the set of candidate obstacles, and for candidate obstacles that meet the preset candidate conditions, map the candidate obstacles to the corresponding bins in the left candidate bin container or the right candidate bin container according to the first position parameter and the second position parameter of the candidate obstacles, and create bin indexes; The candidate obstacles in each compartment are sorted according to the second position parameter, and the left candidate compartment container and the right candidate compartment container are traversed according to the preset hierarchical filtering rules. At least one target obstacle is determined from each compartment to form a target obstacle set. The target obstacle set is sorted according to the first position parameter, and a preset number of key obstacles are output as the filtering results.

2. The obstacle screening method based on binning algorithm according to claim 1, characterized in that, The step of dividing the target space into multiple continuous intervals along the path reference direction of the controlled intelligent agent based on a preset binning interval parameter, and creating left candidate binning containers and right candidate binning containers based on the continuous intervals respectively, includes: The path reference direction is taken as the first direction, and the direction perpendicular to the first direction is taken as the second direction. The first direction and the second direction constitute the local coordinate system of the controlled intelligent agent. The origin of the controlled intelligent agent in the local coordinate system is taken as the reference point, and the axis passing through the reference point and extending along the first direction is taken as the reference line. The second position parameter corresponding to the reference line is zero, and the positive direction of the second direction is defined as the right direction and the negative direction is defined as the left direction. Based on the preset bin spacing parameters, the position range of the target space in the path reference direction is divided into multiple continuous intervals; Based on the multiple continuous intervals, a right candidate bin container and a left candidate bin container are created in the right direction and the left direction, respectively. The right candidate bin container is used to manage the bins of candidate obstacles whose second position parameter is greater than zero, and the left candidate bin container is used to manage the bins of candidate obstacles whose second position parameter is less than zero.

3. The obstacle screening method based on binning algorithm according to claim 1, characterized in that, The step of traversing the candidate obstacle set and mapping candidate obstacles that meet preset candidate conditions to corresponding bins in the left candidate bin container or the right candidate bin container and creating bin indexes based on the first and second position parameters of the candidate obstacles includes: Acquire candidate obstacles, and determine whether the candidate obstacles are within a preset position range along the path reference direction in the target space based on the first position parameter; When the candidate obstacle is within the preset position range, it is determined whether the candidate obstacle meets the preset intersection approach condition, wherein the preset intersection approach condition includes: the angle between the movement direction of the candidate obstacle and the path reference direction is greater than a preset angle threshold; When the candidate obstacle meets the preset intersection approach condition, the interval to which the candidate obstacle belongs in the plurality of consecutive intervals is determined based on the first position parameter; The lateral assignment of the candidate obstacle is determined based on the second position parameter, so as to determine whether the candidate obstacle is mapped to the left candidate bin container or the right candidate bin container; Based on the assigned interval and the lateral affiliation, the bins are queried in the corresponding candidate bin containers. If the bin has not yet been created, the bin is created, the candidate obstacle is placed into the bin, and a bin index is assigned to the bin when the candidate obstacle is first placed into the bin.

4. The obstacle screening method based on binning algorithm according to claim 1, characterized in that, The step of sorting the candidate obstacles in each compartment according to the second position parameter, and traversing the left and right candidate compartment containers respectively according to a preset hierarchical filtering rule to determine at least one target obstacle from each compartment to form a target obstacle set includes: The candidate obstacles in each compartment of the left candidate compartment container and the right candidate compartment container are sorted according to the second position parameter to obtain the candidate sequence corresponding to each compartment. The left-side stratified screening layer number m is determined by the maximum length of the candidate sequence corresponding to each box in the left-side candidate box container, and the right-side stratified screening layer number n is determined by the maximum length of the candidate sequence corresponding to each box in the right-side candidate box container. Based on the left-side layered screening layer m, perform layered screening processing on each sub-box in the left-side candidate sub-box container, and based on the right-side layered screening layer n, perform layered screening processing on each sub-box in the right-side candidate sub-box container, so as to identify at least one target obstacle from each sub-box; The target obstacles identified in each compartment of the left candidate sub-container and the right candidate sub-container are combined to form the target obstacle set.

5. The obstacle screening method based on binning algorithm according to claim 4, characterized in that, The candidate obstacles in any compartment of the left candidate compartment container are sorted in descending order according to the second position parameter to obtain the first candidate sequence; The step of performing stratified filtering on each bin in the left candidate bin container based on the left stratified filtering layer number m includes: Obtain the first candidate sequence corresponding to each bin in the left candidate bin container; In the first layer of screening, for each sub-box in the left candidate sub-box container, the candidate obstacle with the sequence number 1 is selected from the corresponding first candidate sequence as the left first layer target obstacle, and the second position parameter of the left first layer target obstacle is determined as the left reference position parameter of the sub-box in each subsequent layer of screening; In the filtering from the 2nd to the mth layer, the layer number k is traversed layer by layer from 2 to m. At the beginning of each layer filtering, the current position parameter on the left is initialized to the preset left boundary parameter. The preset left boundary parameter is the second position parameter corresponding to the maximum lateral boundary of the preset position range of the target space along the path reference direction on the left. In each layer of filtering, the following steps are performed sequentially for each bin in the left candidate bin container: when the length of the first candidate sequence corresponding to the bin is not less than the layer number k, the candidate obstacle with the sequence number k in the first candidate sequence is obtained as the current candidate obstacle. When the absolute value of the difference between the second position parameter of the current candidate obstacle and the left reference position parameter corresponding to the bin is less than or equal to a preset left distance threshold, and the second position parameter of the current candidate obstacle is greater than the left current position parameter, then the current candidate obstacle is determined as the target obstacle, and the left current position parameter is updated to the second position parameter of the target obstacle.

6. The obstacle screening method based on binning algorithm according to claim 4, characterized in that, The candidate obstacles in any compartment of the right-side candidate compartment container are sorted in ascending order according to the second position parameter to obtain the second candidate sequence; The step of performing stratified filtering on each compartment in the candidate compartment container on the right side based on the number of stratified filtering layers n on the right side includes: Obtain the second candidate sequence corresponding to each bin in the right candidate bin container; In the first layer of screening, for each sub-box in the candidate sub-box container on the right, the candidate obstacle with the sequence number 1 is selected from the corresponding second candidate sequence as the target obstacle of the first layer on the right, and the second position parameter of the target obstacle of the first layer on the right is determined as the right reference position parameter of the sub-box in each subsequent layer of screening; In the filtering of layers 2 to n, the layers k from 2 to n are traversed layer by layer. At the beginning of each layer filtering, the current position parameter on the right is initialized to the preset right boundary parameter. The preset right boundary parameter is the second position parameter corresponding to the maximum lateral boundary of the preset position range of the target space along the path reference direction on the right. In each layer of filtering, the following steps are performed sequentially for each bin in the right candidate bin container: when the length of the second candidate sequence corresponding to the bin is not less than the layer number k, the candidate obstacle with the sequence number k in its second candidate sequence is obtained as the current candidate obstacle. When the absolute value of the difference between the second position parameter of the current candidate obstacle and the right reference position parameter corresponding to the bin is less than or equal to a preset right distance threshold, and the second position parameter of the current candidate obstacle is less than the right current position parameter, then the current candidate obstacle is determined as the target obstacle, and the right current position parameter is updated to the second position parameter of the target obstacle.

7. The obstacle screening method based on binning algorithm according to claim 4, characterized in that, The step of summarizing the target obstacles identified in each compartment of the left candidate sub-container and the right candidate sub-container to form the target obstacle set includes: Construct a unified target container, wherein the unified target container includes a mapping structure with bin index as the key; Write the target obstacles identified in each compartment of the left candidate sub-container and the right candidate sub-container into the target list associated with the corresponding sub-container index in the unified target container; The target list in the unified target container is read sequentially in ascending order of the interval index corresponding to the bin index to form the target obstacle set. Under the same interval index, the target list is read and the target obstacles are output in an alternating manner of left and right bin indexes to maintain the spatial distribution characteristics of the target obstacle set.

8. An electronic device, characterized in that, The electronic device includes: One or more processors; and a memory storing computer program instructions, which, when executed, cause the processors to perform the obstacle screening method based on the binning algorithm as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program and / or instructions stored thereon, characterized in that, When the computer program and / or instructions are executed by the processor, they implement the obstacle screening method based on the binning algorithm as described in any one of claims 1-7.

10. A computer program product comprising a computer program and / or instructions, characterized in that, When the computer program and / or instructions are executed by the processor, they implement the obstacle screening method based on the binning algorithm as described in any one of claims 1-7.

Images