WO2023202313A1 - 位置预测方法、装置、电子设备及存储介质 - Google Patents

位置预测方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2023202313A1
WO2023202313A1 PCT/CN2023/083060 CN2023083060W WO2023202313A1 WO 2023202313 A1 WO2023202313 A1 WO 2023202313A1 CN 2023083060 W CN2023083060 W CN 2023083060W WO 2023202313 A1 WO2023202313 A1 WO 2023202313A1
Authority
WO
WIPO (PCT)
Prior art keywords
neuron
spiking
spiking neuron
abscissa
group
Prior art date
Application number
PCT/CN2023/083060
Other languages
English (en)
French (fr)
Inventor
王源
Original Assignee
北京大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京大学 filed Critical 北京大学
Publication of WO2023202313A1 publication Critical patent/WO2023202313A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present disclosure relates to the field of artificial intelligence technology, and in particular, to a location prediction method, device, electronic equipment and storage medium.
  • the vehicle-mounted detection system can predict the positions of surrounding pedestrians and other motor vehicles through position prediction algorithms, so as to make braking, steering and other operations in advance to reduce the probability of traffic accidents.
  • the existing position prediction of moving targets is mainly based on the sequence-to-sequence prediction network built by the long and short memory network, which predicts the position of the moving target based on multiple position coordinates in the historical path.
  • the sequence-to-sequence prediction network built by the long and short memory network has high computational complexity, resulting in a long input-to-output delay, which leads to low efficiency of prediction results in practical applications and unsatisfactory fast response effects for position prediction.
  • the present disclosure provides a location prediction method, device, electronic equipment and storage medium to solve the shortcomings of high computational complexity and low efficiency of prediction networks in the prior art.
  • the present disclosure provides a location prediction method, including:
  • the second impulse neuron is activated to obtain the third impulse neuron. Prediction of the position of the output of an activated spiking neuron in a two-pulse neuron.
  • the first spiking neuron includes three spiking neuron groups
  • activate the first pulse neuron in the position prediction model to obtain pulses corresponding to the three trajectory points output by the activated neuron in the first pulse neuron include:
  • any group includes abscissa spiking neurons and ordinate spiking neurons;
  • the coordinates applied to the trajectory point corresponding to any group, activating the pulse neurons in any group, and obtaining the pulse output by the activated neurons in any group include:
  • the second spiking neuron includes a second abscissa spiking neuron and a second ordinate spiking neuron;
  • the second pulse neuron in the position prediction model is activated based on the pulses corresponding to the three trajectory points, and the position prediction result output by the activated second pulse neuron is obtained, including:
  • a preset threshold is applied to activate the second abscissa spiking neuron and the second abscissa spiking neuron respectively.
  • the ordinate is the spiking neuron, and the abscissa of the position prediction result output by the activated spiking neuron in the second abscissa spiking neuron is obtained, as well as the abscissa of the output of the activated spiking neuron in the second ordinate spiking neuron.
  • the ordinate of the position prediction result is the spiking neuron, and the abscissa of the position prediction result output by the activated spiking neuron in the second abscissa spiking neuron is obtained, as well as the abscissa of the output of the activated spiking neuron in the second ordinate spiking neuron.
  • the connection strength between the activated abscissa spiking neuron and the second abscissa spiking neuron in any group is controlled based on the corresponding strength of any group. parameters, the index of the activated abscissa spiking neuron in any group, the index of the second abscissa spiking neuron and the total number of the second abscissa spiking neuron; in any group
  • the connection strength between the activated ordinate spiking neuron and the second ordinate spiking neuron is based on the intensity control parameter corresponding to any group and the index of the activated ordinate spiking neuron in any group. , the index of the second ordinate spiking neuron and the total number of the second ordinate spiking neuron are determined.
  • the corresponding intensity control parameters of each group in the three spiking neuron groups are calculated based on the following formula:
  • k represents the prediction time adjustment parameter
  • ⁇ 1 represents the intensity control parameter corresponding to the first group of the three spiking neuron groups
  • ⁇ 2 represents the second group of the three spiking neuron groups.
  • ⁇ 3 represents the intensity control parameter corresponding to the third group of the three spiking neuron groups
  • the trajectory point corresponding to the first group is the trajectory at the latest time among the three trajectory points point
  • the trajectory point corresponding to the second grouping is the trajectory point at the middle time of the three trajectory points
  • the trajectory point corresponding to the third grouping is the trajectory point at the earliest time among the three trajectory points.
  • the present disclosure also provides a location prediction device, including:
  • the determination unit is used to determine the coordinates of three trajectory points at equal time intervals in the historical trajectory
  • the input unit is used to activate the first pulse neuron in the position prediction model based on the coordinates of the three trajectory points, and obtain the first pulses corresponding to the three trajectory points output by the activated first pulse neuron.
  • the prediction unit is configured to activate the second impulse neuron in the position prediction model based on the first impulses corresponding to the three trajectory points, and obtain the position prediction result output by the activated second impulse neuron.
  • the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.
  • the processor executes the program, it implements any one of the above position prediction methods. .
  • the present disclosure also provides a non-transitory computer-readable storage medium on which a computer program is stored.
  • the computer program When executed by a processor, it implements any one of the above position prediction methods.
  • the present disclosure also provides a computer program product, including a computer program, which implements any one of the above position prediction methods when executed by a processor.
  • the position prediction, device, electronic equipment and storage medium activate the first pulse neuron in the position prediction model through three trajectory points with equal time intervals in the historical trajectory, and activate the first pulse neuron in the first pulse neuron based on the time interval in the historical trajectory.
  • the impulse output by the activated neuron, and the connection strength between the first impulse neuron and the second impulse neuron in the position prediction model activate the second impulse neuron, and obtain the position of the output of the activated neuron in the second impulse neuron.
  • Forecasting results the position prediction model is constructed using a spiking neural network. This model predicts position by activating spiking neurons, which reduces the computational complexity, shortens the input-to-output delay, improves the effect of the prediction results, and then Improved the effect of real-time response.
  • Figure 1 is one of the flow diagrams of the location prediction method provided by the present disclosure
  • Figure 2 is the second schematic flowchart of the location prediction method provided by the present disclosure
  • Figure 3 is a network structure diagram of the position prediction model provided by the present disclosure.
  • Figure 4 is a schematic structural diagram of a position prediction device provided by the present disclosure.
  • Figure 5 is a schematic structural diagram of an electronic device provided by the present disclosure.
  • the current moving target position prediction method uses a sequence-to-sequence prediction network built by a long and short memory network. Due to the high computational complexity of the long and short memory network and the long input-to-output delay, it is applied to the fast response effect of position prediction. not ideal. Therefore, how to reduce the computational complexity of predicted position and improve the efficiency of position prediction is an urgent technical problem to be solved in this field.
  • Figure 1 is one of the flow diagrams of the location prediction method provided by the present disclosure. As shown in Figure 1, the method includes:
  • the embodiment of the present disclosure uses the spiking neural network to build a position prediction model, which can reduce Reduce computational complexity and improve the efficiency of position prediction.
  • Step 110 determine the coordinates of three trajectory points at equal time intervals in the historical trajectory
  • the target detection algorithm is used to obtain a historical trajectory of the moving target.
  • the method of obtaining the position information of the moving target can be video, radar, etc.
  • the embodiments of the present disclosure are not limited.
  • the form of the target detection algorithm can be convolutional neural. Networks, traditional computer vision algorithms, etc., the embodiments of the present disclosure do not limit this.
  • Three trajectory points with equal time intervals indicate that three trajectory points A, B, and C are determined in chronological order. Then the time interval between A and B is equal to the time interval between B and C.
  • Step 120 based on the coordinates of the three trajectory points, activate the first pulse neuron in the position prediction model to obtain pulses corresponding to the three trajectory points output by the activated neuron in the first pulse neuron;
  • Step 130 Based on the pulses corresponding to the three trajectory points and the connection strength between the first pulse neuron and the second pulse neuron in the position prediction model, activate the second pulse neuron in the position prediction model to obtain the second pulse. Prediction of the position of the output of the activated spiking neuron in the neuron.
  • the position prediction model includes a first spiking neuron and a second spiking neuron.
  • the spiking neurons in the first spiking neurons corresponding to the coordinates of the three trajectory points are activated, and we get The pulses corresponding to the three trajectory points output by the activated pulse neuron in the first spiking neuron are then corresponding to the pulses corresponding to the three trajectory points, as well as the first spiking neuron and the second spiking neuron in the position prediction model.
  • the connection strength activates the spiking neuron in the second spiking neuron, and obtains the position prediction result of the output of the activated spiking neuron in the second spiking neuron.
  • the coordinate system where the coordinates of the three trajectory points are located is determined when generating the historical trajectory.
  • the first impulse neuron and the second impulse neuron in the position prediction model are based on the maximum future time information for predicting the position of the moving target. , determine the number of spiking neurons and the corresponding relationship between the index of each spiking neuron in the first spiking neuron and the second spiking neuron and the specified coordinate point in the coordinate system.
  • the corresponding relationship can be the specified coordinate point in the coordinate system.
  • any coordinate point in the coordinate system can determine a specified coordinate point through its coordinates, so that Realize the corresponding relationship between the index of each spike neuron in the first spike neuron and the second spike neuron and any coordinate point in the coordinate system.
  • any coordinate point in the coordinate system is calculated by calculating the distance from each specified coordinate point. , determine the specified coordinate point closest to the coordinate point, thereby realizing the corresponding relationship between the index of each spike neuron in the first spike neuron and the second spike neuron and any coordinate point in the coordinate system.
  • the designated coordinate points in the coordinate system can be determined based on certain rules.
  • the rule can be that the designated coordinate points are all coordinate points whose horizontal and vertical coordinates are integers within a certain range of the coordinate system.
  • the range of the coordinate system is If the abscissa is from 0 to 100 and the ordinate is from 0 to 100, the coordinates of the specified coordinate point are (i, j), 0 ⁇ i ⁇ 100, 0 ⁇ j ⁇ 100.
  • activating the spiking neurons in the first spiking neuron corresponding to the coordinates of the three trajectory points can be based on three
  • the corresponding relationship between the overall coordinates of the trajectory point and the index of the first spiking neuron can be used to activate the spiking neuron in the first spiking neuron.
  • the abscissa and ordinate of the three trajectory points can also be compared with the index of the first spiking neuron respectively.
  • the corresponding relationship is to activate the spiking neurons in the first spiking neuron.
  • the activated spiking neurons in the first spiking neuron include spiking neurons activated according to the abscissa and spiking neurons activated according to the ordinate.
  • This paper The disclosed embodiments do not limit this.
  • the corresponding relationship between the index of the spiking neuron in the first spiking neuron and the coordinate point of any point in the coordinate system is based on the correspondence between the index of each spiking neuron in the first spiking neuron and the specified coordinate point in the coordinate system. The relationship is certain.
  • the first spiking neuron in the position prediction model can be a group or three groups of spiking neurons.
  • the first spiking neuron is a group of spiking neurons, it is directly based on the group of spiking neurons according to the three trajectory points. coordinates to activate the spiking neurons in the group.
  • the first spiking neuron is three groups of spiking neurons, it is necessary to determine the three groups based on the index of each group of the three groups of spiking neurons and the index of the three trajectory points.
  • the corresponding trajectory points of each group of spiking neurons activate the spiking neurons in each group according to the corresponding trajectory points of each group.
  • the embodiments of the present disclosure do not limit this.
  • the coordinate point in the coordinate system corresponding to the index of the activated spiking neuron in the second spiking neuron is determined. This coordinate point is the position prediction result. .
  • the second pulse neuron is activated according to the pulses corresponding to the three trajectory points, and the relationship between each pulse neuron in the first pulse neuron and the second pulse neuron.
  • the connection strength of each spiking neuron in the neuron is obtained by activating the second spiking neuron.
  • the connection strength of each spiking neuron in the first spiking neuron and each spiking neuron in the second spiking neuron can be obtained by training.
  • the intensity control parameters are calculated based on the future time information of the predicted moving target position.
  • the corresponding relationship between the index of the spiking neuron in the second spiking neuron and the coordinate point of any point in the coordinate system is determined based on the corresponding relationship between the index of each spiking neuron in the second spiking neuron and the specified coordinate point in the coordinate system. of.
  • the position prediction method activates the first impulse neuron in the position prediction model through three trajectory points with equal time intervals in the historical trajectory, and based on the output of the activated neuron in the first impulse neuron
  • the pulse as well as the connection strength between the first pulse neuron and the second pulse neuron in the position prediction model, activates the second pulse neuron, and obtains the position prediction result of the output of the activated neuron in the second pulse neuron, achieving
  • a position prediction model is built with a spiking neural network. This model uses the coordinate information of three positions to predict the position by activating spiking neurons, which reduces the computational complexity, shortens the input-to-output delay, and improves the accuracy of the prediction results. effect, thereby improving the effect of real-time response.
  • the first spiking neuron in the above embodiment includes three spiking neuron groups, and step 120 includes:
  • the three trajectory points need to be processed in a serial manner, that is, the third trajectory point among the three trajectory points
  • a trajectory point activates the first pulse neuron.
  • the operation of the second trajectory point is performed.
  • the operation of the third trajectory point is performed.
  • this serial method requires waiting for the pulse neuron activated by the previous trajectory point to be reset before activating subsequent trajectory points, which results in low efficiency.
  • the spiking neurons are divided into three spiking neuron groups.
  • the index of the spiking neuron in each spiking neuron group is formed with the specified coordinate point in the coordinate system. According to the corresponding relationship, three trajectory points can be processed in parallel, that is, the spiking neurons in the three spiking neuron groups can be activated in parallel according to the three trajectory points, which improves the execution efficiency.
  • the indexes of the three trajectory points are determined based on the time sequence of the three trajectory points, and the indexes of the three spiking neuron groups are determined when the position prediction model is constructed. After determining the three trajectory points, on the condition that the indexes are the same, any one of the three spiking neuron groups activates the spiking neurons in the group based on the coordinates of the trajectory point with the same index as the group among the three trajectory points. And get the pulse output from the activated spiking neurons in this group.
  • the three groups of spiking neurons are grouped
  • the group with index 0 activates the spiking neuron of the group according to the coordinates of A, and outputs the pulse output by the activated spiking neuron of the group.
  • the group activation operation of index 1 and the group activation operation of index 2 are the same as the group activation of index 0. The operations are the same and will not be repeated here.
  • the number of spiking neurons in the three spiking neuron groups is the same, and there is a corresponding relationship between the index of the spiking neurons in each of the three spiking neuron groups and the specified coordinate point in the coordinate system, and According to the corresponding relationship, the corresponding relationship between the index of the spiking neuron in each group and any coordinate point in the coordinate system is determined.
  • the present disclosure also provides an embodiment, in which any one of the three spiking neuron groups in the above embodiment includes abscissa spiking neurons and ordinate spiking neurons; and in step 120, the corresponding coordinates of the trajectory point, activate the spiking neurons in the group, and obtain the pulse output by the activated neurons in the group, including:
  • the index of the spiking neuron in each of the three spiking neuron groups forms a corresponding relationship with the overall coordinate of the specified coordinate point in the coordinate system, it will result in an excessive number of spiking neurons in each group.
  • the horizontal axis coordinates of the coordinate system are 0 to 100 and the vertical axis coordinates are 0 to 100, then 10,000 spiking neurons are required to form a corresponding relationship with the specified coordinate points, which will cause the number of spiking neurons in each group to vary.
  • the increase in the coordinate range of the horizontal and vertical axes of the coordinate system increases exponentially, which in turn leads to an increase in computational complexity and a decrease in predicted position efficiency.
  • the spiking neurons in each group of spiking neurons are divided into abscissa spiking neurons and ordinate spiking neurons.
  • the index of the abscissa spiking neuron corresponds to the abscissa of the specified coordinate point in the coordinate system.
  • the ordinate spiking neuron The index of the element corresponds to the ordinate of the specified coordinate point in the coordinate system.
  • each of the three spiking neuron groups in the embodiment of the present disclosure only requires 200 spiking neurons to form a corresponding relationship between the abscissa and ordinate in the coordinate system, so that the three The number of spiking neurons in each group of spiking neurons is only the sum of the maximum coordinate values of the horizontal and vertical axes of the coordinate system. It is possible to complete the correspondence between the abscissa and ordinate of the specified coordinate point in the coordinate system with a small number of spiking neurons. relationship, reducing computational complexity and improving prediction efficiency.
  • the abscissa coordinate of the trajectory point corresponding to any one of the three spiking neuron groups activate the abscissa spiking neuron of the group, and obtain the pulse output by the activated abscissa spiking neuron of the group.
  • the corresponding The ordinate of the trajectory point activates the ordinate pulse neuron of the group, and obtains the pulse output by the activated ordinate pulse neuron of the group.
  • FIG. 2 is a second schematic flowchart of the location prediction method provided by the present disclosure. As shown in Figure 2, step 130 includes:
  • Step 131 based on the pulse output by the activated abscissa pulse neuron in any group, combined with the connection strength between the activated abscissa pulse neuron and the second abscissa pulse neuron in the group, obtain the corresponding third value of the group.
  • Step 132 Based on the membrane potential of the second abscissa spiking neuron and the membrane potential of the second ordinate spiking neuron corresponding to each of the three spiking neuron groups, determine the total membrane potential sum of the second abscissa spiking neuron.
  • the second ordinate is the total membrane potential of the spiking neuron;
  • Step 133 based on the total membrane potential of the second abscissa spiking neuron and the total membrane potential of the second ordinate spiking neuron, apply a preset threshold to respectively activate the second abscissa spiking neuron and the second ordinate spiking neuron. , and obtain the abscissa of the position prediction result output by the activated spiking neuron in the second abscissa spiking neuron, and the ordinate of the position prediction result output by the activated spiking neuron in the second ordinate spiking neuron.
  • the second spiking neuron is divided into the second abscissa spiking element used to predict the abscissa and the second ordinate spiking element used to predict the ordinate, a small number of spiking neurons can be used to predict the position and improve improves the efficiency of predicting location.
  • the second spiking neuron includes a second abscissa spiking neuron and a second ordinate spiking neuron.
  • the product of the connection strength of the activated abscissa spiking neuron in the second abscissa spiking neuron, and the pulse output by the activated ordinate spiking neuron in the group and the activated ordinate spiking neuron in the group The product of the connection strength of the neuron and each spiking neuron in the second abscissa spiking neuron is the membrane potential of the second abscissa spiking neuron and the membrane potential of the second ordinate spiking neuron corresponding to the group.
  • the total membrane potential and the second ordinate are the total membrane potential of the spiking neuron.
  • the total membrane potential of the second abscissa spiking neuron and the total membrane potential of the second ordinate spiking neuron obtained in step 132 are compared with the preset threshold respectively. If the second abscissa spiking neuron and the second ordinate spiking neuron are compared If the total membrane potential of the coordinate spiking neurons is greater than the preset threshold, the spiking neuron with the largest total membrane potential among the second abscissa spiking neurons and the spiking neuron with the largest total membrane potential among the second ordinate spiking neurons are respectively activated. element, and respectively obtain the abscissa and ordinate of the position prediction result of the output of the activated spiking neuron in the second abscissa spiking neuron and the second ordinate spiking neuron.
  • the abscissa spiking neuron index of any one of the three spiking neuron groups is the same as the second abscissa spiking neuron index
  • the ordinate spiking neuron index of the group is the same as the second ordinate spiking neuron index. same.
  • the connection strength of each spiking neuron in can be obtained through training, can also be a preset connection strength mapping relationship, or can be obtained dynamically according to the strength control parameters. The embodiments of the present disclosure do not limit this. Among them, the intensity control parameters are calculated based on the future time information of the predicted moving target position.
  • the total membrane potential of the second abscissa spiking neuron and the total membrane potential of the second ordinate spiking neuron are calculated using the following formula:
  • i is the index of the activated abscissa spiking neuron in the group
  • j is the index of the jth abscissa spiking neuron in the second abscissa spiking neuron.
  • S i is the pulse output by the activated abscissa spiking neuron in the group
  • W ij is the jth abscissa spiking neuron in the i-th abscissa spiking neuron and the second abscissa spiking neuron in the group
  • the connection strength; when predicting the ordinate in the position prediction result, i is the index of the activated ordinate spiking neuron in the group, j is the index of the second ordinate spiking neuron, and Si
  • the pulse output by the activated ordinate spiking neuron, W ij is the connection strength of the i-th ordinate spiking neuron in the group and the j-th abscissa spiking neuron in the second abscissa spiking neuron.
  • connection strength acquisition method which method includes:
  • connection strength between the activated abscissa spiking neuron and the second abscissa spiking neuron in any group is based on the intensity control parameter corresponding to the group, the index of the activated abscissa spiking neuron in the group, the second abscissa spiking neuron.
  • the index of the coordinate spiking neuron and the total number of the second abscissa spiking neuron are determined; the connection strength between the activated ordinate spiking neuron and the second ordinate spiking neuron in the group is controlled based on the corresponding strength of the group
  • the parameter, the index of the activated ordinate spiking neuron, the index of the second ordinate spiking neuron and the total number of the second ordinate spiking neuron are determined.
  • connection strength between the activated abscissa spiking neuron and the second abscissa spiking neuron of any of the three spiking neuron groups and the activated ordinate spiking neuron of the group and the second ordinate spiking neuron is calculated by the following formula:
  • ⁇ is the group intensity control parameter.
  • the index of the abscissa spiking neuron, m is the total number of abscissa spiking neurons in the group, and is also the total number of the second abscissa spiking neuron; when predicting the ordinate in the position prediction result, i is the score
  • the index of the activated ordinate spiking neuron in the group, j is the index of the jth ordinate spiking neuron in the second ordinate spiking neuron
  • m is the total number of ordinate spiking neurons in the group, which is also the second
  • the ordinate is the total number of spiking neurons. Among them, 0 ⁇ i ⁇ m, 0 ⁇ j ⁇ m.
  • the intensity control parameter ⁇ of each group in the three spiking neuron groups is different.
  • the intensity control parameter ⁇ of each group can be a preset position, or can also be dynamically determined based on the prediction time adjustment parameter, where the prediction time adjustment
  • the parameter represents the prediction of the position of the moving target after equal time intervals of the parameter. For example: assuming the time interval is ⁇ t and the intensity adjustment coefficient is 1, it means predicting the position of the moving target after one time interval ⁇ t.
  • the intensity adjustment coefficient is 2 means predicting the position of the moving target at the moment after two time intervals ⁇ t, and the embodiment of the present disclosure does not limit this.
  • the present disclosure provides an embodiment of a method for obtaining intensity control parameters corresponding to three spiking neuron groups.
  • the method includes:
  • k represents the prediction time adjustment parameter
  • ⁇ 1 represents the intensity control parameter corresponding to the first group of three spiking neuron groups
  • ⁇ 2 represents the intensity control parameter corresponding to the second group of three spiking neuron groups
  • ⁇ 3 represents the intensity control parameter corresponding to the third group of three spiking neuron groups
  • the trajectory point corresponding to the first group is the trajectory point at the latest moment among the three trajectory points
  • the trajectory point corresponding to the second group is three The trajectory point at the middle time of the trajectory point
  • the trajectory point corresponding to the third group is the trajectory point at the earliest time among the three trajectory points.
  • the predicted future moment position can be determined based on the preset time interval ⁇ t, the current moment t and the prediction time adjustment parameter k.
  • k is equal to 1
  • the predicted future moment is t+ ⁇ t.
  • k is equal to 2
  • the predicted future time is t+2 ⁇ t, that is, the formula for predicting the future time is t+k ⁇ t.
  • FIG. 3 is a network structure diagram of the location prediction model provided by the present disclosure.
  • ⁇ t in the figure represents the time interval
  • t represents the current moment
  • k represents the prediction time adjustment parameter.
  • the position prediction model includes an input layer and a prediction layer. Each input layer Each spiking neuron is fully connected to the prediction layer.
  • the input layer includes three spiking neuron groups.
  • the three spiking neuron groups are processed into three trajectory points respectively.
  • the three trajectory points are trajectories with equal time intervals in the historical trajectory.
  • the three trajectory points are (x 1 ,y 1 ,t), (x 2 ,y 2 ,t- ⁇ t), (x 3 ,y 3 ,t-2 ⁇ t), and the time to be predicted is t+k ⁇ t , (x, y, t+k ⁇ t) is the position that needs to be predicted.
  • the input layer receives the input of coordinate information of three trajectory points on the historical trajectory of the moving target.
  • the time interval of the three trajectory points is ⁇ t, and the moments corresponding to the three positions are t, t- ⁇ t, and t-2 ⁇ t respectively.
  • the integration-firing spiking neuron in the group corresponding to the trajectory point among the three spiking neuron groups in the input layer is activated, and a pulse is generated and transmitted to the prediction layer.
  • the input layer and the prediction layer are fully connected.
  • the second pulse neuron in the prediction layer receives three abscissa pulses and three ordinates from the input layer. According to the connection weight corresponding to the pulse neuron activated by the input layer Accumulated membrane potential.
  • the index corresponding to the second abscissa spiking neuron with the largest membrane potential in the final prediction layer is the abscissa of the position coordinate predicted at the current moment
  • the corresponding index of the second ordinate spiking neuron with the largest membrane potential in the prediction layer is The ordinate of the predicted position coordinate at the current moment.
  • the position prediction device provided by the present disclosure will be described below.
  • the position prediction device described below and the position prediction method described above can be referred to correspondingly.
  • Figure 4 is a schematic structural diagram of a position prediction device provided by the present disclosure. As shown in FIG. 4 , the device includes: a determination unit 410 , an input unit 420 and a prediction unit 430 .
  • the determination unit 410 is used to determine the coordinates of three trajectory points at equal time intervals in the historical trajectory;
  • the input unit 420 is used to activate the first pulse neuron in the position prediction model based on the coordinates of the three trajectory points, and obtain the first pulses corresponding to the three trajectory points output by the activated first pulse neuron;
  • the prediction unit 430 is configured to activate the second impulse neuron in the position prediction model based on the first impulses corresponding to the three trajectory points, and obtain the position prediction result output by the activated second impulse neuron.
  • the determination unit is used to determine the coordinates of three trajectory points at equal time intervals in the historical trajectory; the input unit is used to activate the first position prediction model based on the coordinates of the three trajectory points.
  • Spiking neuron get the three trajectories of the output of the first spiking neuron after activation The first pulse corresponding to each point;
  • the prediction unit is used to activate the second pulse neuron in the position prediction model based on the first pulse corresponding to the three trajectory points, and obtain the position prediction of the output of the activated second pulse neuron.
  • a position prediction model was constructed using a spiking neural network. This model performs position prediction by activating spiking neurons, which reduces the computational complexity, shortens the input-to-output delay, improves the effect of the prediction results, and thereby improves The effect of real-time response.
  • the first spiking neuron in the input unit 420 includes three spiking neuron groups, and the input unit 420 is specifically used to calculate the data based on the index of any one of the three spiking neuron groups and the three trajectory points. Index, determine the trajectory point corresponding to the group, and apply it to the coordinates of the trajectory point corresponding to the group, activate the spiking neurons in the group, and obtain the pulse output by the activated neuron in the group.
  • any one of the three spiking neuron groups in the input unit 420 includes abscissa spiking neurons and ordinate spiking neurons, and the input unit 420 includes:
  • the activation subunit is used to respectively activate the abscissa spiking neurons and the ordinate spiking neurons of the group based on the abscissa and ordinate of the trajectory point corresponding to any group, and obtain the activated abscissa spiking neurons in the group. and the pulses output by activated ordinate spiking neurons respectively.
  • the second spiking neuron in the prediction unit 430 includes a second abscissa spiking neuron and a second ordinate spiking neuron, and the prediction unit 430 includes:
  • the membrane potential calculation subunit is used to obtain based on the pulse output by the activated abscissa spiking neuron in any group, combined with the connection strength between the activated abscissa spiking neuron and the second abscissa spiking neuron in the group, to obtain The membrane potential of the second abscissa spiking neuron corresponding to the group; and based on the pulse output by the activated ordinate spiking neuron in the group, combining the activated ordinate spiking neuron and the second ordinate spiking neuron in the group The connection strength between them is used to obtain the membrane potential of the second ordinate spiking neuron corresponding to the group;
  • the total membrane potential determination subunit is used to determine the second abscissa spiking neuron based on the membrane potential of the second abscissa spiking neuron and the membrane potential of the second ordinate spiking neuron corresponding to each of the three spiking neuron groups.
  • the prediction subunit is used to apply a preset threshold based on the total membrane potential of the second abscissa spiking neuron and the total membrane potential of the second ordinate spiking neuron, respectively activating the second abscissa spiking neuron and the second abscissa spiking neuron.
  • the second ordinate spiking neuron and obtain the abscissa of the position prediction result of the output of the activated spiking neuron in the second abscissa spiking neuron, and the position prediction of the output of the activated spiking neuron in the second ordinate spiking neuron The ordinate of the result.
  • the membrane potential calculation subunit includes:
  • connection strength calculation subunit The connection strength between the activated abscissa pulse neuron and the second abscissa pulse neuron in any group is based on the intensity control parameter corresponding to the group, the activated abscissa pulse in the group The index of the neuron, the index of the second abscissa spiking neuron and the total number of the second abscissa spiking neuron are determined; the connection strength between the activated ordinate spiking neuron and the second ordinate spiking neuron in this group It is determined based on the intensity control parameter corresponding to the group, the index of the activated ordinate spiking neuron in the group, the index of the second ordinate spiking neuron, and the total number of the second ordinate spiking neuron.
  • connection strength calculation subunit includes:
  • the intensity control parameter calculation subunit is used to calculate the corresponding intensity control parameters of each group in the three spiking neuron groups based on the following formula:
  • k represents the prediction time adjustment parameter
  • ⁇ 1 represents the intensity control parameter corresponding to the first group of three spiking neuron groups
  • ⁇ 2 represents the intensity control parameter corresponding to the second group of three spiking neuron groups
  • ⁇ 3 represents the intensity control parameter corresponding to the third group of three spiking neuron groups
  • the trajectory point corresponding to the first group is the trajectory point at the latest moment among the three trajectory points
  • the trajectory point corresponding to the second group is three The trajectory point at the middle time of the trajectory point
  • the trajectory point corresponding to the third group is the trajectory point at the earliest time among the three trajectory points.
  • Figure 5 illustrates a schematic diagram of the physical structure of an electronic device.
  • the electronic device may include: a processor (processor) 510, a communications interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540.
  • the processor 510, the communication interface 520, and the memory 530 complete communication with each other through the communication bus 540.
  • Processor 510 may call the store Logic instructions in the processor 530 to execute a position prediction method, which method includes: determining the coordinates of three trajectory points at equal time intervals in the historical trajectory; activating the first pulse in the position prediction model based on the coordinates of the three trajectory points Neuron, obtain the pulses corresponding to the three trajectory points output by the activated neuron in the first pulse neuron; based on the pulses corresponding to the three trajectory points, as well as the first pulse neuron and the second pulse in the position prediction model The connection strength of the neuron activates the second spiking neuron, and obtains the position prediction result of the output of the activated spiking neuron in the second spiking neuron.
  • the above-mentioned logical instructions in the memory 530 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.
  • the technical solution of the present disclosure is essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .
  • the present disclosure also provides a computer program product.
  • the computer program product includes a computer program.
  • the computer program can be stored on a non-transitory computer-readable storage medium.
  • the computer can Execute the position prediction method provided by each of the above methods, which method includes: determining the coordinates of three trajectory points at equal time intervals in the historical trajectory; activating the first pulse neuron in the position prediction model based on the coordinates of the three trajectory points , obtain the pulses corresponding to the three trajectory points output by the activated neuron in the first pulse neuron; based on the pulses corresponding to the three trajectory points, as well as the first pulse neuron and the second pulse in the position prediction model The connection strength of the neuron activates the second spiking neuron, and obtains the position prediction result of the output of the activated spiking neuron in the second spiking neuron.
  • the present disclosure also provides a non-transitory computer-readable storage medium on which a computer program is stored.
  • the computer program is implemented when executed by a processor to perform the position prediction method provided by each of the above methods.
  • the method includes: Determine the coordinates of three trajectory points at equal time intervals in the historical trajectory; based on the coordinates of the three trajectory points, activate the first pulse neuron in the position prediction model to get The pulses corresponding to the three trajectory points output by the activated neuron in the first spiking neuron; the pulses corresponding to the three trajectory points respectively, as well as the first spiking neuron and the second spiking neuron in the position prediction model
  • the connection strength of the second spiking neuron is activated, and the position prediction result of the output of the activated spiking neuron in the second spiking neuron is obtained.
  • the device embodiments described above are only illustrative.
  • the units described as separate components may or may not be physically separated.
  • the components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Persons of ordinary skill in the art can understand and implement the method without any creative effort.
  • each embodiment can be implemented by software plus a necessary general hardware platform, and of course, it can also be implemented by hardware.
  • the computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., including a number of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Neurology (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

本公开提供一种位置预测方法、装置、电子设备及存储介质，其中方法包括：确定历史轨迹中，等时间间隔的三个轨迹点的坐标；基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到激活后的第一脉冲神经元输出的三个轨迹点分别对应的脉冲；基于三个轨迹点分别对应的脉冲与第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果。该方法通过以脉冲神经网络构建的位置预测模型，该模型根据输入的三个轨迹点的坐标信息，激活脉冲神经元并输出位置预测结果，减少了计算复杂度，缩短了输入到输出的时延，提高了预测结果的效果，进而提高了实时响应的效果。

Description

位置预测方法、装置、电子设备及存储介质
相关申请的交叉引用
本申请要求于2022年04月18日提交的申请号为2022104048413，发明名称为“位置预测方法、装置、电子设备及存储介质”的中国专利申请的优先权，其通过引用方式全部并入本文。
技术领域
本公开涉及人工智能技术领域，尤其涉及一种位置预测方法、装置、电子设备及存储介质。
背景技术
目前，运动目标的位置预测在机器人控制、自动驾驶、安防系统等领域都有重要的应用。例如在自动驾驶场景中，车载探测系统通过位置预测算法可以对周围的行人和其他机动车的位置进行预测，从而提前做出制动、转向等操作，降低交通事故发生的概率。
现有的运动目标的位置预测主要是通过长短记忆时网络搭建的序列到序列预测网络，根据历史路径中的多个位置坐标对运动目标的位置进行预测。但长短记忆时网络搭建的序列到序列预测网络计算复杂度高，导致输入到输出延时较长，从而导致实际应用中预测结果的效率低，位置预测的快速响应效果不理想。
发明内容
本公开提供一种位置预测方法、装置、电子设备及存储介质，用以解决现有技术中预测网络的计算复杂度高效率低的缺陷。
本公开提供一种位置预测方法，包括：
确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到所述第一脉冲神经元中被激活神经元输出的所述三个轨迹点分别对应的脉冲；
基于所述三个轨迹点分别对应的脉冲，以及所述第一脉冲神经元和所述位置预测模型中的第二脉冲神经元的连接强度，激活所述第二脉冲神经元，得到所述第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
根据本公开提供的一种位置预测方法，所述第一脉冲神经元包括三个脉冲神经元分组；
所述基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到所述第一脉冲神经元中被激活神经元输出的所述三个轨迹点分别对应的脉冲，包括：
基于所述三个脉冲神经元分组中任一分组的索引和所述三个轨迹点的索引，确定所述任一分组对应的轨迹点，并应用于所述任一分组对应的轨迹点的坐标，激活所述任一分组中的脉冲神经元，得到所述任一分组中被激活神经元输出的脉冲。
根据本公开提供的一种位置预测方法，所述任一分组包括横坐标脉冲神经元和纵坐标脉冲神经元；
所述应用于所述任一分组对应的轨迹点的坐标，激活所述任一分组中的脉冲神经元，得到所述任一分组中被激活神经元输出的脉冲，包括：
基于所述任一分组对应的轨迹点的横坐标和纵坐标，分别激活所述任一分组的横坐标脉冲神经元和纵坐标脉冲神经元，并得到所述任一分组中被激活横坐标脉冲神经元和被激活纵坐标脉冲神经元分别输出的脉冲。
根据本公开提供的一种位置预测方法，所述第二脉冲神经元包括第二横坐标脉冲神经元和第二纵坐标脉冲神经元；
所述基于所述三个轨迹点分别对应的脉冲，激活所述位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果，包括：
基于所述任一分组中被激活横坐标脉冲神经元输出的脉冲，结合所述任一分组中被激活横坐标脉冲神经元与所述第二横坐标脉冲神经元之间的连接 强度，得到所述任一分组对应的所述第二横坐标脉冲神经元的膜电位；以及基于所述任一分组中被激活纵坐标脉冲神经元输出的脉冲，结合所述任一分组中被激活纵坐标脉冲神经元与所述第二纵坐标脉冲神经元之间的连接强度，得到所述任一分组对应的所述第二纵坐标脉冲神经元的膜电位；
基于所述三个脉冲神经元分组中各分组对应的所述第二横坐标脉冲神经元的膜电位和所述第二纵坐标脉冲神经元的膜电位，确定第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位；
基于所述第二横坐标脉冲神经元的总膜电位和所述第二纵坐标脉冲神经元的总膜电位，应用预设阈值，分别激活所述第二横坐标脉冲神经元和所述第二纵坐标脉冲神经元，并得到所述第二横坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的横坐标，以及所述第二纵坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的纵坐标。
根据本公开提供的一种位置预测方法，所述任一分组中被激活横坐标脉冲神经元与所述第二横坐标脉冲神经元之间的连接强度是基于所述任一分组对应的强度控制参数、所述任一分组中被激活横坐标脉冲神经元的索引、所述第二横坐标脉冲神经元的索引以及所述第二横坐标脉冲神经元的总数确定的；所述任一分组中被激活纵坐标脉冲神经元与所述第二纵坐标脉冲神经元之间的连接强度是基于所述任一分组对应的强度控制参数、所述任一分组中被激活纵坐标脉冲神经元的索引、所述第二纵坐标脉冲神经元的索引以及所述第二纵坐标脉冲神经元的总数确定的。
根据本公开提供的一种位置预测方法，所述三个脉冲神经元分组中各组对应强度控制参数基于如下公式计算得到：


式中，k表示预测时间调整参数；ρ1表示所述三个脉冲神经元分组中的第一分组对应的强度控制参数；ρ2表示所述三个脉冲神经元分组中的第二分组 对应的强度控制参数；ρ3表示所述三个脉冲神经元分组中的第三分组对应的强度控制参数；所述第一分组对应的轨迹点是所述三个轨迹点中最晚时刻的轨迹点；所述第二分组对应的轨迹点是所述三个轨迹点中间时刻的轨迹点；所述第三分组对应的轨迹点是所述三个轨迹点中最早时刻的轨迹点。
本公开还提供一种位置预测装置，包括：
确定单元，用于确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
输入单元，用于基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到激活后的第一脉冲神经元输出的所述三个轨迹点分别对应的第一脉冲；
预测单元，用于基于所述三个轨迹点分别对应的第一脉冲，激活所述位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果。
本公开还提供一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如上述任一种所述位置预测方法。
本公开还提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如上述任一种所述位置预测方法。
本公开还提供一种计算机程序产品，包括计算机程序，所述计算机程序被处理器执行时实现如上述任一种所述位置预测方法。
本公开提供的位置预测、装置、电子设备及存储介质，通过历史轨迹中的时间间隔相等的三个轨迹点，激活位置预测模型中的第一脉冲神经元，并依据第一脉冲神经元中被激活神经元输出的脉冲，以及第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，并得到第二脉冲神经元中被激活神经元输出的位置预测结果，实现了以脉冲神经网络构建位置预测模型，该模型以激活脉冲神经元方式的进行位置预测，减少了计算复杂度，缩短了输入到输出的时延，提高了预测结果的效果，进而提高了实时响应的效果。
附图说明
为了更清楚地说明本公开实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本公开提供的位置预测方法的流程示意图之一；
图2是本公开提供的位置预测方法的流程示意图之二；
图3是本公开提供的位置预测模型网络结构图；
图4是本公开提供的位置预测装置的结构示意图；
图5是本公开提供的电子设备的结构示意图。
具体实施方式
为使本公开的目的、技术方案和优点更加清楚，下面将结合本公开中的附图，对本公开中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本公开一部分实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
目前的运动目标位置预测方法使用长短记忆时网络搭建的序列到序列预测网络，由于长短记忆时网络的计算复杂度高，输入到输出的延时较长，将其应用到位置预测的快速响应效果不理想。因此，如何减少预测位置计算复杂度，提升位置预测的效率是本领域亟待解决的技术问题。
针对于这一技术问题，本公开提供一种位置预测方法。图1是本公开提供的位置预测方法的流程示意图之一。如图1所示，该方法包括：
考虑到脉冲神经网络中脉冲神经元自身带有时间属性，适合处理时序相关的输入信息，并且脉冲神经网络的计算复杂度低，因此，本公开实施例通过脉冲神经网络构建位置预测模型，能够降低计算复杂度，提升位置预测的效率。
步骤110，确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
需要说明的是，使用目标检测算法得到运动目标的一段历史轨迹，运动目标位置信息的获取方式，可以是视频、雷达等，本公开实施例不作限制，目标检测算法的形式，可以是卷积神经网络、传统计算机视觉算法等，本公开实施例对此不作限制。等时间间隔的三个轨迹点表示以时间顺序确定三个轨迹点A、B和C，则A和B之间的时间间隔，与B和C之间的时间间隔相等。
步骤120，基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到第一脉冲神经元中被激活神经元输出的三个轨迹点分别对应的脉冲；
步骤130，基于三个轨迹点分别对应的脉冲，以及第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活位置预测模型中的第二脉冲神经元，得到第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
具体地，位置预测模型中包括第一脉冲神经元和第二脉冲神经元，根据三个轨迹点的坐标，激活与三个轨迹点的坐标对应的第一脉冲神经元中的脉冲神经元，得到第一脉冲神经元中被激活脉冲神经元输出的三个轨迹点分别对应的脉冲，然后根据三个轨迹点分别对应的脉冲，以及第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元中的脉冲神经元，得到第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
需要说明的是，三个轨迹点的坐标所在的坐标系是在生成历史轨迹时确定的，位置预测模型中的第一脉冲神经元和第二脉冲神经元依据预测运动目标位置的最大未来时刻信息，确定脉冲神经元的个数以及第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引与坐标系中指定坐标点的对应关系，该对应关系可以为坐标系中指定坐标点的坐标整体与第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引的对应关系，还可以是坐标系中指定坐标点的横坐标和纵坐标，分别与第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引的对应关系，本公开实施例对此不作限制。在第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引与坐标系中指定坐标点的对应关系之后，坐标系中任一坐标点可以通过其坐标确定一个指定坐标点，从而 实现第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引与坐标系中任一坐标点的对应关系，例如，坐标系中任一点坐标点通过计算与各指定坐标点的距离，确定离该坐标点最近的指定坐标点，从而实现第一脉冲神经元和第二脉冲神经元中的各脉冲神经元的索引与坐标系中任一坐标点的对应关系。其中，坐标系中指定坐标点可以是基于一定规则确定，该规则可以是指定坐标点为在坐标系的某一个范围内的横纵坐标为整数的全部坐标点，例如：坐标系的该范围是横坐标在0到100，纵坐标在0到100，则指定坐标点的坐标为(i,j)，0≤i≤100，0≤j≤100。
依据第一脉冲神经元中的各脉冲神经元的索引与坐标系中任一坐标点的对应关系，激活与三个轨迹点的坐标对应的第一脉冲神经元中的脉冲神经元可以根据三个轨迹点的整体坐标与第一脉冲神经元索引的对应关系，激活第一脉冲神经元中的脉冲神经元，还可以根据三个轨迹点的横坐标和纵坐标分别与第一脉冲神经元索引的对应关系，激活第一脉冲神经元中的脉冲神经元，此时，第一脉冲神经元中被激活的脉冲神经元包括依据横坐标激活的脉冲神经元和依据纵坐标激活的脉冲神经元，本公开实施例对此不作限制。其中，第一脉冲神经元中的脉冲神经元的索引与坐标系中任一点坐标点的对应关系是依据上述第一脉冲神经元中的各脉冲神经元的索引与坐标系中指定坐标点的对应关系确定的。
此外，位置预测模型中的第一脉冲神经元可以是一组或三组脉冲神经元，当第一脉冲神经元是一组脉冲神经元时，则直接基于该组脉冲神经元根据三个轨迹点的坐标，激活该组中的脉冲神经元，当第一脉冲神经元是三组脉冲神经元时，则需要依据三组脉冲神经元中各组的索引和三个轨迹点的索引，确定三组脉冲神经元中各组对应的轨迹点，根据各组对应的轨迹点，激活各组中的脉冲神经元，本公开实施例对此不作限制。
根据第二脉冲神经元的索引和坐标系中任一坐标点的对应关系，确定第二脉冲神经元中被激活脉冲神经元的索引对应的坐标系中坐标点，该坐标点即为位置预测结果。其中，第二脉冲神经元被激活脉冲神经元可以根据三个轨迹点分别对应的脉冲，以及第一脉冲神经元中各脉冲神经元与第二脉冲神 经元中各脉冲神经元的连接强度，激活第二脉冲神经元得到，其中，第一脉冲神经元中各脉冲神经元与第二脉冲神经元中各脉冲神经元的连接强度可以是训练得到的，还可以是预设的连接强度映射关系，或者根据强度控制参数动态获取得到，本公开实施例对此不作限制。其中，强度控制参数是根据预测运动目标位置的未来时刻信息计算得到。第二脉冲神经元中的脉冲神经元的索引与坐标系中任一点坐标点的对应关系是依据上述第二脉冲神经元中的各脉冲神经元的索引与坐标系中指定坐标点的对应关系确定的。
本公开实施例提供的位置预测方法，通过历史轨迹中的时间间隔相等的三个轨迹点，激活位置预测模型中的第一脉冲神经元，并依据第一脉冲神经元中被激活神经元输出的脉冲，以及第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，并得到第二脉冲神经元中被激活神经元输出的位置预测结果，实现了以脉冲神经网络构建位置预测模型，该模型使用三个位置的坐标信息，以激活脉冲神经元方式的进行位置预测，减少了计算复杂度，缩短了输入到输出的时延，提高了预测结果的效果，进而提高了实时响应的效果。
基于上述实施例，本公开还提供一实施例，上述实施例中的第一脉冲神经元包括三个脉冲神经元分组，且步骤120包括：
基于三个脉冲神经元分组中任一分组的索引和三个轨迹点的索引，确定任一分组对应的轨迹点，并应用于该分组对应的轨迹点的坐标，激活该分组中的脉冲神经元，得到该分组中被激活神经元输出的脉冲。
考虑到使用历史轨迹中的三个轨迹点的坐标来进行预测，若只使用一组脉冲神经元来进行位置预测，则需要串行的方式依次处理三个轨迹点，即三个轨迹点中第一个轨迹点激活第一脉冲神经元，得到第一脉冲神经元被激活神经元输出的该轨迹点的脉冲后，再进行第二个轨迹点的操作，第二个轨迹点完成后再进行第三个轨迹点的操作，这种串行的方式需要等待前一轨迹点激活的脉冲神经元复位以后才能进行后续轨迹点的激活，这样导致效率较低，因此，本公开实施例通过将第一脉冲神经元分为三个脉冲神经元分组，每一个脉冲神经元分组中的脉冲神经元的索引分别和坐标系中的指定坐标点形成 对应关系，这样可以并行处理三个轨迹点，即可以并行根据三个轨迹点，分别激活三个脉冲神经元分组中的脉冲神经元，提高了执行效率。
具体地，依据三个轨迹点的时刻顺序确定三个轨迹点的索引，三个脉冲神经元分组的索引，则在位置预测模型构建时确定。在确定三个轨迹点之后，以索引相同为条件，三个脉冲神经元分组中任一分组根据三个轨迹点中与该分组索引相同的轨迹点的坐标，激活该分组中的脉冲神经元，并得到该分组中被激活脉冲神经元输出的脉冲。例如：三个轨迹点A、B和C，则依据A、B和C的时刻顺序，确定A的索引为0，B的索引为1，C的索引为2，则三组脉冲神经元分组中索引为0的分组根据A的坐标，激活该分组的脉冲神经元，并输出该分组被激活脉冲神经元输出的脉冲，索引1的分组激活操作和索引2的分组激活操作与索引0的分组激活操作相同，此处不再赘述。
需要说明的是，三个脉冲神经元分组中脉冲神经元的个数相同，并且三个脉冲神经元分组中每一个分组中的脉冲神经元的索引与坐标系中指定坐标点存在对应关系，并依据该对应关系确定每一个分组中的脉冲神经元的索引与坐标系中任一坐标点的对应关系。
基于上述实施例，本公开还提供一实施例，上述实施例中的三个脉冲神经元分组中任一分组包括横坐标脉冲神经元和纵坐标脉冲神经元；并且步骤120中应用于该分组对应的轨迹点的坐标，激活该分组中的脉冲神经元，得到该分组中被激活神经元输出的脉冲，包括：
基于该分组对应的轨迹点的横坐标和纵坐标，分别激活该分组的横坐标脉冲神经元和纵坐标脉冲神经元，并得到该分组中被激活横坐标脉冲神经元和被激活纵坐标脉冲神经元分别输出的脉冲。
考虑到若三个脉冲神经元分组中每一个分组中的脉冲神经元的索引与坐标系中指定坐标点的整体坐标形成对应关系，则会导致每一个分组中的脉冲神经元的数量过多，例如，坐标系的横轴坐标为0到100，纵轴坐标为0到100，则需要10000个脉冲神经元和指定坐标点形成对应关系，则会导致每一个分组中的脉冲神经元的数量随着坐标系横纵轴坐标范围的增大而成指数级增高，进而导致计算复杂度增大，预测位置效率降低，因此，本公开实施例三 个脉冲神经元分组中每一个分组的脉冲神经元分为横坐标脉冲神经元和纵坐标脉冲神经元，横坐标脉冲神经元的索引和坐标系中指定坐标点的横坐标对应，纵坐标脉冲神经元的索引和坐标系中指定坐标点的纵坐标对应。以上文例子中的坐标系为例，本公开实施例的三个脉冲神经元分组中每一个分组只需要200个脉冲神经元即可完成和坐标系中横坐标与纵坐标形成对应关系，使得三个脉冲神经元分组中每一个分组的脉冲神经元数量仅为坐标系横纵轴最大坐标值的和，实现了以少量的脉冲神经元完成坐标系中指定坐标点的横坐标与纵坐标的对应关系，减少了计算复杂度，提高了预测效率。
具体地，根据三个脉冲神经元分组中任一分组对应的轨迹点的横坐标，激活该分组的横坐标脉冲神经元，得到该分组被激活横坐标脉冲神经元输出的脉冲，根据该分组对应的轨迹点的纵坐标，激活该分组的纵坐标脉冲神经元，得到该分组被激活纵坐标脉冲神经元输出的脉冲。
基于上述实施例，图2是本公开提供的位置预测方法的流程示意图之二。如图2所示，步骤130包括：
步骤131，基于任一分组中被激活横坐标脉冲神经元输出的脉冲，结合该分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元之间的连接强度，得到该分组对应的第二横坐标脉冲神经元的膜电位；以及基于该分组中被激活纵坐标脉冲神经元输出的脉冲，结合该分组中被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元之间的连接强度，得到该分组对应的第二纵坐标脉冲神经元的膜电位；
步骤132，基于三个脉冲神经元分组中各分组对应的第二横坐标脉冲神经元的膜电位和第二纵坐标脉冲神经元的膜电位，确定第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位；
步骤133，基于第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位，应用预设阈值，分别激活第二横坐标脉冲神经元和第二纵坐标脉冲神经元，并得到第二横坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的横坐标，以及第二纵坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的纵坐标。
考虑到将第二脉冲神经元分为用于预测横坐标的第二横坐标脉冲元和用于预测纵坐标的第二纵坐标脉冲元，可以使用少量的脉冲神经元，对位置进行预测，提高了预测位置的效率。
具体地，第二脉冲神经元包括第二横坐标脉冲神经元和第二纵坐标脉冲神经元，根据三个脉冲神经元分组中任一分组中被激活横坐标脉冲神经元输出的脉冲与该分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元中各脉冲神经元的连接强度的乘积，以及该分组中被激活纵坐标脉冲神经元输出的脉冲与该分组中被激活纵坐标脉冲神经元与第二横坐标脉冲神经元中各脉冲神经元的连接强度的乘积，分别得到该分组对应的第二横坐标脉冲神经元的膜电位和第二纵坐标脉冲神经元的膜电位。
对步骤131中得到的三个脉冲神经元分组中各分组对应的第二横坐标脉冲神经元的膜电位和第二纵坐标脉冲神经元的膜电位进行求和，得到第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位。
将步骤132中得到的第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位，分别和预设阈值进行对比，若第二横坐标脉冲神经元和第二纵坐标脉冲神经元中存在总膜电位大于预设阈值，则分别激活第二横坐标脉冲神经元中总膜电位最大的脉冲神经元和第二纵坐标脉冲神经元的中总膜电位最大的脉冲神经元，并分别得到第二横坐标脉冲神经元中和第二纵坐标脉冲神经元中被激活的脉冲神经元输出的位置预测结果的横坐标和纵坐标。
需要说明的是，三个脉冲神经元分组中任一分组的横坐标脉冲神经元索引和第二横坐标脉冲神经元索引相同，该分组的纵坐标脉冲神经元索引和第二纵坐标脉冲神经索引相同。三个脉冲神经元分组中各分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元中各脉冲神经元的连接强度，以及被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元中各脉冲神经元的连接强度可以经过训练得到，还可以是预设的连接强度映射关系，或者根据强度控制参数动态获取得到，本公开实施例对此不作限制。其中，强度控制参数是根据预测运动目标位置的未来时刻信息计算得到。
此外，第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位都使用如下公式计算得到：
Pj＝∑iSi*Wij
式中，在预测位置预测结果中的横坐标时，i为该分组中被激活的横坐标脉冲神经元的索引，j是第二横坐标脉冲神经元中第j个横坐标脉冲神经元的索引，Si为该分组中被激活的横坐标脉冲神经元输出的脉冲，Wij为该分组中第i个横坐标脉冲神经元和第二横坐标脉冲神经元中第j个横坐标脉冲神经元的连接强度；在预测位置预测结果中的纵坐标时，i为该分组中被激活的纵坐标脉冲神经元的索引，j是第二纵坐标脉冲神经元的索引，Si为该分组中被激活的纵坐标脉冲神经元输出的脉冲，Wij为该分组中第i个纵坐标脉冲神经元和第二横坐标脉冲神经元中第j个横坐标脉冲神经元的连接强度。
基于上述实施例，本公开提供连接强度获取方法的实施例，该方法包括：
任一分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元之间的连接强度是基于该分组对应的强度控制参数、该分组中被激活横坐标脉冲神经元的索引、第二横坐标脉冲神经元的索引以及第二横坐标脉冲神经元的总数确定的；该组中被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元之间的连接强度是基于该分组对应的强度控制参数、该中被激活纵坐标脉冲神经元的索引、第二纵坐标脉冲神经元的索引以及第二纵坐标脉冲神经元的总数确定的。
具体地，三个脉冲神经元分组中任一分组被激活横坐标脉冲神经元与第二横坐标脉冲神经元之间的连接强度以及该分组被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元之间的连接强度都是通过以下公式计算得到的：
式中，ρ为该分组强度控制参数，在预测位置预测结果中的横坐标时，i为该分组中被激活的横坐标脉冲神经元的索引，j是第二横坐标脉冲神经元中第j个横坐标脉冲神经元的索引，m为该分组中横坐标脉冲神经元的总数，也是第二横坐标脉冲神经元的总数；在预测位置预测结果中的纵坐标时，i为该分 组中被激活的纵坐标脉冲神经元的索引，j是第二纵坐标脉冲神经元中第j个纵坐标脉冲神经元的索引，m为该分组中纵坐标脉冲神经元的总数，也是第二纵坐标脉冲神经元的总数。其中，0≤i<m，0≤j<m。
需要说明的是，三个脉冲神经元分组中各组的强度控制参数ρ不相同，各组的强度控制参数ρ可以是预先设置的位置，还可以根据预测时间调整参数动态确定，其中预测时间调整参数表示预测该参数个等时间间隔之后的时刻运动目标的位置，例如：设时间间隔为Δt，强度调整系数为1则表示预测1个时间间隔Δt之后的时刻运动目标的位置，强度调整系数为2则表示预测2个时间间隔Δt之后的时刻运动目标的位置，本公开实施例对此不作限制。
基于上述实施例，本公开提供三个脉冲神经元分组对应的强度控制参数获取的方法的实施例，该方法包括：
三个脉冲神经元分组中各组对应强度控制参数基于如下公式计算得到：


式中，k表示预测时间调整参数；ρ1表示三个脉冲神经元分组中的第一分组对应的强度控制参数；ρ2表示三个脉冲神经元分组中的第二分组对应的强度控制参数；ρ3表示三个脉冲神经元分组中的第三分组对应的强度控制参数；第一分组对应的轨迹点是三个轨迹点中最晚时刻的轨迹点；第二分组对应的轨迹点是三个轨迹点中间时刻的轨迹点；第三分组对应的轨迹点是三个轨迹点中最早时刻的轨迹点。
需要说明的是，预测未来时刻的位置可以根据预设时间间隔△t、当前时刻t和预测时间调整参数k确定，当k等于1时，则预测未来时刻为t+△t，当k等于2时则预测未来时刻为t+2△t，即预测未来时刻的公式为t+k△t。
基于上述实施例，本公开提供一优选实施例，图3是本公开提供的位置预测模型网络结构图。如图3所示，图中△t表示时间间隔，t表示当前时刻，k表示预测时间调整参数，位置预测模型包括输入层和预测层，输入层中每一 个脉冲神经元都与预测层全连接，其中，输入层包括三个脉冲神经元分组，三个脉冲神经元分组分别处理是三个轨迹点，三个轨迹点是历史轨迹中时间间隔相等的轨迹点，三个轨迹点分别为(x1,y1,t)，(x2,y2,t-Δt)，(x3,y3,t-2Δt)，要预测的时间为t+kΔt，(x,y,t+kΔt)为需要预测的位置。
具体地，输入层接收运动目标历史轨迹上三个轨迹点的坐标信息的输入，三个轨迹点的时间间隔为Δt，三个位置对应的时刻分别为t，t-Δt，t-2Δt。根据输入轨迹点坐标，激活输入层中三个脉冲神经元分组中与该轨迹点对应的分组中的整合-发放脉冲神经元激活，并产生脉冲传递到预测层。输入层与预测层之间为全连接，预测层中的第二脉冲神经元接收到来自输入层的三个横坐标脉冲和三个纵坐标，根据输入层激活的脉冲神经元对应的连接权值累加膜电位。最终预测层中膜电位最大的第二横坐标脉冲神经元所对应索引即为当前时刻预测得到的位置坐标的横坐标，预测层中膜电位最大的第二纵坐标脉冲神经元所对应索引即为当前时刻预测得到的位置坐标的纵坐标。
下面对本公开提供的位置预测装置进行描述，下文描述的位置预测装置与上文描述的位置预测方法可相互对应参照。
图4是本公开提供的位置预测装置的结构示意图。如图4所示，该装置包括：确定单元410、输入单元420和预测单元430。
其中，
确定单元410，用于确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
输入单元420，用于基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到激活后的第一脉冲神经元输出的三个轨迹点分别对应的第一脉冲；
预测单元430，用于基于三个轨迹点分别对应的第一脉冲，激活位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果。
在本公开实施例中，通过确定单元，用于确定历史轨迹中，等时间间隔的三个轨迹点的坐标；输入单元，用于基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到激活后的第一脉冲神经元输出的三个轨迹 点分别对应的第一脉冲；预测单元，用于基于三个轨迹点分别对应的第一脉冲，激活位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果，实现了以脉冲神经网络构建位置预测模型，该模型以激活脉冲神经元方式的进行位置预测，减少了计算复杂度，缩短了输入到输出的时延，提高了预测结果的效果，进而提高了实时响应的效果。
基于上述任一实施例，输入单元420中的第一脉冲神经元包括三个脉冲神经元分组，输入单元420具体用于基于三个脉冲神经元分组中任一分组的索引和三个轨迹点的索引，确定该分组对应的轨迹点，并应用于该分组对应的轨迹点的坐标，激活该分组中的脉冲神经元，得到该分组中被激活神经元输出的脉冲。
基于上述任一实施例，输入单元420中的三个脉冲神经元分组中任一分组包括横坐标脉冲神经元和纵坐标脉冲神经元，输入单元420包括：
激活子单元，用于基于任一分组对应的轨迹点的横坐标和纵坐标，分别激活该分组的横坐标脉冲神经元和纵坐标脉冲神经元，并得到该分组中被激活横坐标脉冲神经元和被激活纵坐标脉冲神经元分别输出的脉冲。
基于上述任一实施例，预测单元430中的第二脉冲神经元包括第二横坐标脉冲神经元和第二纵坐标脉冲神经元，预测单元430包括：
膜电位计算子单元，用于基于任一分组中被激活横坐标脉冲神经元输出的脉冲，结合该分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元之间的连接强度，得到该分组对应的第二横坐标脉冲神经元的膜电位；以及基于该分组中被激活纵坐标脉冲神经元输出的脉冲，结合该分组中被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元之间的连接强度，得到该分组对应的第二纵坐标脉冲神经元的膜电位；
总膜电位确定子单元，用于基于三个脉冲神经元分组中各分组对应的第二横坐标脉冲神经元的膜电位和第二纵坐标脉冲神经元的膜电位，确定第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位；
预测子单元，用于基于第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位，应用预设阈值，分别激活第二横坐标脉冲神经元和 第二纵坐标脉冲神经元，并得到第二横坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的横坐标，以及第二纵坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的纵坐标。
基于上述任一实施例，膜电位计算子单元包括：
连接强度计算子单元：用于任一分组中被激活横坐标脉冲神经元与第二横坐标脉冲神经元之间的连接强度是基于该分组对应的强度控制参数、该分组中被激活横坐标脉冲神经元的索引、第二横坐标脉冲神经元的索引以及第二横坐标脉冲神经元的总数确定的；该分组中被激活纵坐标脉冲神经元与第二纵坐标脉冲神经元之间的连接强度是基于该分组对应的强度控制参数、该分组中被激活纵坐标脉冲神经元的索引、第二纵坐标脉冲神经元的索引以及第二纵坐标脉冲神经元的总数确定的。
基于上述任一实施例，连接强度计算子单元包括：
强度控制参数计算子单元，用于三个脉冲神经元分组中各组对应强度控制参数基于如下公式计算得到：


式中，k表示预测时间调整参数；ρ1表示三个脉冲神经元分组中的第一分组对应的强度控制参数；ρ2表示三个脉冲神经元分组中的第二分组对应的强度控制参数；ρ3表示三个脉冲神经元分组中的第三分组对应的强度控制参数；第一分组对应的轨迹点是三个轨迹点中最晚时刻的轨迹点；第二分组对应的轨迹点是三个轨迹点中间时刻的轨迹点；第三分组对应的轨迹点是三个轨迹点中最早时刻的轨迹点。
图5示例了一种电子设备的实体结构示意图，如图5所示，该电子设备可以包括：处理器(processor)510、通信接口(Communications Interface)520、存储器(memory)530和通信总线540，其中，处理器510，通信接口520，存储器530通过通信总线540完成相互间的通信。处理器510可以调用存储 器530中的逻辑指令，以执行位置预测方法，该方法包括：确定历史轨迹中，等时间间隔的三个轨迹点的坐标；基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到第一脉冲神经元中被激活神经元输出的三个轨迹点分别对应的脉冲；基于三个轨迹点分别对应的脉冲，以及第一脉冲神经元和位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，得到第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
此外，上述的存储器530中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
另一方面，本公开还提供一种计算机程序产品，所述计算机程序产品包括计算机程序，计算机程序可存储在非暂态计算机可读存储介质上，所述计算机程序被处理器执行时，计算机能够执行上述各方法所提供的位置预测方法，该方法包括：确定历史轨迹中，等时间间隔的三个轨迹点的坐标；基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到第一脉冲神经元中被激活神经元输出的三个轨迹点分别对应的脉冲；基于三个轨迹点分别对应的脉冲，以及第一脉冲神经元和所述位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，得到第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
又一方面，本公开还提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现以执行上述各方法提供的位置预测方法，该方法包括：确定历史轨迹中，等时间间隔的三个轨迹点的坐标；基于三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到 第一脉冲神经元中被激活神经元输出的三个轨迹点分别对应的脉冲；基于三个轨迹点分别对应的脉冲，以及第一脉冲神经元和所述位置预测模型中的第二脉冲神经元的连接强度，激活第二脉冲神经元，得到第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
以上所描述的装置实施例仅仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下，即可以理解并实施。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件。基于这样的理解，上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。
最后应说明的是：以上实施例仅用以说明本公开的技术方案，而非对其限制；尽管参照前述实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本公开各实施例技术方案的精神和范围。

Claims (10)

  1. 一种位置预测方法，其特征在于，包括：
    确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
    基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到所述第一脉冲神经元中被激活神经元输出的所述三个轨迹点分别对应的脉冲；
    基于所述三个轨迹点分别对应的脉冲，以及所述第一脉冲神经元和所述位置预测模型中的第二脉冲神经元的连接强度，激活所述第二脉冲神经元，得到所述第二脉冲神经元中被激活脉冲神经元输出的位置预测结果。
  2. 根据权利要求1所述的位置预测方法，其特征在于，所述第一脉冲神经元包括三个脉冲神经元分组；
    所述基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到所述第一脉冲神经元中被激活神经元输出的所述三个轨迹点分别对应的脉冲，包括：
    基于所述三个脉冲神经元分组中任一分组的索引和所述三个轨迹点的索引，确定所述任一分组对应的轨迹点，并应用于所述任一分组对应的轨迹点的坐标，激活所述任一分组中的脉冲神经元，得到所述任一分组中被激活神经元输出的脉冲。
  3. 根据权利要求2所述的位置预测方法，其特征在于，所述任一分组包括横坐标脉冲神经元和纵坐标脉冲神经元；
    所述应用于所述任一分组对应的轨迹点的坐标，激活所述任一分组中的脉冲神经元，得到所述任一分组中被激活神经元输出的脉冲，包括：
    基于所述任一分组对应的轨迹点的横坐标和纵坐标，分别激活所述任一分组的横坐标脉冲神经元和纵坐标脉冲神经元，并得到所述任一分组中被激活横坐标脉冲神经元和被激活纵坐标脉冲神经元分别输出的脉冲。
  4. 根据权利要求3所述的位置预测方法，其特征在于，所述第二脉冲神经元包括第二横坐标脉冲神经元和第二纵坐标脉冲神经元；
    所述基于所述三个轨迹点分别对应的脉冲，激活所述位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果，包括：
    基于所述任一分组中被激活横坐标脉冲神经元输出的脉冲，结合所述任一分组中被激活横坐标脉冲神经元与所述第二横坐标脉冲神经元之间的连接强度，得到所述任一分组对应的所述第二横坐标脉冲神经元的膜电位；以及基于所述任一分组中被激活纵坐标脉冲神经元输出的脉冲，结合所述任一分组中被激活纵坐标脉冲神经元与所述第二纵坐标脉冲神经元之间的连接强度，得到所述任一分组对应的所述第二纵坐标脉冲神经元的膜电位；
    基于所述三个脉冲神经元分组中各分组对应的所述第二横坐标脉冲神经元的膜电位和所述第二纵坐标脉冲神经元的膜电位，确定第二横坐标脉冲神经元的总膜电位和第二纵坐标脉冲神经元的总膜电位；
    基于所述第二横坐标脉冲神经元的总膜电位和所述第二纵坐标脉冲神经元的总膜电位，应用预设阈值，分别激活所述第二横坐标脉冲神经元和所述第二纵坐标脉冲神经元，并得到所述第二横坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的横坐标，以及所述第二纵坐标脉冲神经元中被激活脉冲神经元输出的位置预测结果的纵坐标。
  5. 根据权利要求4所述的位置预测方法，其特征在于，所述任一分组中被激活横坐标脉冲神经元与所述第二横坐标脉冲神经元之间的连接强度是基于所述任一分组对应的强度控制参数、所述任一分组中被激活横坐标脉冲神经元的索引、所述第二横坐标脉冲神经元的索引以及所述第二横坐标脉冲神经元的总数确定的；所述任一分组中被激活纵坐标脉冲神经元与所述第二纵坐标脉冲神经元之间的连接强度是基于所述任一分组对应的强度控制参数、所述任一分组中被激活纵坐标脉冲神经元的索引、所述第二纵坐标脉冲神经元的索引以及所述第二纵坐标脉冲神经元的总数确定的。
  6. 根据权利要求5所述的位置预测方法，其特征在于，所述三个脉冲神经元分组中各组对应强度控制参数基于如下公式计算得到：


    式中，k表示预测时间调整参数；ρ1表示所述三个脉冲神经元分组中的第一分组对应的强度控制参数；ρ2表示所述三个脉冲神经元分组中的第二分组对应的强度控制参数；ρ3表示所述三个脉冲神经元分组中的第三分组对应的强度控制参数；所述第一分组对应的轨迹点是所述三个轨迹点中最晚时刻的轨迹点；所述第二分组对应的轨迹点是所述三个轨迹点中间时刻的轨迹点；所述第三分组对应的轨迹点是所述三个轨迹点中最早时刻的轨迹点。
  7. 一种位置预测装置，其特征在于，包括：
    确定单元，用于确定历史轨迹中，等时间间隔的三个轨迹点的坐标；
    输入单元，用于基于所述三个轨迹点的坐标，激活位置预测模型中的第一脉冲神经元，得到激活后的第一脉冲神经元输出的所述三个轨迹点分别对应的第一脉冲；
    预测单元，用于基于所述三个轨迹点分别对应的第一脉冲，激活所述位置预测模型中的第二脉冲神经元，得到激活后的第二脉冲神经元输出的位置预测结果。
  8. 一种电子设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现如权利要求1至6任一项所述位置预测方法。
  9. 一种非暂态计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至6任一项所述位置预测方法。
  10. 一种计算机程序产品，包括计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至6任一项所述位置预测方法。
PCT/CN2023/083060 2022-04-18 2023-03-22 位置预测方法、装置、电子设备及存储介质 WO2023202313A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210404841.3 2022-04-18
CN202210404841.3A CN114936331A (zh) 2022-04-18 2022-04-18 位置预测方法、装置、电子设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023202313A1 true WO2023202313A1 (zh) 2023-10-26

Family

ID=82862804

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/083060 WO2023202313A1 (zh) 2022-04-18 2023-03-22 位置预测方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN114936331A (zh)
WO (1) WO2023202313A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114936331A (zh) * 2022-04-18 2022-08-23 北京大学 位置预测方法、装置、电子设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110928993A (zh) * 2019-11-26 2020-03-27 重庆邮电大学 基于深度循环神经网络的用户位置预测方法及系统
CN111275742A (zh) * 2020-01-19 2020-06-12 北京大学 一种目标识别方法、装置、系统及计算机可读存储介质
CN112733452A (zh) * 2021-01-13 2021-04-30 北京嘀嘀无限科技发展有限公司 轨迹预测方法、装置、电子设备和可读存储介质
CN113255905A (zh) * 2021-07-16 2021-08-13 成都时识科技有限公司 脉冲神经网络中神经元的信号处理方法及该网络训练方法
US20220011122A1 (en) * 2020-07-09 2022-01-13 Beijing Tusen Weilai Technology Co., Ltd. Trajectory prediction method and device
CN114936331A (zh) * 2022-04-18 2022-08-23 北京大学 位置预测方法、装置、电子设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110928993A (zh) * 2019-11-26 2020-03-27 重庆邮电大学 基于深度循环神经网络的用户位置预测方法及系统
CN111275742A (zh) * 2020-01-19 2020-06-12 北京大学 一种目标识别方法、装置、系统及计算机可读存储介质
US20220011122A1 (en) * 2020-07-09 2022-01-13 Beijing Tusen Weilai Technology Co., Ltd. Trajectory prediction method and device
CN112733452A (zh) * 2021-01-13 2021-04-30 北京嘀嘀无限科技发展有限公司 轨迹预测方法、装置、电子设备和可读存储介质
CN113255905A (zh) * 2021-07-16 2021-08-13 成都时识科技有限公司 脉冲神经网络中神经元的信号处理方法及该网络训练方法
CN114936331A (zh) * 2022-04-18 2022-08-23 北京大学 位置预测方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN114936331A (zh) 2022-08-23

Similar Documents

Publication Publication Date Title
WO2021012752A1 (zh) 一种基于脉冲神经网络的短程跟踪方法及系统
CN107203134B (zh) 一种基于深度卷积神经网络的前车跟随方法
WO2022121510A1 (zh) 基于随机策略梯度的交通信号控制方法、系统及电子设备
CN108256577B (zh) 一种基于多线激光雷达的障碍物聚类方法
CN113408743B (zh) 联邦模型的生成方法、装置、电子设备和存储介质
WO2021249071A1 (zh) 一种车道线的检测方法及相关设备
KR20230026961A (ko) 장애물 이동 궤적의 예측 방법, 장치 및 자율 운전 차량
CN113264066B (zh) 障碍物轨迹预测方法、装置、自动驾驶车辆及路侧设备
WO2023202313A1 (zh) 位置预测方法、装置、电子设备及存储介质
US20230367934A1 (en) Method and apparatus for constructing vehicle dynamics model and method and apparatus for predicting vehicle state information
CN109606367A (zh) 基于车联网的巡航控制系统的最优线性控制方法及装置
CN112446468A (zh) 用于创建脉冲神经网络的方法、设备和计算机程序
KR20220091607A (ko) 차량-도로 협업 중의 타겟 검출 방법, 장치 및 노변 디바이스
WO2020037500A1 (zh) 智能驾驶方法、装置及存储介质
WO2023082850A1 (zh) 行人轨迹预测方法、装置及存储介质
US20190188542A1 (en) Using Deep Video Frame Prediction For Training A Controller Of An Autonomous Vehicle
WO2022222647A1 (zh) 车辆意图预测方法、装置、设备及存储介质
CN111882047A (zh) 一种基于强化学习与线性规划的快速空管防冲突方法
CN114067166A (zh) 用于确定物理对象的物理特性的设备和方法
WO2023011172A1 (zh) 一种驾驶行为确定方法及其相关设备
CN114419884B (zh) 基于强化学习和相位竞争的自适应信号控制方法及系统
CN113239472B (zh) 一种基于强化学习的导弹制导方法和装置
KR20220042315A (ko) 교통 데이터 예측 방법, 장치 및 전자 기기
CN114882713A (zh) 基于多场景的信号控制方法、系统、设备和存储介质
CN114239405A (zh) 一种脉冲神经网络神经元膜电压计算方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23790974

Country of ref document: EP

Kind code of ref document: A1