Target position classification method and device and computer equipment

Abstract

The invention relates to a target position classification method and device and computer equipment. The method comprises the steps of determining a predicted position type of a target object based on radar image features at a current moment; based on the predicted position types of the target object at the current moment and the historical moment, calculating a dynamic probability corresponding to each position type at the current moment; based on the dynamic probability and the confidence coefficient of each position type of the target object at the historical moment, calculating the confidence coefficient corresponding to each position type of the target object at the current moment; and determining the actual position type of the target object based on the confidence at the current moment. By adopting the method, the stability and confidence accuracy of target position classification can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a target location classification method, apparatus, and computer equipment. Background Technology

[0002] In the field of vehicle perception, millimeter-wave radar is often used to classify the location type of a target by its pitch characteristics in order to assist in path planning and obstacle avoidance.

[0003] Traditional methods for determining a target's elevation characteristics typically rely on fixed thresholds in a single frame of radar data to classify the target as passable, impassable, or subject to low obstacles, using single features such as instantaneous height or signal-to-noise ratio for classification. However, due to environmental disturbances, sensor noise, or the uncertainty of target motion, classification results are prone to fluctuations and misclassifications, leading to unstable output types and discontinuous jumps in confidence levels, thus affecting the reliability of the system's decision-making.

[0004] This shows that the existing target location type classification still suffers from low stability and accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a target location classification method, apparatus, and computer equipment that can improve the stability and confidence accuracy of target location classification in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a target location classification method applied to vehicles, the target location classification method comprising:

[0007] Based on the radar image features at the current moment, determine the predicted location type of the target object;

[0008] Based on the predicted position type of the target object at the current time and historical time, calculate the dynamic probability corresponding to each position type at the current time;

[0009] Based on the dynamic probability and the confidence level of the target object for each location type in historical time, calculate the confidence level of the target object for each location type in the current time.

[0010] Based on the confidence level at the current moment, the actual location type of the target object is determined.

[0011] In one embodiment, the radar image features include a target location range and a signal-to-noise ratio range; determining the predicted location type of the target object based on the radar image features at the current moment includes:

[0012] Obtain the vehicle's own location range;

[0013] Based on the vehicle's position range, the target position range and the signal-to-noise ratio range are classified to predict the position type of the target object.

[0014] In one embodiment, the historical time includes at least one historical time under a preset time sliding window; the step of calculating the dynamic probability corresponding to each location type at the current time based on the predicted location type of the target object at the current time and the historical time includes:

[0015] Based on the predicted location type of the target object at the current time and historical time, calculate the ratio of the occurrence frequency of each location type to the number of times corresponding to the preset time window, and obtain the historical occurrence ratio of each location type;

[0016] Based on the historical occurrence ratio, the dynamic probability corresponding to each location type at the current moment is determined.

[0017] In one embodiment, determining the dynamic probability of each location type at the current moment based on the historical occurrence ratio includes:

[0018] Based on the radar image features and the vehicle's road data, the obstacle type of the target object is determined;

[0019] Based on the motion information of the target object, calculate the motion predictability score of the target object;

[0020] Based on the obstacle type and the motion predictability score, the historical occurrence ratio is weighted to obtain a weighted probability.

[0021] Based on the weighted probabilities, the dynamic probability corresponding to each location type at the current time is determined.

[0022] In one embodiment, determining the dynamic probability corresponding to each location type at the current moment based on the historical occurrence ratio further includes:

[0023] The dynamic probability is smoothed by a preset smoothing factor; the preset smoothing factor is dynamically determined based on the number of consecutive occurrences of the predicted position type at the current time.

[0024] In one embodiment, the location type at a historical moment includes a historical actual location type and a historical unknown location type; calculating the confidence level of the target object at the current moment for each location type based on the dynamic probability and the confidence level of the target object for each location type at a historical moment includes:

[0025] The first calculation factor is determined based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time.

[0026] The second calculation factor is determined based on the dynamic probability of the current location type at the current moment, the confidence level of the historical actual location type at historical moments, and the confidence level of the historical unknown location type at historical moments.

[0027] The confidence level of the current location type at the current moment is determined based on the ratio of the first calculation factor to the second calculation factor.

[0028] In one embodiment, determining the first calculation factor based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time includes:

[0029] The first parameter value is obtained by multiplying the confidence level of the historical actual location type at a historical time with the dynamic probability of the current location type at the current time.

[0030] The second parameter value is obtained based on the confidence level of the historical actual location type at a historical moment and the dynamic probability of the historical unknown location type at the current moment.

[0031] The third parameter value is obtained by multiplying the confidence level of the historical unknown location type at a historical time with the dynamic probability of the current location type at the current time.

[0032] The first calculation factor is obtained based on the sum of the first parameter value, the second parameter value, and the third parameter value.

[0033] In one embodiment, determining the second calculation factor based on the dynamic probability of the current location type at the current time, the confidence level of the historical actual location type at historical times, and the confidence level of the historical unknown location type at historical times includes:

[0034] Based on the confidence level of the actual historical location type at a historical moment and the confidence level of the unknown historical location type at a historical moment, the confidence level of type change is obtained;

[0035] The second calculation factor is determined based on the product of the confidence level of the type change and the dynamic probability of the current location type at the current moment.

[0036] Secondly, this application provides a target location classification device, the target location classification device comprising:

[0037] The type prediction module is used to determine the predicted location type of a target object based on the radar image features at the current moment.

[0038] The probability calculation module is used to calculate the dynamic probability corresponding to each position type at the current time based on the predicted position type of the target object at the current time and historical time.

[0039] The confidence calculation module is used to calculate the confidence level of the target object at the current time for each location type based on the dynamic probability and the confidence level of the target object for each location type at historical times.

[0040] The type determination module is used to determine the actual position type of the target object based on the confidence level at the current moment.

[0041] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0042] The aforementioned target location classification method, apparatus, and computer equipment determine the predicted location type of a target object based on radar image features at the current moment, calculate the dynamic probability corresponding to each location type at the current moment based on the predicted location types of the target object at the current moment and historical moments, calculate the confidence level corresponding to each location type of the target object at the current moment based on the dynamic probability and the confidence level of each location type of the target object at historical moments, determine the actual location type of the target object based on the confidence level at the current moment, perform time-series modeling of dynamic probabilities based on the predicted location types based on radar image features, analyze the time-series evolution trend of dynamic probabilities, and output stable decisions by fusing historical memory and current trends through confidence levels. This achieves time-series smoothing and confidence level optimization of target location types, thereby improving the stability and confidence accuracy of target location classification. Attached Figure Description

[0043] Figure 1 This is a diagram illustrating the application environment of a target location classification method in one embodiment.

[0044] Figure 2 This is a flowchart illustrating a target location classification method in one embodiment;

[0045] Figure 3 This is a flowchart illustrating the target location classification method in another embodiment;

[0046] Figure 4 This is a schematic diagram illustrating the pitch type conversion in one embodiment;

[0047] Figure 5 This is a schematic diagram of the pitch type simulation results in one embodiment;

[0048] Figure 6 This is a structural block diagram of a target location classification device in one embodiment;

[0049] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] The target location classification method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, the processing terminal 102 acquires data transmitted by the radar device 104 via a wired network. The processing terminal 102 processes the radar data to obtain radar image features, and based on the radar image features at the current moment, determines the predicted location type of the target object; based on the predicted location types of the target object at the current moment and historical moments, it calculates the dynamic probability corresponding to each location type at the current moment; based on the dynamic probability and the confidence level of the target object for each location type at historical moments, it calculates the confidence level of the target object for each location type at the current moment; and based on the confidence level at the current moment, it determines the actual location type of the target object. The processing terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and IoT devices; the IoT device can be a smart vehicle device.

[0052] In one embodiment, such as Figure 2 As shown, a target location classification method is provided, which can be applied to... Figure 1 Taking vehicles as an example, target location classification methods include:

[0053] Step S100: Determine the predicted location type of the target object based on the radar image features at the current moment.

[0054] The radar image features can be a set of multi-dimensional parameters characterizing the physical properties of the target extracted from the echo data acquired by the millimeter-wave radar at the current moment. These parameters include, but are not limited to, range, Doppler shift, reflection intensity, signal-to-noise ratio, and elevation angle distribution. They can be used to provide an observable physical representation of the target in space at the current moment. In this embodiment, the radar image features can be extracted by imaging the echo using a radar signal processing unit. Examples of such features include one or more of the following: instantaneous altitude, reflection intensity, scattering area, signal-to-noise ratio, Doppler shift, and elevation angle distribution.

[0055] The target object can be an external entity detected by millimeter-wave radar that has identifiable echo characteristics, whose echo characteristics maintain temporal consistency and can be clustered into independent trajectory units. In this embodiment, the target object can be an independent entity unit with continuous trajectory and consistent characteristics formed by target segmentation of the radar point cloud through a clustering algorithm. For example, the target object can be classified into stationary obstacles, low-speed moving targets, high-speed moving targets, etc., according to different movement states.

[0056] Location type can be a set of discrete classification labels for the travel attributes of a target in a road environment. For example, it can include passable type, impassable type, low-profile static type, etc. Further, the predicted location type can be one of all preset location types, obtained based on the radar image features at the current moment, through preset rules, classifiers, etc., and is used to provide a preliminary classification result for the current frame. In this embodiment, the predicted location type can use a simple classification model or a pre-set threshold division method to map radar image features to the corresponding location type. In an exemplary embodiment, the location type can include passable type, impassable type, low-profile static type, etc.

[0057] Correspondingly, determining the predicted location type of a target object based on the radar image features at the current moment can be achieved by classifying the current radar image features according to preset classification rules, thereby obtaining discrete type labels. For example, when the radar image features include height features and signal-to-noise ratio (SNR) features, determining the predicted location type of a target object based on the radar image features at the current moment can be achieved by determining whether the target is a low obstacle through a combination of a preset height threshold and SNR conditions, and by determining whether the target is an impassable obstacle through a logical combination of the reflection intensity distribution and the elevation angle range.

[0058] Step S200: Based on the predicted position type of the target object at the current time and historical time, calculate the dynamic probability corresponding to each position type at the current time.

[0059] The current moment can be the point in time when the radar system performs a complete perception and classification process. In this embodiment, the current moment can be a timestamp provided by the radar system's clock synchronization module, serving as the completion time of a single radar frame's acquisition and processing. Correspondingly, the historical moments can be a collection of several time points prior to the current moment, with each historical moment corresponding to a predicted location type of the target. For example, the historical moments can be cached to store the predicted location type and confidence sequence corresponding to each past historical moment.

[0060] Dynamic probability can be the tendency probability of each position type at the current moment, reflecting the behavioral inertia of the target type in the time dimension, thereby weakening the impact of single-frame noise interference. For example, dynamic probability can predict the occurrence number, frequency, previous type conversion frequency distribution, and subsequent type conversion frequency distribution of the target in the sequence of the current moment and historical moments, and infer the tendency probability of each position type at the current moment.

[0061] In an exemplary embodiment, based on the predicted location type of the target object at the current time and historical time, the dynamic probability corresponding to each location type at the current time is calculated. The probability transfer weight from the previous time type to the current time type can be calculated by using a first-order Markov chain model. A time decay factor can be introduced to give lower transfer contribution weight to the predictions of more distant historical time. A type preservation preference model can also be constructed to give higher transfer probability to types that appear more than a threshold number of times consecutively. This makes it possible to construct a time-series trend model of target type evolution and reduce the dominant role of single-frame misjudgment in the classification results.

[0062] Step S300: Based on the dynamic probability and the confidence of the target object for each position type in the historical time, calculate the confidence of the target object for each position type in the current time.

[0063] Here, confidence level can be a quantitative assessment of the reliability of a target at a certain location type, used to characterize the credibility of the current classification result. In an exemplary embodiment, confidence level can be a continuous value between 0 and 1 generated based on a weighted combination of the consistency of historical classification results, feature stability, and dynamic probability. Furthermore, confidence level can include, but is not limited to, one or more of the following: feature consistency confidence level, temporal consistency confidence level, and transition stability confidence level.

[0064] In this embodiment, the confidence level can be calculated by using the dynamic probability as a weighting factor and weighting it with the historical confidence level to generate the current confidence level. In an exemplary embodiment, the confidence level can be calculated by using an exponentially weighted average to perform time-decay fusion of the historical confidence level sequence and multiplying it by a dynamic probability correction factor. A verification mechanism can also be introduced to verify the consistency of the confidence level, thereby allowing only historical confidence levels consistent with the dynamic probability trend to participate in the weighting. Furthermore, a confidence level accumulation gating mechanism can be constructed to suppress the confidence level update rate when the dynamic probability is lower than a threshold, thereby enabling adaptive fusion of historical decision memory and current observation trend, and achieving smooth jumps in confidence level.

[0065] Step S400: Determine the actual position type of the target object based on the confidence level at the current moment.

[0066] The actual location type can be the classification decision result output based on the confidence scores of all location types at the current time. For example, the actual location type can be the location type with the highest confidence score among all types at the current time.

[0067] Furthermore, by setting a confidence difference threshold, the result is only output when the difference between the maximum and the second largest value exceeds the set value; otherwise, the result of the previous frame is retained, thereby stabilizing the classification result.

[0068] Furthermore, confidence smoothing filtering can be introduced to perform local weighted averaging of the confidence scores of candidate types before selecting the best output. Target motion speed information can also be combined to adjust the confidence standard for high-speed targets and raise the judgment threshold for stationary targets, thereby outputting a stable classification result that has undergone time-series optimization and reducing decision jitter.

[0069] Taking obstacle classification in a low-speed driving environment on urban roads as an example, the target location classification method in this embodiment can be used to address the issue of a temporarily obstructed curb in front. In a single frame of radar data, the curb's reflection intensity decreases due to partial obstruction, resulting in a lower instantaneous height measurement. In traditional methods, this can easily be misclassified as a low obstacle. However, this solution uses multiple consecutive frames to predict the location type, determining the actual location type as an impassable obstacle, maintaining a consistently high dynamic probability. Although the historical confidence level decreases slightly due to instantaneous interference, it remains high after weighted fusion. Therefore, it can effectively avoid accidental obstacle avoidance actions triggered by single-frame noise, ensuring smooth vehicle passage.

[0070] This embodiment provides a target location classification method that determines the predicted location type of a target object based on radar image features at the current moment. It calculates the dynamic probability corresponding to each location type at the current moment based on the predicted location types of the target object at the current moment and historical moments. Based on the dynamic probability and the confidence level of each location type of the target object at historical moments, it calculates the confidence level corresponding to each location type of the target object at the current moment. Based on the confidence level at the current moment, it determines the actual location type of the target object. By performing temporal modeling of the dynamic probability based on the predicted location type using radar image features, it analyzes the temporal evolution trend of the dynamic probability. By fusing historical memory and current trends through confidence level analysis, it outputs a stable decision, thereby achieving temporal smoothing and confidence level optimization of the target location type, thus improving the stability and accuracy of target location classification.

[0071] In one embodiment, determining the predicted location type of the target object based on the radar image features at the current moment includes:

[0072] Obtain the vehicle's location range;

[0073] Based on the vehicle's position range, the predicted position classification is performed on the target position range and the signal-to-noise ratio range to obtain the predicted position type of the target object.

[0074] The radar image features in this embodiment include the target location range and the signal-to-noise ratio range.

[0075] The vehicle's position range can be the spatial range occupied by the vehicle itself within the current coordinate system, including at least the minimum space required for safe passage. For example, the vehicle's position range can include a closed interval from the ground clearance to the roof height, serving as a reference for determining whether a target constitutes an obstacle. In this embodiment, the vehicle's position range can be pre-stored vehicle design parameters, or it can be obtained in real-time through onboard attitude sensors. Furthermore, it can be dynamically corrected by combining vehicle load and suspension status.

[0076] The signal-to-noise ratio (SNR) range can be the SNR distribution interval of the target echo signal at the current moment, reflecting the fluctuation range of the target's reflection intensity. It can provide consistency information on the target's reflection characteristics, helping to determine whether the target is a real object or noise interference. In a specific embodiment, the SNR range can be calculated by statistically analyzing the SNR of all echo points in the point cloud to which the target belongs, and extracting the minimum and maximum values ​​to form an interval.

[0077] Furthermore, obtaining the vehicle's own position range can be achieved by reading preset ground clearance and vehicle height parameters from the vehicle configuration database and combining them with vehicle model information to determine the static range; alternatively, it can be achieved by collecting the ground clearance distance of the front and rear axles in real time through suspension height sensors to dynamically update the vehicle's own position range; or it can be achieved by combining vehicle load signals and attitude angle data to perform pitch and tilt correction on the preset range to provide benchmark information for the vehicle's own passage space.

[0078] The target location range can be the spatial range occupied by the target object. In one embodiment, the target location range can be the height range occupied by the target object in the vertical direction. This height range can be converted from the distribution range of radar echoes in the pitch angle dimension through triangulation and distance information to determine whether it intrudes into the space required for vehicle passage.

[0079] Taking vertical position classification as an example, based on the vehicle's position range, the target position range and signal-to-noise ratio (SNR) range are predicted for position classification to obtain the predicted position type of the target object. This can be achieved by determining whether the target position range overlaps with the vehicle's position range in the vertical direction, and combining the intensity characteristics of the SNR range to output the predicted position type according to preset rules. Furthermore, it can be classified as an impassable obstacle when the lower limit of the target position range is lower than the vehicle's minimum ground clearance and the upper limit is higher than that height; it can also be classified as a low static obstacle when the target position range is completely below the vehicle's passageway and the SNR range is in the low-reflection range; and it can be marked as an uncertain target in the passable area when the target position range does not overlap with the vehicle's space but the SNR range shows a non-uniform distribution. This allows for classification logic based on spatial conflict detection, using whether the target intrudes into the vehicle's passageway as the criterion for judgment.

[0080] This embodiment provides a target prediction location classification method, which obtains the vehicle's own position range; based on the vehicle's own position range, predicts the target position range and signal-to-noise ratio range to obtain the predicted position type of the target object. It can effectively determine the predicted position classification corresponding to the target position range according to the vehicle's own situation, so that adaptive judgment can be made under different vehicle models, thereby improving the flexibility of predicted position classification.

[0081] In one embodiment, the historical time includes at least one historical time under a preset time sliding window; based on the predicted position type of the target object at the current time and the historical time, the dynamic probability corresponding to each position type at the current time is calculated as follows:

[0082] Based on the predicted location type of the target object at the current time and historical time, calculate the ratio of the occurrence frequency of each location type to the number of times corresponding to the preset time window, and obtain the historical occurrence ratio of each location type.

[0083] Based on historical occurrence rates, determine the dynamic probability of each location type at the current moment.

[0084] The preset time window can be a time window within a historical time range, used to limit the historical observation range on which dynamic probability calculation depends, avoiding response lag due to excessive memory or statistical instability due to excessively short memory.

[0085] The occurrence count can be the frequency of a single location type appearing within a preset time window, reflecting the recent recurrence frequency of that location type. In this embodiment, the occurrence count can be obtained by accumulating the occurrence records of the predicted location type at each moment within the time window. The number of moments can be the total number of valid time points covered by the preset time window, including the current moment and historical moments within the window, and can be calculated based on the window length and radar frame rate.

[0086] Correspondingly, the historical occurrence ratio can be the ratio of the number of times a certain location type appears within a preset time window to the total number of moments, representing the relative frequency of that type in recent observations. This can be used to quantify the persistence of a target being classified as a certain type within a recent time window.

[0087] The historical occurrence ratio of each location type is calculated by dividing the frequency of each location type within the time window by the total number of times. For example, this can be achieved by using a sliding counter to accumulate the prediction results for each type, updating the count and removing expired frames each time a new frame arrives. Alternatively, a weighted counting mechanism can be introduced, assigning higher weights to prediction results from more recent times within the window before calculating the ratio. For instance, the historical time furthest from the current time within the preset time window has the lowest reference value and can be assigned a relatively low weight for counting, while the previous time, being the closest to the current time, can be assigned a relatively higher weight. Furthermore, a minimum effective frame threshold can be set, pausing the ratio output when the actual number of available times falls below the threshold to ensure statistical reliability. This transforms discrete time-series classification results into continuous frequency indicators.

[0088] Furthermore, during the statistical process of occurrence count, the first occurrence time can be recorded for weighted calculation of dynamic probability. For example, if the difference between the first occurrence time of a single location type and the current time is the smallest compared to other location types, then the location type has higher reference value at the current time and can be given a relatively higher weight. Conversely, if the first occurrence time of a single location type is older, then the location type has lower reference value at the current time and can be given a relatively lower weight.

[0089] Accordingly, determining the dynamic probability corresponding to each location type at the current moment can be achieved by using historical occurrence ratios as the dynamic probability, or by using historical occurrence ratios as the basic input and generating the dynamic probability through one or more preset mapping functions, preset correction models, etc. For example, a nonlinear mapping function can be introduced to enhance the ratio value, such as assigning a probability value close to 1 to ratios higher than 0.7; furthermore, the target's motion state can be considered, using the original ratio for stationary targets and adding a ratio attenuation factor for high-speed targets to adapt to rapidly changing scenarios.

[0090] This embodiment provides a method for determining the dynamic probability of target location type. By calculating the ratio of the number of occurrences of each location type to the number of times corresponding to a preset time window based on the predicted location type of the target object at the current time and historical times, the historical occurrence ratio of each location type is obtained. Based on the historical occurrence ratio, the dynamic probability corresponding to each location type at the current time is determined. By using the ratio of the number of occurrences to the number of times to quantify the recent frequency of the location type, the historical occurrence ratio is formed and converted into a dynamic probability. This can realize a classification stability enhancement mechanism based on time series statistics, thereby suppressing classification fluctuations caused by sensor noise or temporary occlusion, and achieving the technical effect of improving the continuity and confidence accuracy of target location type judgment.

[0091] In one embodiment, determining the dynamic probability of each location type at the current moment based on historical occurrence ratios includes:

[0092] Based on radar image features and vehicle road data, the obstacle type of the target object is determined;

[0093] Based on the motion information of the target object, calculate the motion predictability score of the target object;

[0094] Based on obstacle type and motion predictability score, the historical occurrence ratio is weighted to obtain a weighted probability;

[0095] Based on weighted probabilities, determine the dynamic probability corresponding to each location type at the current time.

[0096] The vehicle's road data can be the topological and semantic information of the road the vehicle is currently on, including but not limited to one or more static or semi-static attributes such as road type, number of lanes, speed limit, traffic signs, road curvature, and surrounding facilities. In this embodiment, the road data can provide contextual information about the traffic environment in which the target object is located, assisting in determining its possible behavior patterns and obstacle attributes. For example, the vehicle's road data can be obtained by reading the road attribute data corresponding to the vehicle's current location through a high-precision map module, or it can be obtained by the onboard perception system combining positioning information to identify and update road features in real time.

[0097] The obstacle type of a target object can be its relationship to a road boundary. For example, based on road data and radar image features, it can be determined whether the target object corresponding to the radar image features coincides with the road boundary, thereby determining whether the target object itself is part of the boundary and thus determining the obstacle type of the target object. For example, the obstacle type can include boundary obstacle type and road obstacle type.

[0098] The motion information of the target object can be the temporal motion characteristics exhibited by the target object in consecutive radar frames, including but not limited to one or more of the following: velocity, acceleration, trajectory direction, motion continuity, and trajectory curvature. In an exemplary embodiment, the motion information of the target object can be obtained by calculating the sequence of motion state parameters of the target in space through a multi-frame radar point cloud tracking and trajectory fitting algorithm.

[0099] Motion predictability scoring is a numerical indicator that quantifies the degree to which the future state change trend of a target can be accurately estimated. It reflects the stability of the target's behavior and can thus adjust the weight of historical statistical results in dynamic probability. In this embodiment, motion predictability scoring can be generated by weighting functions, machine learning models, etc., based on one or more of the target's motion information, such as velocity stability, rate of change of acceleration, trajectory continuity, and directional consistency. For example, lower motion predictability means that even if the target is currently passable, it may still have a higher collision risk in the future due to its variable motion; conversely, higher motion predictability indicates a higher reliability of the current passability assessment.

[0100] Correspondingly, calculating the predictability score of a target object's motion can involve analyzing the stability and regularity of the target's trajectory to generate a quantitative predictability value. For example, this can be achieved by linearly fitting the trajectory of five consecutive frames, with a higher score indicating a smaller fitting residual; it can also be achieved by calculating the standard deviation of acceleration changes, assigning a high stability score when the standard deviation is below a threshold; it can also be achieved by detecting the frequency of sudden changes in motion direction and assigning a low consistency score to targets that frequently change direction; other methods can also be used to analyze the degree of regularity through motion features, which are not limited in this embodiment.

[0101] Weighted probability can be a probability value that adjusts the historical occurrence ratio based on obstacle type and motion predictability score. It can reflect the difference in classification weights for different targets in different scenarios, making highly predictable targets more reliant on historical trends, while low-predictability targets focus more on current observations. In a specific embodiment, the weighted probability can be dynamically scaled by setting a base weight coefficient based on obstacle type and combining it with motion predictability score to achieve multiplicative correction of the historical occurrence ratio, thereby intelligently adjusting historical statistical results and balancing classification stability and response sensitivity.

[0102] Based on the weighted probability, the dynamic probability corresponding to each location type at the current time is determined. Accordingly, the weighted probability can be used as the dynamic probability of the location type, or the weighted probability after standardization can be used as the dynamic probability of the location type.

[0103] This embodiment provides a target location classification method based on dynamic probability. It distinguishes the obstacle type of the target object based on radar image features and vehicle road data; and calculates the motion predictability score of the target object based on the motion information of the target object to quantify the stability of the target behavior to support weight adjustment. By weighting the historical occurrence ratio based on obstacle type and motion predictability score, it can realize differentiated adjustment of historical statistical results, enhance the trend continuity of highly predictable targets, improve the response sensitivity of low predictable targets, improve the accuracy and reliability of target location type identification, and achieve the technical effect of enhancing path planning safety and decision robustness.

[0104] In one embodiment, determining the dynamic probability of each location type at the current moment based on historical occurrence ratios further includes:

[0105] Based on a preset smoothing factor, the dynamic probability is smoothed and filtered.

[0106] The preset smoothing factor can be a weighted parameter used to control the rate of change of dynamic probability over time, and can adjust the update rate of dynamic probability. For example, the preset smoothing factor can be dynamically adjusted according to the predicted position type under a preset time sliding window. For instance, it can be adjusted when the state is stable to enhance smoothness, and adjusted when the state changes to improve response speed.

[0107] In one exemplary embodiment, a preset smoothing factor is applied to a preset smoothing filtering algorithm, serving as an input parameter to adjust the smoothing filtering process. In this embodiment, the preset smoothing factor, depending on the preset smoothing filtering algorithm, may include, but is not limited to, one or more of the following: sliding window size, standard deviation, regularization parameter, etc.

[0108] In one exemplary embodiment, the preset smoothing factor is dynamically determined based on the number of consecutive occurrences of the predicted location type at the current moment. For example, the higher the number of consecutive occurrences, the larger the smoothing factor, and vice versa, forming an adaptive adjustment mechanism. Correspondingly, smoothing filtering of the dynamic probability based on the preset smoothing factor can be achieved by weighting the unfiltered dynamic probability at the current moment with the historical filtered dynamic probability according to the preset smoothing factor, thereby outputting the smoothed dynamic probability. Further, dynamically determining the preset smoothing factor based on the number of consecutive occurrences of the predicted location type at the current moment can be achieved by counting the number of consecutive frames of the current predicted location type in the historical sequence, and then looking up or calculating the corresponding smoothing factor value based on this count. For example, a mapping lookup table between the number of consecutive occurrences and the smoothing factor can be established. For instance, 0.3 corresponds to one consecutive moment, and 0.8 corresponds to five or more consecutive moments, thus determining the corresponding preset smoothing factor for different consecutive occurrences. It is understood that the correspondence between the number of consecutive occurrences and the preset smoothing factor is not limited to a mapping lookup table; it can also be obtained by establishing a piecewise linear model, a nonlinear model, or other correspondence models.

[0109] For example, in obstacle recognition scenarios involving sudden changes in lighting at tunnel entrances and exits, the target location classification method in this embodiment might involve a brief anomaly in radar echoes caused by lighting changes the instant a vehicle exits the tunnel. The target might have been classified as an impassable obstacle for five consecutive historical frames, but misclassified as a low obstacle in the sixth frame. Since the number of consecutive occurrences decreases from five to one, the preset smoothing factor should be dynamically adjusted from 0.8 to 0.3 according to the mapping lookup table. This reduces the historical weight, allowing the current observation to quickly influence the dynamic probability. Subsequently, the correct classification is restored in the seventh frame, the number of consecutive occurrences is re-accumulated, and the preset smoothing factor gradually increases. This mechanism effectively avoids the problem of excessive lag or over-response of a fixed smoothing factor in abrupt change scenarios, achieving a balance between rapid correction of misclassifications and maintaining a stable trend.

[0110] This embodiment provides a target location classification method that smooths the dynamic probability based on a preset smoothing factor and dynamically determines the preset smoothing factor based on the number of consecutive occurrences of the predicted location type at the current time. By associating the smoothing intensity with the stability of the target state, the method enhances the influence of historical trends when the state is stable and improves the response sensitivity when the state changes. By combining weighted averaging and adaptive adjustment mechanisms to optimize the continuity and anti-interference ability of the probability output, it can effectively improve the temporal consistency of classification results, suppress confidence jumps caused by sensor noise, and achieve the technical effect of improving the robustness of location classification in complex environments.

[0111] In one embodiment, the location type at a historical moment includes the historical actual location type and the historical unknown location type; based on the dynamic probability and the confidence level of the target object for each location type at a historical moment, the confidence level of the target object for each location type at the current moment is calculated as follows:

[0112] The first calculation factor is determined based on the confidence level of the actual historical location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the unknown historical location type at the current time, and the confidence level of the unknown historical location type at a historical time.

[0113] The second calculation factor is determined based on the dynamic probability of the current location type at the current moment, the confidence level of the actual historical location type at historical moments, and the confidence level of the unknown historical location type at historical moments.

[0114] The confidence level of the current location type at the current moment is determined based on the ratio of the first calculation factor to the second calculation factor.

[0115] The historical actual location type can be the actual location type that has been determined and output by the confidence decision mechanism at a historical moment. As the target type information confirmed in history, it participates in the current confidence calculation and can provide a highly reliable historical reference. In this embodiment, the historical actual location type can be read from the actual location type cache sequence at a historical moment, and its generation method can be consistent with the actual location type at the current moment.

[0116] In this embodiment, the location type may also include an unknown location type, which represents an intermediate state that cannot be clearly classified at this time. This avoids the accumulation of erroneous memories due to forced classification and preserves uncertain historical information. Accordingly, a historical unknown location type can be an intermediate state in a historical moment that has not been clearly classified into the location type of the target object. This indicates that at this historical moment, the classification system cannot form a stable judgment due to insufficient confidence, competition among multiple types, or other reasons. In a specific embodiment, when the confidence of each type at a historical moment is lower than a set threshold, the maximum difference is insufficient, or the confidence of the unknown type is the highest, the location type can be marked as an unknown type, and its confidence level can be recorded.

[0117] The current location type can be the location type whose final confidence level is being calculated at the current moment. In this embodiment, the current location type can be any one of the preset location type set.

[0118] In this embodiment, the first calculation factor can be a confidence index for each location type, based on the current location type time series. In an exemplary embodiment, a preset functional relationship can be used to combine the confidence of historical actual location types, the dynamic probability of the current location type, the dynamic probability of historical unknown location types at the current moment, and the confidence of historical unknown location types at historical moments to obtain the first calculation factor. The confidence of historical actual location types and historical unknown location types can provide historical evidence support, while the dynamic probability of historical unknown location types at the current moment and the dynamic probability of the current location type can provide current evidence support. By fusing evidence from the current moment and historical moments using the above parameters, a probability assessment can be achieved. For example, the first calculation factor can be determined by multiplicative combination or by nonlinear activation of a neural network to output the first calculation factor, thereby constructing a joint support model that includes deterministic memory and the influence of uncertainty, enhancing adaptability to complex scenarios.

[0119] The second calculation factor can be obtained by comprehensively calculating the dynamic probability of the current location type at the current time, the confidence level of the historical actual location type at historical times, and the confidence level of the historical unknown location type at historical times. For example, the confidence levels of the historical actual location type and the historical unknown location type at historical times can reveal the total confidence level of other known categories in the previous frame, which serves as the residual confidence level. When the dynamic probability judgment result conflicts with the dynamic probability of the current location type, it can be used as the denominator to reduce the updated confidence level.

[0120] Accordingly, based on the ratio of the first calculation factor to the second calculation factor, the confidence level of the current position type at the current moment can be determined by dividing the first calculation factor by the second calculation factor to obtain a continuous value between 0 and 1 as the current confidence level.

[0121] This embodiment provides a target location type confidence update method. It determines a first calculation factor based on the confidence of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence of the historical unknown location type at a historical time. It then determines a second calculation factor based on the dynamic probability of the current location type at the current time, the confidence of the historical actual location type at a historical time, and the confidence of the historical unknown location type at a historical time. Finally, it determines the confidence of the current location type at the current time based on the ratio of the first calculation factor to the second calculation factor. By constructing evidence fusion calculation support between historical and current times, the method can reflect not only the current observation intensity but also historical consistency and the rationality of state transitions during the confidence update process. This suppresses the propagation of misjudgments caused by single-frame noise or brief occlusion, thereby effectively improving the robustness of the system under target motion blur or environmental disturbances, and achieving the effect of improving the stability and accuracy of judgment.

[0122] In one embodiment, the first calculation factor is determined based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time.

[0123] The first parameter value is obtained by multiplying the confidence level of the actual location type at a historical time with the dynamic probability of the current location type at the current time.

[0124] The second parameter value is obtained based on the confidence level of the actual historical location type at a historical moment and the dynamic probability of the unknown historical location type at the current moment.

[0125] The third parameter value is obtained by multiplying the confidence level of the unknown location type at a historical time with the dynamic probability of the current location type at the current time.

[0126] The first calculation factor is obtained based on the sum of the first parameter value, the second parameter value, and the third parameter value.

[0127] The first parameter value can be the product of the confidence level of the historical actual location type and the dynamic probability of the current location type at the current moment, used to quantify the sustained support strength of the historical deterministic state for the current target type. In this embodiment, the first parameter value can be obtained by multiplying the confidence level of the confirmed type in the historical moment with the dynamic probability corresponding to that type at the current moment.

[0128] The second parameter value can be the product of the confidence level of the historical actual location type and the dynamic probability of the historical unknown location type at the current moment. It can be used to reflect the potential impact or interference level of the historically confirmed state on the current uncertain state. In a specific embodiment, the second parameter value can be obtained by multiplying the confidence level of the historical actual location type by the dynamic probability of the historical unknown type at the current moment.

[0129] The third parameter value can be the product of the confidence level of the historical unknown location type and the dynamic probability of the current location type at the current moment, used to measure the potential support capability of the historical fuzzy state for the current classification target. In this embodiment, the third parameter value can be obtained by multiplying the confidence level of the historical unknown location type with the dynamic probability of the current location type at the current moment.

[0130] By adding the values ​​of the first, second, and third parameters, a comprehensive support index is generated. This allows for differentiated fusion of different evidence paths, improving numerical robustness in low-evidence scenarios and thus enhancing the joint modeling of historical certainty, uncertainty, and cross-influence.

[0131] This embodiment provides a target location classification method. A first parameter value is obtained by multiplying the confidence level of the historical actual location type at a historical time by the dynamic probability of the current location type at the current time. A second parameter value is obtained by multiplying the confidence level of the historical actual location type at a historical time by the dynamic probability of the unknown historical location type at the current time. A third parameter value is obtained by multiplying the confidence level of the unknown historical location type at a historical time by the dynamic probability of the current location type at the current time. A first calculation factor is obtained by summing the first, second, and third parameter values. The first parameter value reflects the system's accumulated trust in the continuation of a stable state; the second parameter value suppresses the propagation of misclassification caused by transient noise; and the third parameter value characterizes the potential support of historical uncertainty states for the current target type. This method can achieve complex and refined modeling between historical states and current observations, thereby maintaining a smooth evolution of confidence in scenarios such as discontinuous target movement and environmental noise interference. This achieves the technical effect of improving the stability and accuracy of target location classification.

[0132] In one embodiment, the second calculation factor is determined based on the dynamic probability of the current location type at the current moment, the confidence level of the actual historical location type at historical moments, and the confidence level of the unknown historical location type at historical moments, including:

[0133] Based on the confidence scores of the actual historical location type at a given historical moment and the confidence scores of the unknown historical location type at a given historical moment, the confidence score of the type change is obtained.

[0134] The second calculation factor is determined by multiplying the confidence level of type change with the dynamic probability of the current location type at the current moment.

[0135] The type change confidence level can be an evaluation index reflecting whether the system is maintaining a stable state or potentially switching to a new state in the past, used to quantify the stability level of historical decisions. In this embodiment, the type change confidence level can be calculated based on the relative relationship between the confidence level of the actual historical location type and the confidence level of the unknown historical location type. For example, the relative strength of certainty and uncertainty can be measured by using the difference, ratio, or weighted combination method. It is then recursively updated by combining the confidence levels of the actual and unknown types in the previous radar frame with the prior probability of the current radar frame, forming a dynamic evaluation of state changes.

[0136] The second calculation factor is determined by multiplying the type change confidence score by the dynamic probability of the current location type at the current moment. This can be achieved by multiplying the type change confidence score by the dynamic probability of the current location type to generate a baseline term for confidence score normalization. Furthermore, before multiplication, a threshold limit can be applied to the type change confidence score to prevent extreme values ​​from causing computational instability. The product weights can also be dynamically adjusted based on the target's motion state, strengthening stability constraints for stationary targets and relaxing response conditions for moving targets, thereby constructing a coupling relationship between historical stability and current observation trends.

[0137] This embodiment provides a target location classification method that obtains a type change confidence level based on the confidence levels of the actual historical location type and the unknown historical location type at a historical time. A second calculation factor is determined by multiplying the type change confidence level with the dynamic probability of the current location type at the current time. By introducing the type change confidence level, the stability of the historical classification state is explicitly modeled, and this intermediate variable is used to weight and modulate the dynamic probability of the current observation. This suppresses erroneous changes caused by current noise when the historical classification is highly certain, thereby mitigating the confidence level jump problem caused by sensor noise or transient interference. This improves the temporal continuity and decision reliability of target location classification, achieving the technical effect of improving classification stability and accuracy.

[0138] To more clearly illustrate the technical solution of this application, a detailed embodiment is also provided.

[0139] Taking the position type as pitch type as an example, in one embodiment, such as Figure 3 As shown, a target location classification method is provided and applied to a target location classification system. The target location classification system includes a data preprocessing module, a prior probability calculation module, a confidence assessment module, and a result fusion output module. The target location classification method includes:

[0140] Step 1: Extract the height features of the target through the data preprocessing module.

[0141] For target classification, first extract the target feature vector X, = [z max , z min SNR max SNR min ], where z max and z min α-β filtering will be performed, SNR max and SNR min These represent the maximum and minimum SNR among the track-related points. During the data preprocessing stage, the target's pitch type (Type) is initially determined based on the feature vector X. i Where z represents the target's pitch angle, and correspondingly, z max and z min These are the maximum and minimum values ​​of the target's elevation angle, used to characterize the target's elevation angle range. SNR refers to the signal-to-noise ratio of a radar detection point on a target. For example, given multiple data points for a target, each point has a signal-to-noise ratio result, and the SNR is... max and SNR min The highest and lowest signal-to-noise ratio values ​​can be selected from these values ​​and used as a preliminary basis for determining the pitch type.

[0142] The initial determination of the target's pitch type can be achieved using a single-frame pitch classifier. In this embodiment, the single-frame pitch classifier can employ a machine learning algorithm or a threshold-based classifier. For example, it can use pre-set vehicle parameters as the criterion for classifying whether the target can pass; for instance, if the vehicle's maximum height is lower than the pitch angle at the target's lowest point, it can pass underneath.

[0143] Step two involves using the prior probability calculation module to dynamically determine the prior probability (i.e., dynamic probability) based on environmental monitoring and historical observation data. This includes:

[0144] (1) Prior probability initialization: The algorithm's initialization module assigns equal prior probabilities to the four pitch types, i.e., P init (C i )=0.25, where i is the pitch type number and P is the probability value.

[0145] (2) Construct a historical statistics queue of length N to record the occurrence N times of each pitch type in sliding window statistics. Ci ;

[0146] (3) Based on the number of occurrences N Ci Given the proportion of sliding windows, calculate the latest prior probability estimate P for each type. new (C i )=N(Ci ) / N;

[0147] (4) The prior probability is weighted according to the scene attributes and the motion attributes of the trajectory. The scene attributes are obtained from the processing results of other modules in the algorithm, and the motion attributes are the displacement changes of the detected trajectory. Specifically, the scene attributes can be the road fence attributes of the detected target, which are used to determine whether the target itself belongs to the driving boundary of the vehicle, so as to adjust the pitch type of the inaccessible. The motion attributes can be the information of the object's displacement velocity, which are used to determine the motion uncertainty of the target. For example, when the displacement velocity exceeds the standard, it means that the unpredictability of the target is increased, so it is necessary to increase the probability of inaccessibility and decrease the probability of being able to pass from above and from below.

[0148] (5) To prevent abrupt changes in the prior probability and maintain system stability, the prior probability is smoothed by filtering to obtain the final prior probability P(C). i ):

[0149] P(C i )=α×P old (C i )+(1-α)×P new (C i )

[0150] Here, α is the smoothing factor, where 0 < α < 1. The smoothing factor guides the algorithm to place more weight on past or current probability data. It can be set manually or dynamically determined based on the probability distribution within the time window. For example, if an object was considered not an obstacle in the past, but is considered an obstacle only in the current frame, there may be a false detection, and more past data should be considered. Conversely, if the object is considered an obstacle in the current two frames, the probability of a false detection decreases, and to improve response speed, more current data should be considered for probability calculations.

[0151] Step 3: Recursively calculate the confidence score based on the preset Bayesian rules, where:

[0152] In this embodiment, the pitch type includes four types: unknown type (U), passable from above (C1), impassable (C2), and passable from below (C3). The passable from above (C1), impassable (C2), and passable from below (C3) types are independent and cannot be converted into each other. All three types can be converted to the unknown type (U), and the conversion relationship is as follows: Figure 4 As shown.

[0153] For any pitch type, the corresponding confidence assessment formula is:

[0154]

[0155] in: Indicates each type C of the current radar frame i The corresponding prior probability;

[0156] This indicates the final C obtained in the previous radar frame. i The corresponding confidence level;

[0157] This indicates the confidence level for the unknown type in the previous radar frame.

[0158] This represents the prior probability of the unknown type in the current radar frame;

[0159] Represents each type C calculated in the current frame. i The corresponding confidence level.

[0160] Step 4, Confidence Assessment Module: Calculates the confidence level of the output results in real time.

[0161] Calculate the confidence level for each pitch type in real time, and select the maximum confidence level P(C). i ) max The corresponding type C i .

[0162] Substituting the prior probability of each category and the confidence level of the previous frame into the above formula, we calculate the confidence level of each category for the current radar frame, and then select the maximum confidence level P(C). i ) max Its corresponding type C i This refers to the target elevation type output by the radar in the current frame.

[0163] Step 5, Result Fusion Output Module: Generates the final classification result including confidence level assessment, achieving stable and reliable pitch state output. For example, the final classification result may include the pitch type of the output target and its corresponding confidence level, for use by subsequent modules. Simulation results are shown below. Figure 5 As shown, the horizontal axis represents the radar frame number or time step, and the vertical axis represents the confidence level of the current pitch type. It can be seen that starting from around the second moment, the confidence level of each frame gradually increases, and the confidence level approaches 1 around the 20th moment.

[0164] This embodiment provides a target location classification method that treats the confidence level of the target's pitch type as a time-varying state variable. Based on the Bayesian criterion, it recursively updates the confidence level using the latest observation data, effectively utilizing observation information from multiple frames to improve the accuracy of target pitch classification. Simultaneously, it outputs a stable, time-smoothly changing confidence level result. Specifically, by dynamically setting the prior probability of the pitch type, the model's sensitivity to obstacle detection is increased, improving the robustness of the classification model. The application of the Bayesian criterion provides a unified framework for probability estimation and confidence assessment, increasing the reliability of the classification results.

[0165] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0166] Based on the same inventive concept, this application also provides a target location classification device for implementing the target location classification method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more target location classification device embodiments provided below can be found in the limitations of the target location classification method described above, and will not be repeated here.

[0167] In one embodiment, such as Figure 6 As shown, a target location classification device is provided, the target location classification device comprising:

[0168] The type prediction module 100 is used to determine the predicted location type of the target object based on the radar image features at the current moment.

[0169] The probability calculation module 200 is used to calculate the dynamic probability corresponding to each position type at the current time based on the predicted position type of the target object at the current time and historical time.

[0170] The confidence calculation module 300 is used to calculate the confidence of the target object at the current time for each position type based on the dynamic probability and the confidence of the target object at each position type in the historical time.

[0171] The type determination module 400 is used to determine the actual position type of the target object based on the confidence level at the current moment.

[0172] In one embodiment, the radar image features include a target location range and a signal-to-noise ratio range; the type prediction module 100 is further configured to:

[0173] Obtain the vehicle's own location range;

[0174] Based on the vehicle's position range, the target position range and the signal-to-noise ratio range are classified to predict the position type of the target object.

[0175] In one embodiment, the historical moment includes at least one historical moment under a preset time sliding window; the probability calculation module 200 is further configured to:

[0176] Based on the predicted location type of the target object at the current time and historical time, calculate the ratio of the occurrence frequency of each location type to the number of times corresponding to the preset time window, and obtain the historical occurrence ratio of each location type;

[0177] Based on the historical occurrence ratio, the dynamic probability corresponding to each location type at the current moment is determined.

[0178] In one embodiment, the probability calculation module 200 is further configured to:

[0179] Based on the radar image features and the vehicle's road data, the obstacle type of the target object is determined;

[0180] Based on the motion information of the target object, calculate the motion predictability score of the target object;

[0181] Based on the obstacle type and the motion predictability score, the historical occurrence ratio is weighted to obtain a weighted probability.

[0182] Based on the weighted probabilities, the dynamic probability corresponding to each location type at the current time is determined.

[0183] In one embodiment, the probability calculation module 200 is further configured to:

[0184] The dynamic probability is smoothed by a preset smoothing factor; the preset smoothing factor is dynamically determined based on the number of consecutive occurrences of the predicted position type at the current time.

[0185] In one embodiment, the location type at a historical moment includes a historical actual location type and a historical unknown location type; the confidence calculation module 300 is further used for:

[0186] The first calculation factor is determined based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time.

[0187] The second calculation factor is determined based on the dynamic probability of the current location type at the current moment, the confidence level of the historical actual location type at historical moments, and the confidence level of the historical unknown location type at historical moments.

[0188] The confidence level of the current location type at the current moment is determined based on the ratio of the first calculation factor to the second calculation factor.

[0189] In one embodiment, the confidence calculation module 300 is further configured to:

[0190] The first parameter value is obtained by multiplying the confidence level of the historical actual location type at a historical time with the dynamic probability of the current location type at the current time.

[0191] The second parameter value is obtained based on the confidence level of the historical actual location type at a historical moment and the dynamic probability of the historical unknown location type at the current moment.

[0192] The third parameter value is obtained by multiplying the confidence level of the historical unknown location type at a historical time with the dynamic probability of the current location type at the current time.

[0193] The first calculation factor is obtained based on the sum of the first parameter value, the second parameter value, and the third parameter value.

[0194] In one embodiment, the confidence calculation module 300 is further configured to:

[0195] Based on the confidence level of the actual historical location type at a historical moment and the confidence level of the unknown historical location type at a historical moment, the confidence level of type change is obtained;

[0196] The second calculation factor is determined based on the product of the confidence level of the type change and the dynamic probability of the current location type at the current moment.

[0197] Each module in the aforementioned target location classification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0198] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a target location classification method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0199] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0200] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target location classification method of any of the above embodiments:

[0201] Based on the radar image features at the current moment, determine the predicted location type of the target object;

[0202] Based on the predicted position type of the target object at the current time and historical time, calculate the dynamic probability corresponding to each position type at the current time;

[0203] Based on the dynamic probability and the confidence level of the target object for each location type in historical time, calculate the confidence level of the target object for each location type in the current time.

[0204] Based on the confidence level at the current moment, the actual location type of the target object is determined.

[0205] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0206] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0208] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A target location classification method, characterized in that, Applied to vehicles, the target location classification method includes: Based on the radar image features at the current moment, determine the predicted location type of the target object; Based on the predicted position type of the target object at the current time and historical time, calculate the dynamic probability corresponding to each position type at the current time; Based on the dynamic probability and the confidence level of the target object for each location type in historical time, calculate the confidence level of the target object for each location type in the current time. Based on the confidence level at the current moment, the actual location type of the target object is determined.

2. The target location classification method according to claim 1, characterized in that, The radar image features include the target location range and the signal-to-noise ratio range; determining the predicted location type of the target object based on the radar image features at the current moment includes: Obtain the vehicle's own location range; Based on the vehicle's position range, the target position range and the signal-to-noise ratio range are classified to predict the position type of the target object.

3. The target location classification method according to claim 2, characterized in that, The historical moments include at least one historical moment under a preset time sliding window; the calculation of the dynamic probability corresponding to each position type at the current moment based on the predicted position type of the target object at the current moment and the historical moments includes: Based on the predicted location type of the target object at the current time and historical time, calculate the ratio of the occurrence frequency of each location type to the number of times corresponding to the preset time window, and obtain the historical occurrence ratio of each location type; Based on the historical occurrence ratio, the dynamic probability corresponding to each location type at the current moment is determined.

4. The target location classification method according to claim 3, characterized in that, The determination of the dynamic probability corresponding to each location type at the current moment based on the historical occurrence ratio includes: Based on the radar image features and the vehicle's road data, the obstacle type of the target object is determined; Based on the motion information of the target object, calculate the motion predictability score of the target object; Based on the obstacle type and the motion predictability score, the historical occurrence ratio is weighted to obtain a weighted probability. Based on the weighted probabilities, the dynamic probability corresponding to each location type at the current time is determined.

5. The target location classification method according to claim 3, characterized in that, The step of determining the dynamic probability corresponding to each location type at the current moment based on the historical occurrence ratio further includes: The dynamic probability is smoothed by a preset smoothing factor; the preset smoothing factor is dynamically determined based on the number of consecutive occurrences of the predicted position type at the current time.

6. The target location classification method according to claim 1, characterized in that, The location type at a historical moment includes the historical actual location type and the historical unknown location type; the calculation of the confidence level of the target object at the current moment for each location type based on the dynamic probability and the confidence level of the target object for each location type at a historical moment includes: The first calculation factor is determined based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time. The second calculation factor is determined based on the dynamic probability of the current location type at the current moment, the confidence level of the historical actual location type at historical moments, and the confidence level of the historical unknown location type at historical moments. The confidence level of the current location type at the current moment is determined based on the ratio of the first calculation factor to the second calculation factor.

7. The target location classification method according to claim 6, characterized in that, The determination of the first calculation factor based on the confidence level of the historical actual location type at a historical time, the dynamic probability of the current location type at the current time, the dynamic probability of the historical unknown location type at the current time, and the confidence level of the historical unknown location type at a historical time includes: The first parameter value is obtained by multiplying the confidence level of the historical actual location type at a historical time with the dynamic probability of the current location type at the current time. The second parameter value is obtained based on the confidence level of the historical actual location type at a historical moment and the dynamic probability of the historical unknown location type at the current moment. The third parameter value is obtained by multiplying the confidence level of the historical unknown location type at a historical time with the dynamic probability of the current location type at the current time. The first calculation factor is obtained based on the sum of the first parameter value, the second parameter value, and the third parameter value.

8. The target location classification method according to claim 6, characterized in that, The determination of the second calculation factor based on the dynamic probability of the current location type at the current time, the confidence level of the historical actual location type at historical times, and the confidence level of the historical unknown location type at historical times includes: Based on the confidence level of the actual historical location type at a historical moment and the confidence level of the unknown historical location type at a historical moment, the confidence level of type change is obtained; The second calculation factor is determined based on the product of the confidence level of the type change and the dynamic probability of the current location type at the current moment.

9. A target location classification device, characterized in that, The target location classification device includes: The type prediction module is used to determine the predicted location type of a target object based on the radar image features at the current moment. The probability calculation module is used to calculate the dynamic probability corresponding to each position type at the current time based on the predicted position type of the target object at the current time and historical time. The confidence calculation module is used to calculate the confidence level of the target object at the current time for each location type based on the dynamic probability and the confidence level of the target object for each location type at historical times. The type determination module is used to determine the actual position type of the target object based on the confidence level at the current moment.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

Images