Pedestrian trajectory prediction methods, devices, equipment and computer-readable storage media

Abstract

A method, apparatus, device, and computer-readable storage medium for predicting pedestrian trajectories are disclosed. The method includes: acquiring the actual trajectory coordinates of a pedestrian at multiple acquisition times; calculating, one by one, the initial predicted coordinates of the pedestrian at multiple times within a future period based on a preset first model and the actual trajectory coordinates; wherein each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times; optimizing the initial predicted coordinates at the target time based on a preset second model, the actual trajectory coordinates and / or the initial predicted coordinates before the target time, and the initial predicted coordinates after the target time, to obtain the final predicted coordinates at the target time. This method comprehensively considers pedestrian trajectories using global trajectory data, making the final prediction result more consistent with the global movement logic of pedestrians and achieving higher accuracy.

Description

Technical Field

[0001] This application relates to the field of pedestrian trajectory prediction technology, specifically to a pedestrian trajectory prediction method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] Pedestrian trajectory prediction is a key research area in computer vision and autonomous driving, playing a crucial role in driver assistance, video surveillance, and service robots. Pedestrian trajectory prediction is the task of predicting the future trajectory coordinates of each pedestrian based on their historical trajectories. Previous approaches have primarily involved designing models based on social forces, which model pedestrian interactions in dynamic scenes by incorporating various forces (attraction and repulsion). Generally, pedestrian trajectory prediction is modeled in two dimensions (spatial and temporal). Some existing techniques simulate social interactions between pedestrians in the spatial dimension (interaction), utilizing Social-LSTM (Long Short-Term Memory) models to integrate social features through social merging. Social-GAN (Generative Adversarial Networks) models extend the ability of social aggregation with several new merging schemes. The STGAT model employs a Graph Attention Network (GAT), which propagates spatial and temporal interactions between different pedestrians. In addition, some recent technical solutions based on Transformer utilize the self-attention mechanism in Transformer to fully explore the temporal relationship of the historical trajectory features of a single pedestrian, which can also solve this task well.

[0003] Existing Transformer-based solutions employ a step-by-step modeling approach. They predict the next frame's trajectory coordinates based on the temporal relationships of pedestrian trajectory points and the observed coordinates. The model calculates the feature attention relationships between each trajectory point using a self-attention mechanism based on the input trajectory features of the first frame, predicting the coordinates of the next frame's trajectory points frame by frame. However, this process only considers local trajectory point feature attention within the entire trajectory sequence, neglecting the crucial information of global trajectory feature attention. This leads to a problem: if the model only considers local trajectory feature attention, and the predicted trajectory point coordinates in a certain frame deviate significantly from the true value, the model will incorporate the incorrect prediction from the previous step into subsequent trajectory point predictions, resulting in accumulated errors. This leads to poor prediction performance in complex scenarios and for irregular pedestrian movements such as acceleration, deceleration, and stillness, resulting in a certain degree of performance degradation. Summary of the Invention

[0004] This application provides a method, apparatus, device, and computer-readable storage medium for predicting pedestrian trajectories, which can solve the related technical problems existing in the prior art.

[0005] In a first aspect, embodiments of this application provide a method for predicting pedestrian trajectories, employing the following technical solution:

[0006] A method for predicting pedestrian trajectories, characterized in that the method includes:

[0007] Obtain the actual trajectory coordinates of pedestrians at multiple data collection times;

[0008] Based on the preset first model and the actual trajectory coordinates, the initial predicted coordinates of pedestrians at multiple times in the future are calculated one by one; wherein, each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times.

[0009] Based on the preset second model, the actual trajectory coordinates before the target time and / or the initial predicted coordinates, and the initial predicted coordinates after the target time, the initial predicted coordinates at the target time are optimized to obtain the final predicted coordinates at the target time.

[0010] In conjunction with the first aspect, in one implementation, the step of optimizing the initial predicted coordinates at the target time based on a preset second model, the actual trajectory coordinates before the target time and / or the initial predicted coordinates, and the initial predicted coordinates after the target time, to obtain the final predicted coordinates at the target time, is as follows:

[0011] The process involves optimizing multiple moments within a future time period as target moments to obtain the final predicted coordinate points at these target moments. The optimization steps for each target moment include:

[0012] The final predicted coordinates at the target time are obtained by optimizing the actual trajectory coordinates and the final predicted coordinates before the target time, as well as the initial predicted coordinates after the target time.

[0013] In conjunction with the first aspect, in one implementation, based on a preset second model, the actual trajectory coordinates before and after the target time, and the initial predicted coordinates, the initial predicted coordinates at the target time are optimized to obtain the final predicted coordinates at the target time.

[0014] The target time is optimized based on the actual trajectory coordinates and the initial predicted coordinates before the multiple target times, and the initial predicted coordinates after the corresponding target times, to obtain the final predicted coordinates at the multiple target times.

[0015] In conjunction with the first aspect, in one implementation, the second model is obtained by performing a global training method on an initial second model, the global training method comprising the following steps:

[0016] Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0017] Obtain multiple first predicted coordinates of pedestrians within the second time period, based on the first real trajectory coordinates of the first model;

[0018] The second model obtains the first optimized coordinates of the pedestrian within the second time period based on the first real trajectory coordinates and the first predicted coordinates.

[0019] Based on the loss between the first optimized coordinate point and the second true trajectory coordinate point, the second model is trained and updated until it meets the set loss requirements.

[0020] In conjunction with the first aspect, in one implementation, the second model is obtained by performing a global training method on an initial second model, the global training method comprising the following steps:

[0021] Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0022] Obtain multiple first predicted coordinates of pedestrians within the second time period, based on the first real trajectory coordinates of the first model;

[0023] By fusing the first predicted coordinate point and the second actual trajectory coordinate point, a second predicted coordinate point is obtained;

[0024] The second optimized coordinates of the pedestrian within the second time period are obtained by the second model based on the first real trajectory coordinates and the second predicted coordinates.

[0025] Based on the loss between the second optimized coordinate point and the second true trajectory coordinate point, the second model is trained and updated until it meets the set loss requirements.

[0026] In conjunction with the first aspect, in one implementation, the second model is obtained by performing a global training method on an initial second model, the global training method comprising the following steps:

[0027] Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0028] Obtain multiple first predicted coordinates of pedestrians within the second time period, based on the first real trajectory coordinates of the first model;

[0029] By fusing the first predicted coordinate point and the second actual trajectory coordinate point, a second predicted coordinate point that is closer to the actual trajectory is obtained;

[0030] The second model on the first training route is used to obtain the first optimized coordinate point based on the first real trajectory coordinate point and the first predicted coordinate point;

[0031] The second model on the second training route obtains the second optimized coordinates of the pedestrian in the second time period based on the first real trajectory coordinates and the second predicted coordinates.

[0032] Based on the loss of the first optimized coordinate point and the second optimized coordinate point, the second model is trained and updated on the first training path until it meets the set loss requirements.

[0033] In conjunction with the first aspect, in one implementation, after obtaining the second optimized coordinates of the pedestrian within the second time period based on the first true trajectory coordinates and the second predicted coordinates on the second training route, the process includes the following steps:

[0034] Based on the loss between the second optimized coordinate point and the second true trajectory coordinate point, the second model is trained and updated on the second training route until it meets the set loss requirements.

[0035] Secondly, embodiments of this application provide a pedestrian trajectory prediction device, which adopts the following technical solution:

[0036] A pedestrian trajectory prediction device, the device comprising:

[0037] The acquisition module is configured to acquire the actual trajectory coordinates of a pedestrian at multiple acquisition times;

[0038] The first prediction module is configured to calculate the initial predicted coordinates of a pedestrian at multiple times within a future period based on a preset first model and the actual trajectory coordinates; wherein each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times.

[0039] The second prediction module is configured to optimize the initial prediction coordinates at the target time based on a preset second model, the actual trajectory coordinates before the target time and / or the initial prediction coordinates, and the initial prediction coordinates after the target time, to obtain the final prediction coordinates at the target time.

[0040] Thirdly, embodiments of this application provide a pedestrian trajectory prediction device, which adopts the following technical solution:

[0041] A pedestrian trajectory prediction device includes a processor, a memory, and a pedestrian trajectory prediction program stored in the memory and executable by the processor, wherein when the pedestrian trajectory prediction program is executed by the processor, it implements the steps of the pedestrian trajectory prediction method as described above.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium, employing the following technical solution:

[0043] A computer-readable storage medium storing a pedestrian trajectory prediction program, wherein when executed by a processor, the pedestrian trajectory prediction program implements the steps of the pedestrian trajectory prediction method as described above.

[0044] The beneficial effects of the technical solutions provided in this application include:

[0045] After the first model calculates the initial predicted coordinates for each time period in the future based on the relevant coordinates already acquired, the second model will again optimize each initial predicted coordinate by using all the acquired and predicted real trajectory coordinates and initial predicted coordinates. This achieves comprehensive consideration of pedestrian trajectory using global trajectory data. Compared with existing technologies that only use historical data before the current time for local prediction, the final prediction result is more in line with the global movement logic of pedestrians and has higher accuracy. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an embodiment of the pedestrian trajectory prediction method of this application;

[0047] Figure 2 This is a schematic diagram of the initial prediction coordinate point prediction process in the pedestrian trajectory prediction method of this application;

[0048] Figure 3 This is a schematic diagram of the final predicted coordinate point prediction process in the pedestrian trajectory prediction method of this application;

[0049] Figure 4 This is a schematic diagram of the functional modules of an embodiment of the pedestrian trajectory prediction device of this application;

[0050] Figure 5 This is a schematic diagram of the hardware structure of the pedestrian trajectory prediction device involved in the embodiments of this application. Detailed Implementation

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0052] First, some of the technical terms used in this application will be explained to help those skilled in the art understand this application.

[0053] A method for predicting pedestrian trajectories, characterized in that the method includes:

[0054] Obtain the actual trajectory coordinates of pedestrians at multiple data collection times;

[0055] Based on the preset first model and the actual trajectory coordinates, the initial predicted coordinates of pedestrians at multiple times in the future are calculated one by one; wherein, each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times.

[0056] Based on the preset second model, the actual trajectory coordinates before the target time and / or the initial predicted coordinates, and the initial predicted coordinates after the target time, the initial predicted coordinates at the target time are optimized to obtain the final predicted coordinates at the target time.

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0058] In a first aspect, embodiments of this application provide a method for predicting pedestrian trajectories.

[0059] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the pedestrian trajectory prediction method of this application. Figure 1 As shown, pedestrian trajectory prediction methods include:

[0060] S100: Obtain the actual trajectory coordinates of the pedestrian at multiple data collection times;

[0061] S200. Based on the preset first model and the actual trajectory coordinates, calculate the initial predicted coordinates of the pedestrian at multiple times in the future period; wherein, each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times.

[0062] S300. Based on the preset second model, the actual trajectory coordinates before the target time and / or the initial predicted coordinates, and the initial predicted coordinates after the target time, optimize the initial predicted coordinates at the target time to obtain the final predicted coordinates at the target time.

[0063] In this embodiment, after the first model calculates the initial predicted coordinate points for each time period in the future based on the relevant coordinate points already acquired, the second model will again optimize each initial predicted coordinate point by using all the acquired and predicted real trajectory coordinate points and initial predicted coordinate points. This achieves the comprehensive consideration of pedestrian trajectory using global trajectory data. Compared with the existing technology that only uses historical data before the current time for local prediction, the final prediction result is more in line with the global movement logic of pedestrians and has higher accuracy.

[0064] The process by which the first model obtains the initial predicted coordinate points one by one from the real trajectory coordinate points and the initial predicted coordinate points at all previous moments can be referred to as follows: Figure 2The process by which the second model optimizes the initial predicted coordinates at the target time using the true trajectory coordinates and / or the initial predicted coordinates before the target time, and the initial predicted coordinates after the target time, can be found in [reference needed]. Figure 3 .

[0065] Furthermore, in some embodiments, step S300 involves optimizing the initial predicted coordinates at the target time based on a preset second model, the actual trajectory coordinates before the target time, and / or the initial predicted coordinates after the target time, to obtain the final predicted coordinates at the target time.

[0066] The process involves optimizing multiple moments within a future time period as target moments to obtain the final predicted coordinate points at these target moments. The optimization steps for each target moment include:

[0067] The final predicted coordinates at the target time are obtained by optimizing the actual trajectory coordinates and the final predicted coordinates before the target time, as well as the initial predicted coordinates after the target time.

[0068] This configuration ensures that when optimizing the initial predicted coordinates at each target time, the previously optimized final predicted coordinates and the actual trajectory coordinates are used, guaranteeing that the initial predicted coordinates being optimized can be optimized with the highest accuracy.

[0069] In other embodiments, step S300, based on a preset second model, the actual trajectory coordinates before and after the target time, and the initial predicted coordinates, optimizes the initial predicted coordinates at the target time to obtain the final predicted coordinates at the target time.

[0070] The target time is optimized based on the actual trajectory coordinates and the initial predicted coordinates before the multiple target times, and the initial predicted coordinates after the corresponding target times, to obtain the final predicted coordinates at the multiple target times.

[0071] This configuration allows for simultaneous optimization of initial predicted coordinates for multiple target time points without waiting for optimization results from previous target time points, thus improving optimization efficiency.

[0072] Furthermore, in one embodiment, the second model is obtained by completing a global training method from the initial second model. The global training method may be different in different embodiments. This application specifically discloses the following three global training methods: global training method A, global training method B, and global training method C.

[0073] For global training method A, it includes the following steps:

[0074] A100: Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0075] A200. Obtain multiple first predicted coordinate points of pedestrians within the second time period, based on the first real trajectory coordinate points predicted by the first model.

[0076] A300: Obtain the first optimized coordinates of the pedestrian in the second time period based on the first real trajectory coordinates and the first predicted coordinates of the second model;

[0077] A400. Based on the loss between the first optimized coordinate point and the second true trajectory coordinate point, train and update the second model until it meets the set loss requirements.

[0078] For global training method B, it includes the following steps:

[0079] B100: Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0080] B200: Obtain multiple first predicted coordinate points of pedestrians within the second time period, based on the first real trajectory coordinate points predicted by the first model;

[0081] B300, merge the first predicted coordinate point and the second actual trajectory coordinate point to obtain the second predicted coordinate point;

[0082] B400: Obtain the second optimized coordinates of the pedestrian within the second time period, based on the first real trajectory coordinates and the second predicted coordinates of the second model.

[0083] B500. Based on the loss between the second optimized coordinate point and the second true trajectory coordinate point, train and update the second model until it meets the set loss requirements.

[0084] For global training method C, it includes the following steps:

[0085] C100: Obtain the first true trajectory coordinates of a pedestrian within a first time period and the second true trajectory coordinates of a pedestrian within a second time period; wherein the first time period and the second time period are continuous.

[0086] C200: Obtain multiple first predicted coordinate points of pedestrians within the second time period, based on the first real trajectory coordinate points predicted by the first model;

[0087] C300: Merge the first predicted coordinate point and the second true trajectory coordinate point to obtain a second predicted coordinate point that is closer to the true trajectory;

[0088] C400: Obtain the first optimized coordinate point by the second model on the first training route based on the first real trajectory coordinate point and the first predicted coordinate point;

[0089] C500: Obtain the second optimized coordinates of the pedestrian in the second time period based on the first real trajectory coordinates and the second predicted coordinates on the second training route;

[0090] C600. Based on the loss of the first optimized coordinate point and the second optimized coordinate point, train and update the second model on the first training route until it meets the set loss requirements.

[0091] C700. Based on the loss of the second optimized coordinate point and the second true trajectory coordinate point, train and update the second model on the second training route until it meets the set loss requirements.

[0092] Among them, the global training method A has a faster training process, but it requires a large number of first true trajectory coordinate points and second true trajectory coordinate points to form the first optimized coordinate points and compare the first optimized coordinate points.

[0093] However, if only the initial predictions of the first model are used as input to the second model, errors may exist in the initial predictions of the first model at intermediate steps. If these errors are not corrected, the accumulated error in subsequent modeling processes will become increasingly large, thus failing to achieve optimized results. To address this issue, a Global Temporal Guidance Module (GTG) is introduced during the training phase to work in conjunction with the second model to learn predictions that more closely approximate the true trajectory. Therefore, in step B300 of global training method B and step C300 of global training method C, the initial predictions of the first model and the true trajectory are mapped to the same feature space. These embedded features are then concatenated to form a fused feature representation, resulting in second predicted coordinates that more closely approximate the true trajectory. Through the guidance of GTG, the overall model can learn more accurate trajectory prediction embedding features, facilitating more efficient training of the second model using the second predicted coordinates.

[0094] In addition, in global training method C, besides training the second model in the first route (the same as global training method B) quickly and efficiently, the second model using the initial prediction results as input will also be trained in the second route, so that the final effect can still be close to the real trajectory. This makes the second model trained in the second route more in line with the user's analysis and prediction process.

[0095] For the step of training and updating the second model based on loss in the above methods, the difference between the two can be compared and evaluated using a loss function. The loss function is a non-negative real-valued function that measures the difference between the model's predicted value and the actual trajectory value. Generally, the smaller the loss calculated by the loss function, the better the model's robustness. Initially, the parameters in the entire model are not at the values ​​corresponding to the best model performance; they are default parameters. In each round of neural network training, the difference between the feedforward result and the true value of each iteration is calculated, thus guiding the model's parameters to calculate in the correct direction for the next step. After a sufficient number of iterations of training, the model's calculated loss becomes sufficiently small, at which point the model's prediction performance reaches its optimal level.

[0096] Secondly, embodiments of this application also provide a pedestrian trajectory prediction device.

[0097] In one embodiment, reference is made to Figure 4 , Figure 4 This is a schematic diagram of the functional modules of an embodiment of the pedestrian trajectory prediction device of this application. Figure 4 As shown, the pedestrian trajectory prediction device includes:

[0098] The acquisition module is configured to acquire the actual trajectory coordinates of a pedestrian at multiple acquisition times;

[0099] The first prediction module is configured to calculate the initial predicted coordinates of a pedestrian at multiple times within a future period based on a preset first model and the actual trajectory coordinates; wherein each initial predicted coordinate is obtained based on the actual trajectory coordinates and the initial predicted coordinates at all previous times.

[0100] The second prediction module is configured to optimize the initial prediction coordinates at the target time based on a preset second model, the actual trajectory coordinates before the target time and / or the initial prediction coordinates, and the initial prediction coordinates after the target time, to obtain the final prediction coordinates at the target time.

[0101] The functions of each module in the above-mentioned pedestrian trajectory prediction device correspond to the steps in the above-mentioned pedestrian trajectory prediction method embodiment, and their functions and implementation processes will not be described in detail here.

[0102] Thirdly, embodiments of this application provide a pedestrian trajectory prediction device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0103] Reference Figure 5 , Figure 5 This is a schematic diagram of the hardware structure of the pedestrian trajectory prediction device involved in the embodiments of this application. In the embodiments of this application, the pedestrian trajectory prediction device may include a processor, a memory, a communication interface, and a communication bus.

[0104] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0105] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting internal components of the pedestrian trajectory prediction device, as well as interfaces used for interconnecting the pedestrian trajectory prediction device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0106] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0107] The processor can be a general-purpose processor, which can call the pedestrian trajectory prediction program stored in the memory and execute the pedestrian trajectory prediction method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the pedestrian trajectory prediction program is called can be referred to in the various embodiments of the pedestrian trajectory prediction method of this application, and will not be repeated here.

[0108] Those skilled in the art will understand that Figure 5The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0109] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0110] The present application provides a computer-readable storage medium storing a pedestrian trajectory prediction program, wherein when the pedestrian trajectory prediction program is executed by a processor, it implements the steps of the pedestrian trajectory prediction method described above.

[0111] The method implemented when the pedestrian trajectory prediction program is executed can be referred to in various embodiments of the pedestrian trajectory prediction method of this application, and will not be repeated here.

[0112] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0113] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0114] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0115] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0116] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0117] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0118] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for pedestrian trajectory prediction, the method comprising: The method comprises: acquiring real trajectory coordinate points of pedestrians at multiple collection time points; according to a preset first model and the real trajectory coordinate points, calculating initial prediction coordinate points of the pedestrians at multiple time points in a future period of time one by one; wherein each initial prediction coordinate point is obtained according to all the real trajectory coordinate points at previous time points and the initial prediction coordinate points; according to a preset second model, the real trajectory coordinate points before a target time point and / or the initial prediction coordinate points, and the initial prediction coordinate points after the target time point, optimizing the initial prediction coordinate points at the target time point to obtain final prediction coordinate points at the target time point, the second model being obtained by completing a global training method on an initial second model; in the step of optimizing the initial prediction coordinate points at the target time point according to the preset second model, the real trajectory coordinate points before the target time point and / or the initial prediction coordinate points, and the initial prediction coordinate points after the target time point to obtain the final prediction coordinate points at the target time point, optimizing each of the multiple time points in the future period of time one by one to obtain final prediction coordinate points at multiple target time points, wherein the step of optimizing the target time point comprises: optimizing the target time point according to the real trajectory coordinate points before the target time point and the final prediction coordinate points, and the initial prediction coordinate points after the target time point to obtain the final prediction coordinate points at the target time point.

2. The pedestrian trajectory prediction method of claim 1, wherein, in the step of optimizing the initial prediction coordinate points at the target time point according to the preset second model, the real trajectory coordinate points before the target time point and the initial prediction coordinate points, and the initial prediction coordinate points after the target time point to obtain the final prediction coordinate points at the target time point, optimizing the target time point according to the real trajectory coordinate points before the target time point and the initial prediction coordinate points, and the initial prediction coordinate points after the target time point to obtain the final prediction coordinate points at the target time point. 3.The pedestrian trajectory prediction method of claim 1, wherein, The global training method comprises the following steps: acquiring first real trajectory coordinate points of pedestrians in a first time period and second real trajectory coordinate points of pedestrians in a second time period; wherein the first time period and the second time period are continuous; acquiring multiple first prediction coordinate points of the pedestrians in the second time period predicted by the first model based on the first real trajectory coordinate points; acquiring first optimization coordinate points of the pedestrians in the second time period obtained by the second model based on the first real trajectory coordinate points and the first prediction coordinate points; training and updating the second model according to the loss between the first optimization coordinate points and the second real trajectory coordinate points until the loss meets a set requirement.

4. The pedestrian trajectory prediction method of claim 1, wherein, The global training method comprises the following steps: acquiring first real trajectory coordinate points of pedestrians in a first time period and second real trajectory coordinate points of pedestrians in a second time period; wherein the first time period and the second time period are continuous; acquiring multiple first prediction coordinate points of the pedestrians in the second time period predicted by the first model based on the first real trajectory coordinate points; fuse the first predicted coordinate point and the second real trajectory coordinate point to obtain a second predicted coordinate point; obtain a second optimized coordinate point of the pedestrian in the second time period based on the first real trajectory coordinate point and the second predicted coordinate point obtained by the second model; train and update the second model according to the loss of the second optimized coordinate point and the second real trajectory coordinate point until the set loss requirement is met.

5. The pedestrian trajectory prediction method of claim 1, wherein, The global training method comprises the following steps: obtain a first real trajectory coordinate point of a pedestrian in a first time period and a second real trajectory coordinate point of a pedestrian in a second time period; wherein the first time period and the second time period are continuous; obtain a plurality of first predicted coordinate points of the pedestrian in the second time period predicted by the first model based on the first real trajectory coordinate point; fuse the first predicted coordinate point and the second real trajectory coordinate point to obtain a second predicted coordinate point that is closer to the real trajectory; obtain a first optimized coordinate point of the second model based on the first real trajectory coordinate point and the first predicted coordinate point on a first training route; obtain a second optimized coordinate point of the pedestrian in the second time period obtained by the second model based on the first real trajectory coordinate point and the second predicted coordinate point on a second training route; train and update the second model on the first training route according to the loss of the first optimized coordinate point and the second optimized coordinate point until the set loss requirement is met.

6. The pedestrian trajectory prediction method of claim 5, wherein, After obtaining the second optimized coordinate point of the pedestrian in the second time period obtained by the second model based on the first real trajectory coordinate point and the second predicted coordinate point on the second training route, the following steps are included: train and update the second model on the second training route according to the loss of the second optimized coordinate point and the second real trajectory coordinate point until the set loss requirement is met.

7. A pedestrian trajectory prediction apparatus characterized by comprising: The device comprises: an acquisition module configured to acquire real trajectory coordinate points of a pedestrian at a plurality of collection time points; a first prediction module configured to calculate initial predicted coordinate points of the pedestrian at a plurality of time points in a future period of time one by one according to a preset first model and the real trajectory coordinate points; wherein each initial predicted coordinate point is obtained according to all previous time point real trajectory coordinate points and the initial predicted coordinate points; a second prediction module configured to optimize the initial predicted coordinate point at a target time point according to a preset second model, the real trajectory coordinate points and / or the initial predicted coordinate points before the target time point, and the initial predicted coordinate points after the target time point, to obtain a final predicted coordinate point at the target time point, the second model being obtained after the initial second model completes a global training method; wherein the second prediction module is further configured to: optimize the initial predicted coordinate point at a target time point according to a preset second model, the real trajectory coordinate points and / or the initial predicted coordinate points before the target time point, and the initial predicted coordinate points after the target time point, to obtain a final predicted coordinate point at the target time point, The future time period is divided into multiple time points, and each time point is taken as a target time point to perform optimization, to obtain final prediction coordinate points at the multiple target time points; The final prediction coordinate points at the target time point are obtained by performing optimization on the real trajectory coordinate points before the target time point, the final prediction coordinate points, and the initial prediction coordinate points after the target time point.

8. A pedestrian trajectory prediction device characterized by comprising: The pedestrian trajectory prediction device comprises a processor, a memory, and a pedestrian trajectory prediction program stored on the memory and executable by the processor, wherein the pedestrian trajectory prediction program, when executed by the processor, implements the steps of the pedestrian trajectory prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a pedestrian trajectory prediction program, wherein the pedestrian trajectory prediction program, when executed by the processor, implements the steps of the pedestrian trajectory prediction method according to any one of claims 1 to 6.

Images