CN115470418B - Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography - Google Patents
Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography Download PDFInfo
- Publication number
- CN115470418B CN115470418B CN202211124537.XA CN202211124537A CN115470418B CN 115470418 B CN115470418 B CN 115470418B CN 202211124537 A CN202211124537 A CN 202211124537A CN 115470418 B CN115470418 B CN 115470418B
- Authority
- CN
- China
- Prior art keywords
- queuing
- target detection
- time
- detection point
- point
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
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
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
- G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Marketing (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Operations Research (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Physiology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of data prediction, in particular to a queuing point recommending method and a queuing point recommending system based on unmanned aerial vehicle aerial photography, which are characterized in that a YOLOv5 algorithm is used as a real-time target detection algorithm, and compared with other algorithms, the single-stage algorithm is faster in running speed, higher in average detection precision, the queuing situation of target detection points at future time points is predicted through a Markov prediction model, the number of queuing people and the speed after a user starts to reach the target detection points in a certain travel mode can be predicted, the accuracy and the practicability of the example can be improved, and in the case of statistics of queuing people with a plurality of target detection points nearby, optimal options can be selected for users according to the existing data and historical data, so that the users can conveniently make travel planning.
Description
Technical Field
The invention relates to the technical field of data prediction, in particular to a queuing point recommending method and system based on unmanned aerial vehicle aerial photography.
Background
In recent years, with the rapid development of target detection, data processing and real-time sharing technologies, technologies of shooting specific scenes such as classrooms, offices, site safety detection and the like through fixed shooting equipment and processing images by utilizing a high-performance server have been mature, for example, chinese patent CN202110301889.7 discloses a subway platform guide queuing system based on YOLOv3 face detection, a face detection subsystem is used for acquiring passenger queuing images of waiting areas of various platforms of a subway, a YOLOv3 algorithm is used for carrying out face real-time detection, queuing number information is matched with the waiting areas, queuing numbers corresponding to the matched waiting areas of various platforms are visually displayed, but in the prior art, the counting of outdoor scenes is not influenced by multiple factors, for example, if residents can determine the time of going to nucleic acid detection points by knowing the queuing numbers of the nucleic acid detection points, and a great deal of time can be saved. For example, in a stadium (track and field), a certain sports group committee wants to know the number of people in the field, but for dense occasions, the naked eye counting efficiency is extremely low, the flexibility of fixed positions is poor, and a method for counting unmanned aerial vehicles in real time is available nowadays, such as a method and a system for counting unmanned aerial vehicle platform people based on image deep learning, but a feasible method for predicting the number of people in a queue is still lacking.
Disclosure of Invention
In view of the above, the invention aims to provide a queuing point recommending method and a queuing point recommending system based on unmanned aerial vehicle aerial photography, so as to solve the problem that the prior art lacks a method for predicting the number of queuing people in an outdoor scene.
Based on the above purpose, the invention provides a queuing point recommending method based on unmanned aerial vehicle aerial photography, which comprises the following steps:
intercepting a picture captured by an unmanned aerial vehicle to obtain queuing images of all target detection points;
counting the number of people in the queuing images of the detection points through a YOLOv5 algorithm to obtain the number of people in the queuing of each target detection point in real time;
clustering queuing data of each target detection point by using dynamic time warping as a distance measure;
inputting the clustered historical queuing data into a Markov original model, and training the original model to obtain a prediction model;
calculating the time when the user reaches the target detection point according to the user specified time, the specified position and the target detection point position;
inputting current queuing data into a Markov prediction model, predicting the number of queuing people when a user reaches a target detection point, and predicting the time required by queuing;
calculating the estimated total consumption time according to the time required for queuing and the time spent for the user to go to and from the target detection point;
and comparing the estimated total time spent by the user in selecting different target detection points, and outputting the target detection point corresponding to the minimum estimated total time spent as the optimal detection point.
Preferably, the method further comprises:
acquiring preference factors of users;
after calculating the estimated total time consumption, the method is replaced by:
calculating the recommendation degree of different target detection points according to the preference factors of the user and the calculated estimated total consumption time;
and comparing the recommendation degree of the target detection points, and outputting the target detection point with the highest recommendation degree as the optimal detection point.
Preferably, obtaining the preference factor of the user includes:
and obtaining travel willingness coefficients of the user under the conditions of different terrains, different weather and different time periods by means of questionnaires or historical data analysis.
Preferably, the formula for calculating the recommendation degree of different target detection points is:
wherein m is a trip willingness coefficient under the current weather, n is a trip willingness coefficient of a time period corresponding to the appointed time, and k 1 、k 2 ……k n The travel willingness coefficients of different terrains are respectively d 1 、d 2 ……d n Respectively the path lengths of different terrains, d is the total path length between the designated position and the target detection point, t is the estimated total consumption time, r is the correction coefficient, r 1 、r 2 ……r n Respectively d/t are correspondingly belonging to different sets A 1 、A 2 ……A n Correction coefficients at that time.
Preferably, clustering the queuing data for each target detection point using dynamic time warping as a distance metric includes:
calculating intra-cluster and inter-cluster dynamic time warping distance ratios, including:
calculating the dynamic time regular distance of any two pieces of historical daily load data in the cluster, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance in the cluster;
calculating the dynamic time regular distance between any two center points between clusters, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance between clusters;
and obtaining the intra-cluster dynamic time warping distance ratio according to the intra-cluster dynamic time warping average distance and the inter-cluster dynamic time warping average distance.
Preferably, the method further comprises:
performing path planning on the unmanned aerial vehicle at a plurality of target detection points based on a genetic algorithm, including:
a1, adopting binary coding and real number coding, taking each individual as a path sequence of a group of detection points, taking each detection point as a chromosome, randomly initializing the path sequence in matlab by using an np.array function, shuffling by using a shuffle function, and determining each individual to realize population initialization;
a2, defining the fitness as the reciprocal of the distance, and calculating the fitness;
a3, adopting roulette selection, firstly determining the selected probability of each individual, dividing the fitness of each individual by the fitness of the total individual according to a formula, namely converting each probability into an accumulated probability when p=fi/Σfi is realized, randomly generating a number in [0,1], and selecting a corresponding individual according to a relative interval falling in the accumulated probability;
a4, under the real number coding, partial mapping crossover is adopted to define a random number of [0,1], and when the random number is larger than crossover probability cross-prob=0.9, partial chromosomes (a plurality of adjacent detection points) are crossed, and upper and lower individual conflict parts are exchanged to solve conflicts;
a5, adopting turnover mutation to define a [0,1] random number, and performing mutation when the random number is smaller than the mutation probability mut-prob=0.1;
repeating steps A2-A5 until the termination condition is satisfied: the fitness value of the optimal individuals in the population is continuous.
Preferably, in crowd counting the queuing images of the detection points, the neural network parameter obtaining method of the YOLOv5 algorithm includes:
preprocessing the input end of the neural network for the queuing image;
slicing the queuing image through a Focus structure of the backstone, and performing convolution operation on the sliced image to obtain a double downsampling characteristic diagram under the condition of no information loss;
carrying out convolution processing on the double downsampled feature map through a CBL convolution module, and then carrying out feature learning on the model by utilizing a CSP structure;
processing the feature tensor through the FPN+PAN structure of the Neck;
the Head layer predicts the image features, generates a bounding box and a prediction class, and outputs the convolutional neural network parameters of YOLOv 5.
Preferably, the preprocessing comprises Mosaic data enhancement, adaptive anchor frame calculation and adaptive picture scaling.
Preferably, the method further comprises: real-time feedback is carried out on the number of people in real time queuing, the time required by predicting queuing of each target detection point and the total predicted consumption time of each target detection point.
The specification also provides a queuing point recommendation system based on unmanned aerial vehicle aerial photography, comprising:
the image processing module is used for intercepting pictures captured by the unmanned aerial vehicle to obtain queuing images of all target detection points;
the counting module is used for counting the crowd of the queuing images of the detection points through a YOLOv5 algorithm to obtain the real-time queuing crowd of each target detection point;
the prediction module is used for clustering queuing data of each target detection point by using dynamic time regularity as distance measurement, inputting clustered historical queuing data into a Markov original model, training the original model to obtain a prediction model, calculating the time when a user arrives at the target detection point according to the user appointed time, the appointed position and the target detection point position, inputting the current queuing data into the Markov prediction model, predicting the queuing number when the user arrives at the target detection point, predicting the queuing required time, and calculating the predicted total consumption time according to the queuing required time and the time spent by the user to come and go to the target detection point;
and the recommendation module is used for comparing the estimated total time spent by the user for selecting different target detection points and outputting the shortest target detection point with the estimated total time spent as the optimal detection point.
The invention has the beneficial effects that: the YOLOv5 algorithm is a real-time target detection algorithm, and is used as a single-stage algorithm, compared with other algorithms, the single-stage algorithm has higher running speed and higher average detection precision, the queuing situation of target detection points at future time points is predicted through a Markov prediction model, the queuing number and the queuing speed after a period of time from a user departure point to the arrival of the target detection points through a certain travel mode can be predicted, the accuracy and the practicability of the example can be further improved, and in the case of statistics of queuing numbers with a plurality of target detection points nearby, optimal options can be selected for users to select according to the existing data and the historical data, so that the user can conveniently make travel planning.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only of the invention and that other drawings can be obtained from them without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a queuing point recommendation method according to an embodiment of the invention;
FIG. 2 is a flowchart of a queuing point recommendation method based on recommendation degree according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an unmanned aerial vehicle, an upper computer and a server system according to an embodiment of the invention.
Detailed Description
The present invention will be further described in detail with reference to specific embodiments in order to make the objects, technical solutions and advantages of the present invention more apparent.
It is to be noted that unless otherwise defined, technical or scientific terms used herein should be taken in a general sense as understood by one of ordinary skill in the art to which the present invention belongs. The terms "first," "second," and the like, as used herein, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.
As shown in fig. 1, an embodiment of the present disclosure provides a queuing point recommendation method based on unmanned aerial vehicle aerial photography, including the following steps:
intercepting a picture captured by an unmanned aerial vehicle to obtain queuing images of all target detection points;
counting the number of people in the queuing images of the detection points through a YOLOv5 algorithm to obtain the number of people in the queuing of each target detection point in real time;
for example, first the model of the YOLOv5 algorithm can be essentially divided into four parts: network architecture of input, backhaul, neg, output.
An input end: firstly, the picture is subjected to the processing (preprocessing) of a neural network input end: mosaics data enhancement, adaptive anchor frame calculation, adaptive picture scaling:
the practical use method is referred as follows:
mosained data enhancement: firstly, 1 picture and random three pictures of a transmitted data set are combined into 4 pictures, and the pictures are spliced in a random scaling, random cutting and random arrangement mode.
Self-adaptive anchor frame calculation: the preset frames are framed out of the targets at the possible positions and then adjusted based on the preset frames.
And calculating the position coordinates of the target and the height and width of the frame through the self-adaptive anchor frame. Adaptive picture scaling: and (5) adopting a letterbox self-adaptive picture scaling technology to scale and fill the picture to a specified size. Then, the CSP structure performs feature extraction on the preprocessed tensor through the Focus structure of the backbox, and a slicing operation is adopted to split a high-resolution picture (feature map) into a plurality of low-resolution pictures/feature maps, namely column-separated sampling and splicing.
Firstly, before a picture enters a backlight in v5, the Focus module performs slicing operation on the picture, namely, taking a value in every other pixel in one picture, similar to adjacent downsampling, so that four pictures are taken, the four pictures are complementary and almost long, but no information is lost, W, H information is concentrated into a channel space, an input channel is expanded by 4 times, namely, the spliced picture becomes 12 channels relative to the original RGB three-channel mode, and finally, the obtained new picture is subjected to convolution operation, so that a double downsampling characteristic diagram without information loss is finally obtained.
Then, CBL convolution module: CBL is a convolution block: consists of three network layers, conv, batch Normalization, leak ReLU.
The Conv layer is a convolution layer, and a plurality of different convolution kernels are adopted for processing the input image to obtain different response characteristic diagrams.
The BN (Batch Normalization) layer is a batch normalization layer, and the BN layer is used as a layer of the neural network after the convolutional layer is placed before the activation function. When the number of the obtained feature images is m and the size of the feature images is w×h (i.e. the number of the image pixels), the data quantity of the BN is m×w×h. The main operation step of the BN layer is to calculate the mean value and variance of all batch data, then to normalize the pixel value and the mean value by dividing the difference by the variance, and to add an offset factor and a scale change factor to control the normalized value, wherein the value of the factor is learned by a neural network in training.
The leak ReLU function is a variant of the ReLU function in which the learning speed of the ReLU may become very slow when the input is negative, even rendering neurons directly ineffective, since the input is now less than zero and the gradient is zero, so that its weight cannot be updated and will remain silent throughout the rest of the training. When the input value of ReLU is negative, the output is always 0, and the first derivative is always 0, which results in that the Neuron cannot update the parameter, i.e. the Neuron does not learn, and this phenomenon is called "Dead Neuron". To reduce the occurrence of silent neurons, allowing gradient-based learning (although slow) to address the disadvantage of the ReLU function, a leakage (leakage) value is introduced in the negative half-interval of the ReLU function, so called the leakage ReLU function, which outputs a small gradient to negative inputs. Because the derivative is always non-zero, the problem that neurons do not learn after the ReLU function enters the negative interval is solved.
The CSP structure divides the original input into two branches, the convolution operation is carried out to halve the number of channels, then one branch carries out the Bottleneck's operation, then the two branches are concat, so that the input and the output of the Bottleneck CSP are the same size, and the model can learn more characteristics.
And then, processing the feature tensor through the FPN+PAN structure of Neck, combining the FPN+PAN, conveying the strong semantic features from top to bottom by the FPN layer, conveying the strong positioning features from bottom to top by the feature pyramid containing the PAN structure, and carrying out parameter aggregation on different detection layers from different trunk layers.
The Head layer predicts the image features, generates a bounding box and predicts the category
YOLOv5 employs giou_loss in the class Loss function. IOU_Loss: the overlapping area of the detection frame and the target frame is mainly considered.
GIOU_Loss: on the basis of the IOU, the problem that the bounding boxes are not coincident is solved. And NMS non-maximum suppression is employed. The GIoU focuses not only on the overlapping area, but also on other non-overlapping areas, so that the overlapping ratio of the two areas can be better reflected, and the prediction frame and the real frame can be continuously close by minimizing the GIOU_Loss.
Input: two arbitrary convex A, B epsilon S n ，
And (3) outputting: GIoU, find the smallest convex C in S space containing a, B,
giou_loss function:
IoU function:
the queuing data of each target detection point is clustered by using dynamic time warping as a distance measure, and the similarity of the queuing data is compared.
For example, calculating intra-cluster and inter-cluster dynamic time warping distance ratios includes:
calculating the dynamic time regular distance of any two pieces of historical data in the cluster, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance in the cluster;
calculating the dynamic time regular distance between any two center points between clusters, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance between clusters;
and obtaining the intra-cluster dynamic time warping distance ratio according to the intra-cluster dynamic time warping average distance and the inter-cluster dynamic time warping average distance.
Inputting clustered historical queuing data into a Markov original model, and training the original model to obtain a prediction model M1; the construction of the prediction model of M1 mainly has the following characteristics:
the unmanned aerial vehicle imaging system is used for acquiring and outputting a plurality of historical queuing data;
a clustering module for clustering the historical daily load data using dynamic time warping as a distance measure;
the training module is used for inputting the clustered historical data into a Markov chain original model, training the Markov chain original model and obtaining a Markov chain prediction model;
and the prediction module is used for inputting the current daily load data into the Markov chain prediction model so as to predict the data of the next stage.
The time when the user arrives at the target detection point is calculated according to the user specified time, the specified position and the target detection point position, for example, the user specified time can be the current time or a future time point, and the specified position can be the current position or the position selected by the user.
Inputting current queuing data into a Markov prediction model, predicting the number of queuing people when a user reaches a target detection point, and predicting the time required by queuing;
calculating the estimated total consumption time according to the time required for queuing and the time spent for the user to go to and from the target detection point;
comparing the estimated total time spent by the user for selecting different target detection points, and outputting the target detection point at the left end of the estimated total time spent as an optimal detection point.
According to the queuing point recommending method based on unmanned aerial vehicle aerial photography, the YOLOv5 algorithm is used as a real-time target detection algorithm, and compared with other algorithms, the single-stage algorithm is faster in running speed and higher in average detection precision, the queuing situation of the target detection point at the future time point is predicted through the Markov prediction model, the queuing number and the queuing speed from the starting point of a user to a period of time taken for reaching the target detection point through a certain travel mode can be predicted, the accuracy and the practicability of the example can be improved, and in the case of statistics of the queuing number with a plurality of target detection points nearby, optimal options can be selected for users to select according to existing data and historical data, and a user can conveniently conduct travel planning.
As an embodiment, the method further comprises:
acquiring preference factors of users;
after calculating the estimated total time consumption, the method is replaced by:
calculating the recommendation degree of different target detection points according to the preference factors of the user and the calculated estimated total consumption time;
and comparing the recommendation degree of the target detection points, and outputting the target detection point with the highest recommendation degree as the optimal detection point.
The obtaining of the preference factors of the user comprises the following steps:
and obtaining travel willingness coefficients of users under the conditions of different terrains, different weather and different time periods by means of questionnaire investigation or historical data analysis, wherein the formula for calculating the recommendation degree of different target detection points is as follows:
wherein m is a trip willingness coefficient under the current weather, n is a trip willingness coefficient of a time period corresponding to the appointed time, and k 1 、k 2 ……k n The travel willingness coefficients of different terrains are respectively d 1 、d 2 ……d n Respectively the path lengths of different terrains, d is the total path length between the designated position and the target detection point, t is the estimated total consumption time, r is the correction coefficient, r 1 、r 2 ……r n Respectively d/t are correspondingly belonging to different sets A 1 、A 2 ……A n Correction coefficients at that time.
In the method, when a user selects a target queuing point, the time is generally not only taken as the only consideration factor, when the time difference between two or more queuing points is not large, the factors such as weather conditions and road conditions of the day also influence the selection of the user, for example, when weather conditions and road conditions are good, the user may prefer to reach a queuing point which is far away and has little time consumption, for example, when the weather is bad, a muddy road exists in a journey, the user can take more time, but the queuing point with a short distance can influence the judgment of the user even before work, after work, on weekends and in evening, and travel coefficients of the user under the conditions of different terrains, different weather conditions and different time periods are comprehensively considered, and a formula for calculating the recommendation degree of different target detection points is designed.
As an embodiment, the method further includes planning a path of the unmanned aerial vehicle at a plurality of target detection points based on a genetic algorithm, including:
a1, firstly, encoding is performed.
For TSP, binary coding and real coding are generally used, real coding is more convenient (the later variation and exchange are realized), each individual is the path sequence of a group of detection points, each detection point is a chromosome, the path sequence is randomly initialized by using an np.array function in matlab, each individual is determined by using a shuffle function, and thus population initialization is realized.
A2, then, calculate the fitness
The higher the adaptation to the environment, the easier it is to make a natural choice, whereas for TSP the total path is the shortest, which is defined for convenience as the reciprocal of the distance, fit=1.distance
A3, selecting.
The item adopts roulette selection, firstly, the selected probability of each individual is determined, the formula is the fitness of each individual divided by the fitness of the total individual, each probability is converted into the accumulated probability when p=fi/Σfi is realized, then numbers are randomly generated in [0,1], and the numbers are seen in the relative interval of the accumulated probabilities, so that the corresponding individual is selected.
A4, then, crossing.
Under real number coding, a random number of [0,1] is defined by using partial mapping crossover, and when the random number is larger than crossover probability cross-prob=0.9, that is, partial chromosomes (a plurality of adjacent detection points) are crossed, but in TSP, the crossover of detection point paths can generate conflict, and repeated staining individuals can possibly occur. The upper and lower individual conflict parts exchange to resolve conflicts.
A5, finally, mutating.
A random number of [0,1] is defined by inversion mutation, and mutation is performed when the random number is smaller than the mutation probability mut-prob=0.1.
A6, repeating the steps A2-A5 until the termination condition is met.
Termination condition: there are generally three types: the maximum iteration times, the reaching precision and the fitness value of the optimal individual in the population are reached continuously, and the project adopts the third type, namely the iteration times are adjusted until the finally obtained fitness value is unchanged.
By adopting the track planning of the genetic algorithm, the efficiency of the unmanned aerial vehicle in image acquisition of a plurality of target detection points can be improved, energy sources are saved, and the endurance of the unmanned aerial vehicle is improved.
The embodiment of the specification also provides a queuing point recommendation system based on unmanned aerial vehicle aerial photography, which comprises the following steps:
the image processing module is used for intercepting pictures captured by the unmanned aerial vehicle to obtain queuing images of all target detection points;
the counting module is used for counting the crowd of the queuing images of the detection points through a YOLOv5 algorithm to obtain the real-time queuing crowd of each target detection point;
the prediction module is used for clustering queuing data of each target detection point by using dynamic time regularity as distance measurement, inputting clustered historical queuing data into a Markov original model, training the original model to obtain a prediction model, calculating the time when a user arrives at the target detection point according to the user appointed time, the appointed position and the target detection point position, inputting the current queuing data into the Markov prediction model, predicting the queuing number when the user arrives at the target detection point, predicting the queuing required time, and calculating the predicted total consumption time according to the queuing required time and the time spent by the user to come and go to the target detection point;
and the recommendation module is used for comparing the expected total time consumption of the user for selecting different target detection points and outputting the target detection point at the left end of the expected total time consumption as an optimal detection point.
Furthermore, the system can also be responsible for supervising the unmanned aerial vehicle and synchronizing real-time data by arranging an upper computer and a server, as shown in fig. 3.
Those of ordinary skill in the art will appreciate that: the discussion of any of the embodiments above is merely exemplary and is not intended to suggest that the scope of the invention (including the claims) is limited to these examples; the technical features of the above embodiments or in the different embodiments may also be combined within the idea of the invention, the steps may be implemented in any order and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.
The present invention is intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omission, modification, equivalent replacement, improvement, etc. of the present invention should be included in the scope of the present invention.
Claims (7)
1. A queuing point recommending method based on unmanned aerial vehicle aerial photography is characterized by comprising the following steps:
intercepting a picture captured by an unmanned aerial vehicle to obtain queuing images of all target detection points;
counting the number of people in the queuing images of the detection points through a YOLOv5 algorithm to obtain the number of people in the queuing of each target detection point in real time;
clustering queuing data of each target detection point by using dynamic time warping as a distance measure;
inputting the clustered historical queuing data into a Markov original model, and training the original model to obtain a prediction model;
calculating the time when the user reaches the target detection point according to the user specified time, the specified position and the target detection point position;
inputting current queuing data into a Markov prediction model, predicting the number of queuing people when a user reaches a target detection point, and predicting the time required by queuing;
calculating the estimated total consumption time according to the time required for queuing and the time spent for the user to go to and from the target detection point;
acquiring preference factors of users;
calculating the recommendation degree of different target detection points according to the preference factors of the user and the calculated estimated total consumption time;
comparing the recommendation degree of the target detection points, and outputting the target detection point with the highest recommendation degree as an optimal detection point;
the obtaining the preference factor of the user comprises:
obtaining travel willingness coefficients of users under the conditions of different terrains, different weather and different time periods in a questionnaire or historical data analysis mode;
the formula for calculating the recommendation degree of different target detection points is as follows:
wherein m is a trip willingness coefficient under the current weather, n is a trip willingness coefficient of a time period corresponding to the appointed time, and k 1 、k 2 ……k n The travel willingness coefficients of different terrains are respectively d 1 、d 2 ……d n Respectively the path lengths of different terrains, d is the total path length between the designated position and the target detection point, t is the estimated total consumption time, r is the correction coefficient, r 1 、r 2 ……r n Respectively d/t are correspondingly belonging to different sets A 1 、A 2 ……A n Correction coefficients at that time.
2. The unmanned aerial vehicle-based queuing point recommendation method of claim 1, wherein clustering the queuing data for each target detection point using dynamic time warping as a distance metric comprises:
calculating intra-cluster and inter-cluster dynamic time warping distance ratios, including:
calculating the dynamic time regular distance of any two pieces of historical daily load data in the cluster, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance in the cluster;
calculating the dynamic time regular distance between any two center points between clusters, and averaging the dynamic time regular distances to obtain the dynamic time regular average distance between clusters;
and obtaining the intra-cluster dynamic time warping distance ratio according to the intra-cluster dynamic time warping average distance and the inter-cluster dynamic time warping average distance.
3. Queuing point recommendation method based on unmanned aerial vehicle aerial photography as claimed in claim 1, further comprising:
performing path planning on the unmanned aerial vehicle at a plurality of target detection points based on a genetic algorithm, including:
a1, adopting binary coding and real number coding, wherein each individual is a path sequence of a group of detection points, each detection point is a chromosome, randomly initializing the path sequence by using an np.array function in matlab, shuffling by using a shuffle function, and determining each individual to realize population initialization;
a2, defining the fitness as the reciprocal of the distance, and calculating the fitness;
a3, adopting roulette selection, firstly determining the selected probability of each individual, wherein the formula is the fitness of each individual divided by the fitness of the total individual, namelyWhen implemented, the probabilities are converted to cumulative probabilities, then at [0,1]The corresponding individuals are selected according to the relative interval falling within the accumulated probability;
a4, under the real number coding, partial mapping crossover is adopted to define a random number of [0,1], and when the random number is larger than crossover probability cross-prob=0.9, partial chromosomes are crossed, and upper and lower individual conflict parts are exchanged to solve conflicts;
a5, adopting turnover mutation to define a [0,1] random number, and performing mutation when the random number is smaller than the mutation probability mut-prob=0.1;
repeating steps A2-A5 until the termination condition is satisfied: the fitness value of the optimal individuals in the population is continuous.
4. The unmanned aerial vehicle-based queuing point recommendation method according to claim 1, wherein in the crowd counting of queuing images of detection points, the neural network parameter obtaining method of the YOLOv5 algorithm comprises:
preprocessing the input end of the neural network for the queuing image;
slicing the queuing image through a Focus structure of the backstone, and performing convolution operation on the sliced image to obtain a double downsampling characteristic diagram under the condition of no information loss;
carrying out convolution processing on the double downsampled feature map through a CBL convolution module, and then carrying out feature learning on the model by utilizing a CSP structure;
processing the feature tensor through the FPN+PAN structure of the Neck;
the Head layer predicts the image features, generates a bounding box and a prediction class, and outputs the convolutional neural network parameters of YOLOv 5.
5. The unmanned aerial vehicle-based queuing point recommendation method according to claim 4, wherein the preprocessing comprises mosaics data enhancement, adaptive anchor frame calculation and adaptive picture scaling.
6. Queuing point recommendation method based on unmanned aerial vehicle aerial photography as claimed in claim 1, further comprising: real-time feedback is carried out on the number of people in real time queuing, the time required by predicting queuing of each target detection point and the total predicted consumption time of each target detection point.
7. Queuing point recommendation system based on unmanned aerial vehicle is characterized by comprising:
the image processing module is used for intercepting pictures captured by the unmanned aerial vehicle to obtain queuing images of all target detection points;
the counting module is used for counting the crowd of the queuing images of the detection points through a YOLOv5 algorithm to obtain the real-time queuing crowd of each target detection point;
the prediction module is used for clustering queuing data of each target detection point by using dynamic time regularity as distance measurement, inputting clustered historical queuing data into a Markov original model, training the original model to obtain a prediction model, calculating the time when a user arrives at the target detection point according to the user appointed time, the appointed position and the target detection point position, inputting the current queuing data into the Markov prediction model, predicting the queuing number when the user arrives at the target detection point, predicting the queuing required time, and calculating the predicted total consumption time according to the queuing required time and the time spent by the user to come and go to the target detection point;
the recommendation module is used for acquiring preference factors of users, calculating recommendation degrees of different target detection points according to the preference factors of the users and the calculated expected total consumption time, comparing the recommendation degrees of the target detection points, and outputting the target detection point with the highest recommendation degree as an optimal detection point, wherein the acquisition of the preference factors of the users comprises the following steps:
obtaining travel willingness coefficients of users under the conditions of different terrains, different weather and different time periods in a questionnaire or historical data analysis mode;
the formula for calculating the recommendation degree of different target detection points is as follows:
wherein m is a trip willingness coefficient under the current weather, n is a trip willingness coefficient of a time period corresponding to the appointed time, and k 1 、k 2 ……k n The travel willingness coefficients of different terrains are respectively d 1 、d 2 ……d n Respectively the path lengths of different terrains, d is the total path length between the designated position and the target detection point, t is the estimated total consumption time, r is the correction coefficient, r 1 、r 2 ……r n Respectively d/t corresponds toBelonging to different sets A 1 、A 2 ……A n Correction coefficients at that time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211124537.XA CN115470418B (en) | 2022-09-15 | 2022-09-15 | Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211124537.XA CN115470418B (en) | 2022-09-15 | 2022-09-15 | Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115470418A CN115470418A (en) | 2022-12-13 |
| CN115470418B true CN115470418B (en) | 2023-07-25 |
Family
ID=84333885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211124537.XA Active CN115470418B (en) | 2022-09-15 | 2022-09-15 | Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115470418B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116483210B (en) * | 2023-06-25 | 2023-09-08 | 安徽大学 | Brain-controlled unmanned aerial vehicle method and system based on deep learning and sliding mode control |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108629323A (en) * | 2018-05-11 | 2018-10-09 | 哈尔滨工业大学 | A kind of integrated providing method of scenic spot tourist chain type trip |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108288498A (en) * | 2018-01-15 | 2018-07-17 | 浙江大学 | A kind of hospital intelligent attendance system based on machine vision |
| CN111160617B (en) * | 2019-12-06 | 2022-11-18 | 北京国电通网络技术有限公司 | Power daily load prediction method and device |
| CN111160631B (en) * | 2019-12-17 | 2024-01-19 | 西北工业大学 | Conflict detection and resolution method based on four-dimensional track operation |
| CN112561142B (en) * | 2020-12-04 | 2022-07-01 | 电子科技大学成都学院 | Queuing information inquiry system |
| CN114399552B (en) * | 2022-03-24 | 2022-06-14 | 武汉视合远方科技有限公司 | Indoor monitoring environment behavior identification and positioning method |
| CN114741618A (en) * | 2022-04-19 | 2022-07-12 | 中国工商银行股份有限公司 | Offline service point recommendation method, recommendation device, equipment and medium |
-
2022
- 2022-09-15 CN CN202211124537.XA patent/CN115470418B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108629323A (en) * | 2018-05-11 | 2018-10-09 | 哈尔滨工业大学 | A kind of integrated providing method of scenic spot tourist chain type trip |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115470418A (en) | 2022-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106683048B (en) | Image super-resolution method and device | |
| CN110569875B (en) | Deep neural network target detection method based on feature multiplexing | |
| CN111144329A (en) | Light-weight rapid crowd counting method based on multiple labels | |
| CN111815665B (en) | Single image crowd counting method based on depth information and scale perception information | |
| WO2021218671A1 (en) | Target tracking method and device, and storage medium and computer program | |
| CN114898284B (en) | Crowd counting method based on feature pyramid local difference attention mechanism | |
| CN115470418B (en) | Queuing point recommendation method and system based on unmanned aerial vehicle aerial photography | |
| CN114360067A (en) | Dynamic gesture recognition method based on deep learning | |
| CN112288776B (en) | Target tracking method based on multi-time step pyramid codec | |
| CN111414931A (en) | Multi-branch multi-scale small target detection method based on image depth | |
| CN111104831B (en) | Visual tracking method, device, computer equipment and medium | |
| CN111401113A (en) | Pedestrian re-identification method based on human body posture estimation | |
| CN110991317A (en) | Crowd counting method based on multi-scale perspective sensing type network | |
| CN110909625A (en) | Computer vision basic network training, identifying and constructing method and device | |
| CN115359366A (en) | Remote sensing image target detection method based on parameter optimization | |
| CN115187786A (en) | Rotation-based CenterNet2 target detection method | |
| CN115810149A (en) | High-resolution remote sensing image building extraction method based on superpixel and image convolution | |
| CN115544239A (en) | Deep learning model-based layout preference prediction method | |
| CN114373104A (en) | Three-dimensional point cloud semantic segmentation method and system based on dynamic aggregation | |
| CN110070023A (en) | A kind of self-supervisory learning method and device based on sequence of motion recurrence | |
| WO2022029847A1 (en) | Turbulence prediction system and turbulence prediction method | |
| CN113989612A (en) | Remote sensing image target detection method based on attention and generation countermeasure network | |
| CN110991230B (en) | Method and system for detecting ship by using remote sensing images in any direction based on rotation candidate frame | |
| KR102014288B1 (en) | Development pressure prediction method based on artificial intelligence using drone | |
| CN117131991A (en) | Urban rainfall prediction method and platform based on hybrid neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |