CN116092037B - Vehicle type identification method integrating track space-semantic features - Google Patents
Vehicle type identification method integrating track space-semantic features Download PDFInfo
- Publication number
- CN116092037B CN116092037B CN202310171830.XA CN202310171830A CN116092037B CN 116092037 B CN116092037 B CN 116092037B CN 202310171830 A CN202310171830 A CN 202310171830A CN 116092037 B CN116092037 B CN 116092037B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- micro
- track
- point
- speed
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a vehicle type identification method integrating track space-semantic features, which is used for preprocessing original vehicle track data, and comprises repeated track filtering, stay point identification and micro-stroke division; calculating vehicle motion characteristics and associated geographic semantic characteristics based on the vehicle track data, the stop points and the micro-travel; and identifying the vehicle type based on the vehicle motion characteristics and the associated geographic semantic characteristics. The specific operation of the stay point identification is as follows: and selecting track points with the moving distance smaller than a distance threshold and the theoretical speed smaller than a speed threshold from certain vehicle track points to perform DBSCAN clustering, wherein the clustering is constrained by using the space distance and the speed difference as double neighborhoods, and each cluster after clustering corresponds to one stop point of the vehicle respectively. The micro-travel division is to break the vehicle track when the time interval of the adjacent track points is larger than the time threshold value second or meets the stop point, and then the micro-travel of each vehicle is obtained. And the vehicle type recognition precision is effectively improved.
Description
Technical Field
The invention belongs to the field of vehicle type identification, and relates to a vehicle type identification method integrating track space-semantic features.
Background
In recent years, along with continuous maturation and popularization of the Internet of things, a sensing network and a precise positioning technology, a large amount of rich movement track data is accumulated. The track data contains rich and valuable resident trip activities and urban traffic running information, and can provide important decision basis for ecological livability environment construction and healthy social management. On the other hand, the method is beneficial to accurately estimating road traffic load, carbon emission of motor vehicles and the like, and is also beneficial to formulating personalized vehicle navigation guidance and realizing intelligent city traffic control.
Currently, vehicle type identification methods are largely divided into two main categories, invasive and non-invasive. The traditional invasive method mainly relies on a fixed position sensor (such as an induction coil, a dynamic weighing system and the like) to directly record the type of the vehicle, and the method has the advantages of accurate identification result and high efficiency, but has high operation and maintenance cost, is easy to interfere with traffic, and is difficult to cover all road sections. The non-invasive method mainly utilizes sensor equipment such as radar, infrared, acoustic, visual cameras and the like to identify the type of the running vehicle in a non-contact manner, has mature technology and accurate result without interfering with traffic, but also has the problems of higher operation and maintenance cost, difficult large-scale popularization and the like, and has certain limitation in severe weather conditions. In recent years, under the crowded sensing mode that people are sensors, continuously accumulated vehicle track data provides continuous dynamic traffic monitoring data with wide coverage and lower acquisition cost, so that the problems of high operation and maintenance cost, sensitivity to environment, easiness in generating traffic interference and the like of the existing method are overcome to a certain extent, and the method becomes an important means for vehicle type identification, traffic mode classification and driving behavior analysis.
The method is characterized in that the track data is utilized to extract the vehicle types, the principle is similar to that of traffic mode detection, the running characteristics such as speed, acceleration and deceleration are calculated according to the track time-space information, and then the support vector machine, random forest, deep learning and other classical classification models are utilized to identify different vehicle types. For example SUN Z, BAN X, vehicle classification using GPS data [ J]Transportation Research Part C A Emerging Technologies, volume 37,2013, pages102-117 calculated the standard deviation of the acceleration and deceleration and the acceleration and deceleration were greater than 1m/s 2 Characteristic indexes such as track ratio of the (E) and the like, combined with a support vector machine (Support Vector Machine, SVM)The two classes of the passenger car and the truck are realized by the secondary kernel function; on the basis of this SUN Z, BAN X.identification multiclass vehicles using global positioning system data [ J]Journal of Intelligent Transportation Systems,2018,22 (1): 1-9 propose the use of additive and subtractive interval accumulation frequencies and improved SVM classification to achieve three classifications of passenger cars, minivans and large vans. Thereafter, SIMONCINI M, TACCARI L, SAMBOF, et al Vehicle classification from low-frequency GPS data with recurrent neural networks [ J]Transportation Research Part C A-Emerging Technologies, volume 91,2018,Pages 176-191, proposes a vehicle classification method based on a long-short time memory (LSTM) recurrent neural network (Recurrent Neural Network, RNN) applied to light and heavy vehicle classification and light, medium and heavy vehicle classification respectively, the accuracy reaches 86.7% and 75.8%, and the accuracy is still to be improved compared with a fixed sensor.
Disclosure of Invention
The embodiment of the invention aims to provide a vehicle type identification method integrating track space and semantic features, which aims to solve the problem that the accuracy of the existing method for extracting the vehicle type by utilizing track data is to be improved.
The technical scheme adopted by the embodiment of the invention is as follows: the vehicle type identification method integrating the track space and the semantic features comprises the following steps:
s1, preprocessing original vehicle track data, including repeated track filtering, stay point identification and micro-program division;
step S2, calculating vehicle motion characteristics and associated geographic semantic characteristics based on vehicle track data, stay points and micro-travel;
and step S3, identifying the vehicle type based on the vehicle motion characteristics and the associated geographic semantic characteristics.
Further, the specific operation of the stay point identification in step S1 is as follows: and selecting track points with the moving distance smaller than a distance threshold and the theoretical speed smaller than a speed threshold from certain vehicle track points to perform DBSCAN clustering, wherein the clustering is constrained by using the space distance and the speed difference as double neighborhoods, and each cluster after clustering corresponds to one stop point of the vehicle respectively.
Further, a spatial distance constraint threshold Eps of DBSCAN cluster s =500 m, speed difference threshold Eps v =3 m/s, minimum point threshold minpts=10.
Further, the micro-travel division in the step S1 is to break the vehicle track when the time interval between adjacent track points is greater than the time threshold or encounters a stop point, so as to obtain the micro-travel of each vehicle.
Further, the vehicle motion characteristics in step S2 include a speed standard deviation, an acceleration standard deviation, a proportion of a track point of each speed section, a proportion of a track point of each acceleration section, a driving distance and a stopping rate of each micro-stroke.
Further, the calculation formula of the proportion of the track point of each acceleration section is as follows:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is in absolute acceleration interval AR k P is the micro-travel TS (i) Is a trace point>Is a micro-stroke TS (i) Theoretical absolute acceleration of the middle track point p, count { } is the number of elements of the corresponding set;
the calculation formula of the proportion of the track points of each speed interval is as follows:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is in the speed interval VR k Track point proportion of->Is a micro-stroke TS (i) The theoretical speed of the middle trace point p, count { }, is the number of elements of the corresponding set.
Further, the calculation formula of the travel distance of each micro-travel is as follows:
wherein d i Is a micro-stroke TS (i) The distance between two adjacent points D j (i) Is a micro-stroke TS (i) Is a travel distance of (2);
the calculation formula of the stopping rate of each section of micro-travel is as follows:
wherein p is a micro-stroke TS (i) Is arranged at the position of a certain track point,is a micro-stroke TS (i) The theoretical speed of the middle trace point p, count { } is the number of elements of the corresponding set, +.>Is a micro-stroke TS (i) Track point with medium speed of 0.
Further, the vehicle movement characteristics in step S2 include vehicle route road characteristics of each micro-trip, and the vehicle route road characteristics have the following calculation formula:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is matched to the r j Track point ratio on level road, r j For road class, j=1-5, respectively corresponding to expressway, national road, provincial road, county road and village road; r is R p Road class matched with the track point p is counted { } which is the number of corresponding set elements.
Further, the vehicle movement characteristics of step S2 include a vehicle stay place characteristic, and the vehicle stay place characteristic calculation formula is as follows:
in the method, in the process of the invention,is a vehicle V (i) All dwell points SS of (2) (i) S of the middle j The proportion of the stay points with highest class interest point proportion, s j J=1 to 5 for the type of the interest point, and corresponds to a bus stop, a service area, a bus stop, a railway station and others respectively; q is vehicle V (i) POI (point of stay) q The type of interest point with the highest duty ratio in the range of 500 meters of the stay point q.
Further, in step S3, based on the vehicle motion feature and the associated geographic semantic feature, a multi-core support vector machine and a probabilistic neural network grading random forest are adopted to perform vehicle type identification.
The embodiment of the invention has the beneficial effects that:
(1) According to the embodiment of the invention, the DBSCAN clustering method is utilized to effectively identify the vehicle stay point information, and the vehicle track is subjected to travel segmentation according to the vehicle stay point information, so that the stability of the travel track movement characteristics is ensured, and the separability of different types of vehicles is improved.
(2) Besides the motion characteristics of the vehicle, the embodiment of the invention extracts the geographic semantic characteristics of the road type, the stay place and the like of the vehicle approach by utilizing the navigation road map and the interest point data, and expands the characteristic set for identifying the vehicle type.
(3) The experiment adopts three classical machine learning methods of a multi-core support vector machine, a probability neural network and a random forest to carry out comprehensive verification, and results show that the random forest classification accuracy is highest, the average accuracy and recall rate reach more than 92%, and the method can be effectively applied to classification of buses, trucks and tractors.
(4) Comparing the fixed length stroke segmentation with the variable length stroke segmentation based on the stay point recognition, the latter having higher classification accuracy; compared with different feature combination modes, the vehicle classification precision of comprehensively considering the motion features, the road ways and the stay place information is obviously improved, the precision reaches 83% -93%, and the vehicle type recognition precision can be effectively improved by combining the vehicle motion features and the geographic semantic features.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a technical roadmap of a vehicle type recognition method that fuses trajectory space-semantic features.
Fig. 2 is a schematic diagram of identifying a stop point and dividing a micro-stroke on a driving track.
Fig. 3 is a graph of statistical correlation of average speed versus standard deviation of speed for passenger car, truck, and tractor travel tracks.
Fig. 4 is a graph showing statistical correlation between the speed variation coefficient and the maximum speed of the driving track of the passenger car, the freight car and the tractor.
Fig. 5 is a graph showing a traveling speed range of a passenger car, a truck, or a tractor.
Fig. 6 is a graph showing a traveling acceleration range profile of a passenger car, a truck, or a tractor.
Fig. 7 is a road class distribution diagram of a passenger car, a truck, and a tractor.
Fig. 8 is a plot of the residence point distribution for a passenger car, truck, tractor.
Fig. 9 is a basic structural diagram of a probabilistic neural network.
Fig. 10 is a schematic diagram of a random forest algorithm.
FIG. 11 is a graph of classification accuracy versus different classifiers with different combinations of features and stroke splitting patterns.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is obvious that the described embodiments are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
The embodiment provides a vehicle type identification method integrating track space and semantic features, as shown in fig. 1, comprising the following steps:
step S1, preprocessing original vehicle track data, including repeated track filtering, stay point identification and micro-travel division, so as to ensure reliable track travel and stable motion characteristics of vehicles involved in vehicle classification;
s2, extracting spatial semantic features based on vehicle track data, the identified stay points and the divided micro-strokes, and calculating vehicle motion features and associated geographic semantic features;
and S3, identifying the type of the vehicle by adopting a vehicle classification model.
The track data is a series of position data sets with time stamps for recording individual or group movement tracks, and is provided with vehicles V in the original track data set (i) Is TV (i) ＝{P 1 (i) …P j (i) …P N (i) }, wherein the trace point data P j (i) ＝(V (i) ,t j (i) ,p j (i) ，v j (i) ) Record vehicle V (i) At t j (i) Spatial position p of time j (i) And instantaneous velocity v j (i) 。
Based on the above observation information, the theoretical speed of each track point can be calculated firstAnd theoretical acceleration->The basic motion characteristics are as follows:
wherein d j (i) Is a vehicle V (i) Euclidean distance of the trajectory point j of (1) to the previous trajectory point j-1,is a vehicle V (i) Euclidean distance from track point j+1 to the previous track point j; />Is a vehicle V (i) Time corresponding to trace point j+1, +.>Is a vehicle V (i) Time corresponding to trace point j, +.>Is a vehicle V (i) The time corresponding to the trace point j-1; />Is a vehicle V (i) Theoretical speed of trajectory point j, +.>Is a vehicle V (i) Theoretical velocity of trajectory point j-1 of (c).
Because of uncertainty of vehicle track positioning accuracy, preprocessing is needed for original vehicle track data, and the method comprises three steps of repeated track filtering, stay point identification and micro-stroke division:
step S11, repeated track point filtering: deleting the repeated track points recorded by the same vehicle at the same moment, if the vehicle V (i) Is { p } h (i) …p j (i) ...p k (i) Meeting |t j (i) t j+1 (i) I=0, j=h … k-1, then the first trace point p is divided h (i) Besides, deleting other track points of the section, and filtering to obtain the TV (i) ＝{P 1 (i) …P h (i) ,P k+1 (i) …P N (i) }；
Step S12, identifying a stay point: selecting track points with a moving distance smaller than a set distance threshold and a theoretical speed smaller than a set speed threshold from certain vehicle track points to perform DBSCAN clustering, wherein the clustering uses space distance and speed difference as double neighborhood constraint, each cluster corresponds to one parking point of the vehicle after clustering, and other vehicle tracks identify the parking points according to the processes; the distance threshold and the speed threshold can be selected according to the distance between two stay points in the vehicle history track point data and the speed of the stay points and vehicles nearby, and the embodiment obtains the spatial distance constraint threshold Eps of the DBSCAN cluster with the distance threshold of 100m, the speed threshold of 10m/s and the DBSCAN through a large number of experiments and analysis s =500 m, speed difference threshold Eps v When the minimum point threshold value MinPts=10, the DBSCAN clustering judges that the accuracy of the stay point is highest;
step S13, micro-program division: when the time interval of the adjacent track points is larger than the set time threshold or meets the stopping point, the track of the vehicle is broken, and the micro-travel of each vehicle is obtained after breaking. The time threshold can be selected according to the residence time between two sections of micro-programs in the vehicle history track point data, and the residence time is compounded on the basis of the residence point judgment by considering the running speed and the distance in step S12, so that the accuracy of micro-program division is improved.
As shown in fig. 2, the horizontal axis is longitude and latitude, the vertical axis is a time axis, the red dot (stop point 1), the green dot (stop point 2) and the blue dot (stop point 3) are 3 stop points identified respectively, the black track section is a micro-stroke divided by the corresponding vehicle, it can be seen that the micro-stroke 1 and the micro-stroke 3 are two long-distance micro-strokes of round trip, and the micro-stroke 2 is a short-distance micro-stroke.
The driving behavior, the stay position, the path selection and the like of different types of vehicles are unique, and the embodiment explores the driving behavior and the vehicle type information contained in the track data from two aspects of vehicle motion characteristics and geographic semantic characteristics. After the micro-stroke is divided, the embodiment first calculates the statistical indexes such as the maximum speed, the minimum speed, the average speed, the standard deviation of the speed, the variation coefficient of the speed and the like of each section of micro-stroke respectively. Fig. 3 is a statistical correlation characteristic of average speed and standard deviation of speed of a driving track of a passenger car, a truck or a tractor, and fig. 4 is a statistical correlation characteristic of speed variation coefficient and maximum speed of the driving track of the passenger car, the truck or the tractor, wherein each scattered point represents one micro-travel of a corresponding type of vehicle. As can be seen from the figure, the scattered points of different types of vehicles are distributed in a staggered manner, and the vehicles are difficult to classify accurately only by using the speed statistical features.
Further analyzing the interval distribution condition of the speed and the acceleration in the micro-travel of each type of vehicle, namely respectively calculating the interval of each speed (in km/h) and the interval of absolute acceleration (in m/s) in the micro-travel of each type of vehicle 2 Unit), and the results are shown in fig. 5 to 6.
Assume an original vehicle travel track TV (i) Is divided into m sections of micro-travel TS (i) And n dwell points SS (i) Then is in absolute acceleration interval (in m/s 2 Unit) of the track pointsThe calculation formula of (2) is as follows:
in the method, in the process of the invention,calculated as micro-travel TS (i) Is in absolute acceleration interval AR k The ratio of the track points of (2) absolute acceleration interval AR k The absolute acceleration interval number can be divided according to the required precision, such as [0,0.1 ], [0.1,0.2 ], [0.2,0.3 ], [0.3,0.4 ], [0.4,0.5 ], [0.5, + ]; p is micro-travel TS (i) Is arranged at the position of a certain track point,is a micro-stroke TS (i) The theoretical absolute acceleration of the middle track point p is calculated by a formula (2); count { } is the corresponding set { p | (a) p ∈AR k &p∈TS (i) Number of elements.
The calculation formula of the proportion of the track points in the speed interval (in km/h) is as follows:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is in the speed interval VR k Track point proportion of->Is a micro-stroke TS (i) The theoretical speed of the middle track point p is calculated by a formula (1); count { } is the number of corresponding set elements. Speed interval VR k The number of speed intervals can be divided according to the required precision, such as [0,20 ], [20,40 ], [40,60 ], [60,80 ], [80,100 ], [100, ++ infinity A kind of electronic device.
Further calculating the driving distance of each micro-stroke:
wherein d i Is a micro-stroke TS (i) The distance between two adjacent points in the middle,is a micro-stroke TS (i) Is a traveling distance of the vehicle.
Further calculating the stopping rate of each section of micro-travel:
wherein p is a micro-stroke TS (i) Is arranged at the position of a certain track point,is a micro-stroke TS (i) The theoretical velocity of the middle trace point p, count { } is the corresponding set { p|v } p ＝0}、{p|p∈TS (i) Number of elements, +.>Is a micro-stroke TS (i) Track point with medium speed of 0.
As can be seen from fig. 5, the tractor travel speed is relatively high and is mainly focused on 60-80km/h due to the high speed of the tractor along the long distance route. And the running speed distribution of the passenger car and the freight car is relatively uniform and has no obvious difference. As can be seen from FIG. 6, the traction vehicle has a smaller running acceleration than the passenger car or truck, and an acceleration of 0-0.1m/s 2 Is obviously large in distribution proportionAnd is used in passenger car and truck. Therefore, vehicle type identification cannot be effectively performed by simply using vehicle motion characteristics, and because the vehicle running speed is sensitive to the running road type, traffic conditions and other environmental factors, and the running acceleration is closely related to the vehicle factors such as load, performance and the like, the geographic semantic characteristics such as the fusion road type, the stay place and the like are considered.
The embodiment utilizes the navigation road map and the interest point data to analyze and obtain geographic semantic features such as the road type of the vehicle approach and the stay place, and the like, and is used for assisting in vehicle type identification. Considering that a single micro-trip is difficult to accurately and completely represent typical travel activities of a vehicle, calculating the proportion of the micro-trips of the vehicle on different levels of roads and the proportion of the micro-trips of the vehicle on different types of interest points in all the stay points, and taking the proportion as the geographic semantic characteristics of all the micro-trips of the same vehicle.
First, a matching relation between each track point and the road is established based on the space distance, and then the road and the stay point characteristics of each vehicle approach are calculated.
The calculation formula of the road characteristics of the vehicle approach is as follows:
in RV j (i) Is a vehicle V (i) Matched to the r in the running micro-travel j Track point ratio on level road, r j For road class, j=1-5, respectively corresponding to expressway, national road, provincial road, county road and village road; r is R p Road class matched for trajectory point p, count { } is corresponding set { p|R p ∈r j &p∈TS (i) }、{p|p∈TS (i) Number of elements.
The vehicle stay point feature calculation formula is as follows:
wherein SV is j (i) Is a vehicle V (i) All dwell points SS of (2) (i) S of the middle j The proportion of the stay points with highest class interest point proportion, s j J=1 to 5 for the type of the interest point, and corresponds to a bus stop, a service area, a bus stop, a railway station and others respectively; q is vehicle V (i) POI (point of stay) q For the type of interest point (Point of Interest) with the highest proportion in the 500m range of the stay point q, count { } is the number of corresponding set elements.
As shown in fig. 7-8, there are significant differences in road characteristics and stopping points for different types of vehicle approaches. As shown in fig. 7, more than 60% of the tractor track points are on the expressway, more than 90% of the truck track points are on the non-expressway, and nearly 80% of the bus track points are on the provincial and lower roads. As shown in fig. 8, although the stop ratio of different types of vehicles is high at bus stops, the stop ratio of the buses at other stop points is different, for example, the stop ratio of the buses at bus stops is far higher than that of the other vehicles. In addition, considering that a single micro-trip is not representative, the present embodiment uses the vehicle route road characteristics and the vehicle stop location characteristics calculated by the same vehicle micro-trip and stop point as the geographical semantic characteristics of each micro-trip of the vehicle for vehicle classification.
Based on the motion features and the geographic semantic features, the vehicle type recognition is performed by using three classical supervised learning methods of a multi-core Support Vector Machine (SVM), a probabilistic neural network and a random forest.
As shown in formula (9), the multi-core SVM is a learning model that seeks a linear weighted combination based on a plurality of basic kernel functions as a synthetic kernel to obtain higher classification accuracy:
wherein X is i The motion characteristic matrix is input into the vehicle, each row of the motion characteristic matrix represents the motion characteristic of one micro-stroke, and the motion characteristic comprises the proportion of the track points of 6 speed intervals, the proportion of the track points of 6 acceleration intervals and each row of the motion characteristic matrixSpeed standard deviation, acceleration standard deviation, driving distance and stopping rate of the section micro-travel; x is X j For an input vehicle geosemantic feature matrix, vehicle geosemantic features include 5 vehicle route road features and 5 vehicle stop location features, K (X i ，X j ) Is a synthetic kernel function, k m (X i ，X j ) Is a basic kernel function, d m Is the corresponding basic kernel function k m (X i ，X j ) Is the parameter to be learned, M is the total number of basic kernel functions. In this embodiment, a linear kernel function, a polynomial kernel function, a radial kernel function, and a Sigmoid kernel function are selected as basic kernel functions, calculated vehicle motion features and geographic semantic features are mapped to optimal kernel functions, and the weight d of each kernel function is determined by training a classifier m And obtaining a vehicle classification result.
The basic structure of the probabilistic neural network is shown in fig. 9, and includes an input layer, an implicit layer, a summation layer, and an output layer. Input layer accepts sample data x= (X) 1 …X M ) And transferred to the hidden layer, the number of neurons of the input layer is equal to the length M of the input vector; the second hidden layer calculates the similarity between the input vector X and each training sample according to a formula (10), so that the number of neurons is consistent with the number of training samples; the third layer of summation layer calculates the posterior probability v of the input vector X belonging to various types according to the formula (11) i So the number of neurons is the same as the number of categories; the output layer receives the output of the summation layer and selects the neuron class with the greatest posterior probability as the final class of the input variable:
wherein X is ij For the j-th training sample in class i, delta is the smoothing factor to be set,representing an input vector X and training samples X ij Similarity between;
wherein L is i For the number of training samples corresponding to the ith neuron in the hidden layer, v i Is the output of the ith (class) neuron in the summation layer.
The main idea of random forest is as shown in fig. 10, firstly, a bootstrap self-help sampling method is utilized to randomly generate K training sets containing n samples; respectively constructing decision classification trees for the K randomly generated training sets to form a random forest; based on the generated random forest, K classification results of the test data are obtained, and then a final classification result is determined according to a voting method.
Example 2
And (3) carrying out comprehensive verification of space semantic feature analysis and vehicle type identification by using the driving track data of the 1958 vehicle acquired on the 10 th and 19 th of 2018. The experimental travel track data comprises 2359290 travel track points of passenger cars, trucks and tractors corresponding to 657, 665 and 636 vehicles. As shown in table 1, the raw trajectory data has an average sampling interval of 45 seconds and an average sampling interval of 126.63 meters. After the stay point identification and the micro-travel division, the passenger car, the truck and the tractor are respectively divided into 2413, 1127 and 1288 sections of micro-travel, and the number of stay points and the number of micro-travel are contained in a single vehicle running track shown in table 1.
Because the passenger car micro-travel data are relatively more, 1254 passenger car micro-travels are randomly selected for the classification experiment according to the embodiment of the invention, so that the data of different types of vehicles are ensured to be similar in scale and the classification fitting is avoided. And further calculating 16 motion characteristics such as speed and acceleration interval distribution of each micro-stroke, 10 geographical semantic characteristics such as vehicle path characteristics and stay place characteristics, and the like, wherein the total of 26 track characteristics are calculated, and in the embodiment, 80% of micro-strokes are used for training a model and 20% of micro-strokes are used for classification test.
Table 2 shows the confusion matrix of different vehicle classification methods using the trajectory space semantic features, the diagonal elements indicate the number of micro-strokes for correctly identifying the vehicle type, and the non-diagonal elements indicate the number of micro-strokes for incorrectly identifying the vehicle type. In general, the three methods have classification accuracy of 76.61% -93.80% and recall rate of 76.11% -93.42%, wherein the random forest classification accuracy is highest, the probability neural network is inferior, and the support vector machine classification accuracy is lowest. Compared with the classification precision of different types of vehicles, the classification precision of the passenger car is highest, the precision and recall rate are up to 93.80% and 93.42% respectively, the tractor is the lowest, and the misclassified vehicles are mainly concentrated between the tractor and the truck.
TABLE 1 Primary trajectory data, vehicle stop-and-go basic statistical characteristics
TABLE 2 SVM, probabilistic neural network and random forest based vehicle classification confusion matrix
Table 3 comparison of classification accuracy under different feature combinations and Stroke segmentation modes
Fig. 11 and table 3 show the accuracy statistics and the improvement amplitude of the three vehicle classification methods under different feature combinations and different travel division modes. On the one hand, through stay point identification and variable length micro-travel division, the vehicle classification precision is improved by 7.83% -17.18% compared with the fixed length segmentation. On the other hand, the geographical semantic features such as the approach roads and the stay places are considered, so that the vehicle classification precision is obviously improved compared with the classification precision based on the motion features alone, and the random forest method is improved by 25.07%. Finally, compared with different classification methods, the random forest classification method always keeps the highest precision although adopting different feature combinations and different stroke segmentation modes, and the probability neural network is secondly the lowest in classification precision of the support vector machine.
The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.
Claims (5)
1. The vehicle type identification method integrating the track space and the semantic features is characterized by comprising the following steps of:
s1, preprocessing original vehicle track data, including repeated track filtering, stay point identification and micro-program division;
step S2, calculating vehicle motion characteristics and associated geographic semantic characteristics based on vehicle track data, stay points and micro-travel;
step S3, identifying the type of the vehicle based on the motion characteristics of the vehicle and the associated geographic semantic characteristics;
the vehicle motion characteristics of the step S2 comprise the speed standard deviation, the acceleration standard deviation, the proportion of the track points of each speed interval, the proportion of the track points of each acceleration interval, the driving distance and the stopping rate of each micro-stroke;
the calculation formula of the proportion of the track points of each acceleration interval is as follows:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is in absolute acceleration interval AR k P is the micro-travel TS (i) Is a trace point>Is a micro-stroke TS (i) Theoretical absolute acceleration of the middle trace point p, count { } is the pairThe number of elements to be assembled;
the calculation formula of the proportion of the track points of each speed interval is as follows:
in the method, in the process of the invention,is a vehicle V (i) Is of micro-program TS (i) Is in the speed interval VR k Track point proportion of->Is a micro-stroke TS (i) The theoretical speed of the middle track point p, count { } is the number of elements of the corresponding set;
the calculation formula of the travel distance of each micro-travel is as follows:
wherein d i Is a micro-stroke TS (i) The distance between two adjacent points in the middle,is a micro-stroke TS (i) Is a travel distance of (2);
the calculation formula of the stopping rate of each section of micro-travel is as follows:
wherein p is a micro-stroke TS (i) Is arranged at the position of a certain track point,is a micro-stroke TS (i) Middle trackThe theoretical speed of point p, count { } is the number of elements of the corresponding set, +.>Is a micro-stroke TS (i) Track point with medium speed of 0 accounts for proportion;
the vehicle movement characteristics of the step S2 include vehicle route road characteristics of each micro-trip, and the vehicle route road characteristics have the following calculation formula:
in RV j (i) Is a vehicle V (i) Is of micro-program TS (i) Is matched to the r j Track point ratio on level road, r j For road class, j=1-5, respectively corresponding to expressway, national road, provincial road, county road and village road; r is R p Road grade matched with the track point p, count { } is the number of corresponding set elements;
the vehicle motion characteristics of step S2 include a vehicle stay place characteristic, and the vehicle stay place characteristic calculation formula is as follows:
in the method, in the process of the invention,is a vehicle V (i) All dwell points SS of (2) (i) S of the middle j The proportion of the stay points with highest class interest point proportion, s j J=1 to 5 for the type of the interest point, and corresponds to a bus stop, a service area, a bus stop, a railway station and others respectively; q is vehicle V (i) POI (point of stay) q For the interest point type with the highest proportion in the 500m range of the stay point q, count { } is the number of corresponding set elements.
2. The vehicle type identification method of merging trajectory space-semantic features according to claim 1, wherein the specific operation of the stay point identification of step S1 is: and selecting track points with the moving distance smaller than a distance threshold and the theoretical speed smaller than a speed threshold from certain vehicle track points to perform DBSCAN clustering, wherein the clustering is constrained by using the space distance and the speed difference as double neighborhoods, and each cluster after clustering corresponds to one stop point of the vehicle respectively.
3. The method for identifying a vehicle type with fusion of trajectory space-semantic features according to claim 2, characterized in that the spatial distance constraint threshold Eps of DBSCAN clusters s =500 m, speed difference threshold Eps v =3 m/s, minimum point threshold minpts=10.
4. The method for identifying the vehicle type by fusing track space and semantic features according to claim 1, wherein the micro-travel division in the step S1 is to break the track of the vehicle when the time interval between adjacent track points is greater than a time threshold or a stop point is encountered, so as to obtain the micro-travel of each vehicle.
5. The vehicle type recognition method based on the fusion track space-semantic features according to any one of claims 1 to 4, wherein in step S3, the vehicle type recognition is performed by using a multi-core support vector machine and a probabilistic neural network hierarchical random forest based on the vehicle motion features and the associated geographic semantic features.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310171830.XA CN116092037B (en) | 2023-02-13 | 2023-02-13 | Vehicle type identification method integrating track space-semantic features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310171830.XA CN116092037B (en) | 2023-02-13 | 2023-02-13 | Vehicle type identification method integrating track space-semantic features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116092037A CN116092037A (en) | 2023-05-09 |
| CN116092037B true CN116092037B (en) | 2023-07-28 |
Family
ID=86204472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310171830.XA Active CN116092037B (en) | 2023-02-13 | 2023-02-13 | Vehicle type identification method integrating track space-semantic features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116092037B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116561604B (en) * | 2023-07-10 | 2023-09-29 | 北京千方科技股份有限公司 | Truck identification method and device, electronic equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340427A (en) * | 2020-03-02 | 2020-06-26 | 西南交通大学 | Freight car running state identification method based on track data |
| CN113887659A (en) * | 2021-10-22 | 2022-01-04 | 沈阳世纪高通科技有限公司 | Vehicle type classification algorithm based on naive Bayes |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107463940B (en) * | 2017-06-29 | 2020-02-21 | 清华大学 | Vehicle type identification method and device based on mobile phone data |
| CN109686085B (en) * | 2018-12-17 | 2020-05-05 | 北京交通大学 | GPS data based dangerous cargo transport vehicle stop node activity type identification method |
| EP3680858A1 (en) * | 2019-01-11 | 2020-07-15 | Tata Consultancy Services Limited | Dynamic multi-camera tracking of moving objects in motion streams |
| CN110570658B (en) * | 2019-10-23 | 2022-02-01 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
| CN114117261B (en) * | 2022-01-29 | 2022-04-22 | 腾讯科技（深圳）有限公司 | Track detection method and device, electronic equipment and storage medium |
| CN115330521A (en) * | 2022-08-17 | 2022-11-11 | 奇瑞徽银汽车金融股份有限公司 | Vehicle loan overdue risk assessment method and system based on GPS track data |
| CN115436972A (en) * | 2022-08-31 | 2022-12-06 | 湖南省交通科学研究院有限公司 | Method, system and medium for identifying stop point based on heavy-duty truck GPS track |
| CN115619103B (en) * | 2022-11-15 | 2023-03-31 | 湖南省交通科学研究院有限公司 | Typical industry logistics combined transportation analysis method and system based on truck driving track |
-
2023
- 2023-02-13 CN CN202310171830.XA patent/CN116092037B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340427A (en) * | 2020-03-02 | 2020-06-26 | 西南交通大学 | Freight car running state identification method based on track data |
| CN113887659A (en) * | 2021-10-22 | 2022-01-04 | 沈阳世纪高通科技有限公司 | Vehicle type classification algorithm based on naive Bayes |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116092037A (en) | 2023-05-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107742193B (en) | A kind of driving Risk Forecast Method based on time-varying state transition probability Markov chain | |
| CN105825310A (en) | Taxi passenger-searching path recommendation method based on information entropy | |
| CN108062857B (en) | Prediction technique for cab-getter's trip purpose | |
| CN106846806A (en) | Urban highway traffic method for detecting abnormality based on Isolation Forest | |
| Kumar et al. | A model based approach to predict stream travel time using public transit as probes | |
| CN105869402B (en) | Express highway section speed modification method based on polymorphic type floating car data | |
| CN116092037B (en) | Vehicle type identification method integrating track space-semantic features | |
| Xue et al. | A context-aware framework for risky driving behavior evaluation based on trajectory data | |
| Wang et al. | Vehicle reidentification with self-adaptive time windows for real-time travel time estimation | |
| Moosavi et al. | Driving style representation in convolutional recurrent neural network model of driver identification | |
| Wang et al. | ARIMA Model and Few‐Shot Learning for Vehicle Speed Time Series Analysis and Prediction | |
| Wu et al. | Clustering of several typical behavioral characteristics of commercial vehicle drivers based on GPS data mining: Case study of highways in China | |
| Yu et al. | A feature-oriented vehicle trajectory data processing scheme for data mining: A case study for Statewide truck parking behaviors | |
| CN117273964B (en) | Intelligent vehicle insurance policy generation system and method for self-adaptive driving data | |
| CN116341706A (en) | Urban mobile load probability prediction system and method based on comprehensive energy perception | |
| CN114091581A (en) | Vehicle operation behavior type identification method based on sparse track | |
| Paredes et al. | Intelligent collision risk detection in medium-sized cities of developing countries, using naturalistic driving: A review | |
| Scora et al. | Real-time roadway emissions estimation using visual traffic measurements | |
| CN117475624A (en) | Mileage pile layout method for monitoring traffic state of basic road section of expressway | |
| Chen et al. | Detecting and analyzing unlicensed taxis: A case study of Chongqing City | |
| CN106157657A (en) | A kind of moving state identification system and method for mobile subscriber | |
| Van Hinsbergh et al. | Vehicle point of interest detection using in-car data | |
| Muñoz-Organero et al. | Detecting different road infrastructural elements based on the stochastic characterization of speed patterns | |
| CN114495519B (en) | Intelligent visual display system for expressway | |
| Raslan et al. | IoT for measuring road network quality index |
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 |