CN109918977A - Determine the method, device and equipment of free time parking stall - Google Patents

Abstract

This application discloses a kind of methods of determining idle parking stall, belong to image identification technical field.This method comprises: obtaining the vehicle image for containing at least two vehicle, determine that the two neighboring respective two dimension of vehicle surrounds frame at least two vehicles according to vehicle image；The location information that frame determines at least one feature locations point on the external frame in the respective bottom of two neighboring vehicle is surrounded according to the respective two dimension of two neighboring vehicle；According to the location information of at least one feature locations point on the external frame in the bottom of two neighboring vehicle, determine the idle parking stall between two neighboring vehicle, it only needs to have adjacent two vehicle that can recognize that two dimension surrounds frame in vehicle image, the idle parking stall between the two vehicles can be determined, to improve the recognition effect of idle parking stall.

Description

Determine the method, device and equipment of free time parking stall

Technical field

This application involves image identification technical field, in particular to a kind of method, device and equipment of determining idle parking stall.

Background technique

Automatic parking is a function important in automatic Pilot field, directly reflects the degree of intelligence of vehicle.Automatically The realization for function of parking needs vehicle that can automatically determine the position where idle parking stall.

In the related art, the automatic identification of idle parking stall is mainly realized by identification parking stall line.Specifically, vehicle The image of the image collection assembly acquisition vehicle periphery of middle setting, parking stall identifies equipment to image collection assembly acquired image It is identified, to determine the parking stall line in image, and according to the position of the determining idle parking stall of the parking stall line identified.

However in practical applications, the type of parking stall line is varied, there is printing lines, embedding careless brick, cement brick etc., and Scheme in the related technology can only realize the parking stall line for identifying color clear and clearly printing lines, for other types of Parking stall line can not then identify that the effect for causing parking stall to identify is poor.

Summary of the invention

In order to improve the recognition effect of parking stall, embodiments herein provides a kind of method of determining idle parking stall, dress It sets and equipment.

In a first aspect, providing a kind of method of determining idle parking stall, which comprises

Obtain the vehicle image for containing at least two vehicle；It is determined at least two vehicle according to the vehicle image The two neighboring respective two dimension of vehicle surrounds frame, and it is the boundary rectangle that vehicle is corresponded in the vehicle image that the two dimension, which surrounds frame, Frame；Frame is surrounded according to the two neighboring respective two dimension of vehicle, determines the external frame in the respective bottom of two neighboring vehicle On at least one feature locations point location information, the external frame in bottom is the orthographic projection of corresponding vehicle on the ground Boundary rectangle frame, the feature locations point are used to indicate the point on the predeterminated position of the corresponding external frame in bottom；According to the phase The location information of at least one feature locations point on the external frame in bottom of adjacent two vehicles, determine the two neighboring vehicle it Between idle parking stall.

In this scenario, by determining that the two dimension of the two neighboring vehicle in vehicle image surrounds frame, and according to adjacent two The two dimension of a vehicle surrounds the position that frame determines at least one feature locations point in the external frame in the respective bottom of two neighboring vehicle Confidence breath, and then determine the idle parking stall between the two neighboring vehicle, it is only necessary to existing in vehicle image can recognize that bottom Adjacent two vehicle of the external frame in portion, it can the idle parking stall between the two vehicles is determined, to improve idle vehicle The recognition effect of position.

Optionally, at least one described feature locations point includes three vertex of the corresponding external frame in bottom, and described three Vertex is used to indicate three vertex of the visible external frame in bottom in the vehicle image, and the location information is used to indicate car body Coordinate in coordinate system, it is described that frame is surrounded according to the two neighboring respective two dimension of vehicle, determine the two neighboring vehicle The location information of at least one feature locations point on the respective external frame in bottom, comprising: each according to the two neighboring vehicle From two dimension surround frame, determine three vertex of the external frame in the respective bottom of two neighboring vehicle in the bodywork reference frame In coordinate；

The location information of at least one feature locations point on the external frame in bottom according to the two neighboring vehicle, Determine the idle parking stall between the two neighboring vehicle, comprising: according to the three of the external frame in bottom of the two neighboring vehicle Coordinate of a vertex in the bodywork reference frame determines that the idle parking stall between the two neighboring vehicle is sat in the car body Coordinate in mark system.

In the optinal plan, by the two dimension encirclement frame of two neighboring vehicle, outside the bottom for identifying two neighboring vehicle Coordinate of three vertex of frame in bodywork reference frame is connect, and further determines that coordinate of the idle parking stall in bodywork reference frame, The accurate location of idle parking stall can be directly obtained by vehicle image, to realize the accurate positionin to idle parking stall.

Optionally, described that frame is surrounded according to the two neighboring respective two dimension of vehicle, determine the two neighboring vehicle Coordinate of three vertex of the respective external frame in bottom in the bodywork reference frame, comprising:

For any vehicle i in the two neighboring vehicle, the two dimension of the vehicle i is determined by key point identification model Surround the relative position of three key points on frame；Three key points are three with the external frame in bottom of the vehicle i Three characteristic points of vertex correspondence；The key point identification model is the knowledge for carrying out machine training previously according to sample image and obtaining Other model, the sample image are the images for being labeled with three key points on two dimension encirclement frame in advance；According to the vehicle i's Two dimension surrounds the relative position of three key points on frame, and three key points calculated on the two dimension encirclement frame of the vehicle i exist Two-dimensional coordinate in the coordinate system of the vehicle image；Three key points on frame are surrounded in institute according to the two dimension of the vehicle i The two-dimensional coordinate in the coordinate system of vehicle image is stated, calculates three vertex in the external frame in bottom of the vehicle i in the vehicle Coordinate in body coordinate system.

In the optinal plan, first passes through surround the sample image progress machine that frame subscript is poured in three key points in two dimension in advance Device learning training, in the application stage, wraps two dimension to be identified by key point identification model with obtaining key point identification model Image in peripheral frame is identified, with determination three key points therein, three passes on frame are surrounded to two dimension to realize The automatic identification of key point, in addition, surrounding three key points on frame in two dimension according to three vertex correspondences of the external frame in bottom Relative position calculates coordinate of three vertex of the external frame in bottom of vehicle in bodywork reference frame, provides a kind of according to two Dimension image determines seat calibration method of the external frame in the bottom of vehicle in bodywork reference frame.

Optionally, the relative position for three key points that the two dimension of the vehicle i is surrounded on frame includes: the first key point phase For the vehicle i two dimension surround frame bottom left vertex offset and the vehicle i two dimension surround frame left side line length it Between the first percentage, the second key point relative to the vehicle i two dimension surround frame bottom left vertex offset and the vehicle The two dimension of i surrounds the two dimension of the second percentage and third key point relative to the vehicle i between the following line length of frame The two dimension of the offset and the vehicle i of surrounding frame bottom right vertex surrounds the third percentage on the right of frame between line length, wherein First key point is that the two dimension of the vehicle i surrounds the characteristic point on frame left side bearing, and second key point is the vehicle The two dimension of i surrounds the characteristic point on frame lower sideline, and the third key point is that the two dimension of the vehicle i is surrounded on frame right side bearing Characteristic point；

The relative position that three key points on frame are surrounded according to the two dimension of the vehicle i, calculates the vehicle i's Two dimension surrounds two-dimensional coordinate of three key points in the coordinate system of the vehicle image on frame, comprising:

According to the two dimension packet of first percentage, second percentage, the third percentage and the vehicle i Coordinate of four vertex of peripheral frame in the coordinate system of the vehicle image calculates first key point, second key Two-dimensional coordinate in the coordinate system of point and each leisure vehicle image of the third key point.

In the optinal plan, three key points of the external frame in bottom for determining vehicle on two dimension encirclement frame are being identified When, it surrounds the excursion percentage on the vertex of frame relative to two dimension on two dimension encirclement frame by key point to indicate key point, makes The parameter section for obtaining three key points of the vehicle of different distances is consistent, and can be good at coping with the scale in different scenes Variation.

Optionally, the two dimension according to the vehicle i surrounds three key points on frame in the seat of the vehicle image Two-dimensional coordinate in mark system, calculates coordinate of three vertex of the external frame in bottom of the vehicle i in the bodywork reference frame, It include: the internal reference matrix and outer ginseng matrix for obtaining image capture device, described image acquisition equipment is the acquisition vehicle image Equipment；Two dimension of three key points in the coordinate system of the vehicle image on frame is surrounded according to the two dimension of the vehicle i Coordinate and the internal reference matrix and the outer ginseng matrix calculate three vertex of the external frame in bottom of the vehicle i described Coordinate in bodywork reference frame.

In the optinal plan, according to the internal reference matrix of image capture device and outer ginseng matrix by key point in vehicle image Coordinate system in two-dimensional coordinate be converted into vehicle the external frame in bottom coordinate of three vertex in bodywork reference frame, realize From the conversion of two-dimensional coordinate in image the coordinate into bodywork reference frame.

Optionally, three vertex of the external frame in bottom according to the two neighboring vehicle are in the bodywork reference frame In coordinate, determine coordinate of the idle parking stall between the two neighboring vehicle in the bodywork reference frame, comprising:

According to seat of three vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame Mark calculates the 4th coordinate of the vertex in the bodywork reference frame of the external frame in the two neighboring respective bottom of vehicle； According to coordinate of four vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame, institute is calculated State coordinate of the central point of the external frame in the respective bottom of two neighboring vehicle in the bodywork reference frame；According to parking stall long side side Determine that parking stall type, the parking stall longitudinal direction are according in the two neighboring vehicle to the angle between central point line Direction determined by the longitudinal direction of the external frame in the bottom of at least one vehicle, the central point line are the two neighboring vehicle The external frame in bottom central point between line, the parking stall type includes vertical parking stall, parallel space or oblique vehicle Position；Determine that parking stall quantity n, the parking stall quantity n are the parking stall numbers between the two neighboring vehicle according to the parking stall type Amount；When the n is the integer more than or equal to 2, determine n-1 Along ent on the central point line in the car body The central point line is divided into n equal portions by the coordinate in coordinate system, the n-1 Along ent；According to the parking stall long side side To and coordinate of the n-1 Along ent in the bodywork reference frame, determine the corresponding sky of n-1 Along ent Coordinate of the not busy parking stall in the bodywork reference frame.

In the optinal plan, provide a kind of external frame in bottom of two neighboring vehicle of combination central point line and Parking stall direction obtains seat calibration method of the idle parking stall between the two neighboring vehicle in the bodywork reference frame.

Optionally, it is described according to the parking stall longitudinal direction and the n-1 Along ent in the bodywork reference frame Coordinate, determine coordinate of the corresponding idle parking stall of the n-1 Along ent in the bodywork reference frame, comprising:

For Along ent i any in the n-1 Along ent, from seat of the Along ent i in the bodywork reference frame Mark starts, and moves h/2 respectively along the corresponding positive and negative both direction of the parking stall longitudinal direction, it is corresponding to obtain the Along ent i Coordinate of the midpoint of two short sides of idle parking stall in the bodywork reference frame；From the corresponding idle parking stall the Along ent i Coordinate of the midpoint in the bodywork reference frame of two short sides start, along the short of the corresponding idle parking stall the Along ent i The corresponding positive and negative both direction of edge direction moves w/2 respectively, and four vertex for obtaining the corresponding idle parking stall the Along ent i exist Coordinate in the bodywork reference frame；Wherein, h is default parking stall length, and w is default parking stall width, and the Along ent i is corresponding The short side direction of idle parking stall is vertical with the parking stall longitudinal direction；

Alternatively, for Along ent i any in the n-1 Along ent, from the Along ent i in the bodywork reference frame Coordinate start, the corresponding positive and negative both direction of short side direction along the corresponding idle parking stall the Along ent i moves w/ respectively 2, obtain coordinate of the midpoint of two long sides of the corresponding idle parking stall the Along ent i in the bodywork reference frame；From institute It states coordinate of the midpoint of two long sides of the corresponding idle parking stall Along ent i in the bodywork reference frame to start, along the vehicle The corresponding positive and negative both direction of bit length edge direction moves h/2 respectively, obtains four tops of the corresponding idle parking stall the Along ent i Coordinate of the point in the bodywork reference frame；Wherein, h is default parking stall length, and w is default parking stall width, i pairs of the Along ent The short side direction for the idle parking stall answered is vertical with the parking stall longitudinal direction.

The optional scheme provides two kinds of seats according to parking stall longitudinal direction and Along ent i in bodywork reference frame Mark determines seat calibration method of four vertex of idle parking stall in bodywork reference frame.

Optionally, the angle according between parking stall longitudinal direction and central point line determines parking stall type, comprising:

When the angle is 90 °, determine that the parking stall type is vertical parking stall；

When the angle is 0 °, determine that the parking stall type is parallel space；

When the angle is in (30 °, 60 °) or (120 °, 150 °) section, determine that the parking stall type is oblique Parking stall.

It is optionally, described that quantity n in parking stall is calculated according to the parking stall type, comprising:

When the parking stall type is vertical parking stall, n is the maximum integer less than or equal to d/w；

When the parking stall type is parallel space, n is the maximum integer less than or equal to d/h；

When the parking stall type is oblique parking stall, n is the maximum integer less than or equal to d/ (w/sin α)；

Wherein, d is the length of the central point line, and w is default parking stall width, and h is default parking stall length, and α is described Angle.

The optional scheme provides a kind of center of external frame in the bottom according to parking stall type and two neighboring vehicle The method that the length of two lines of point determines the quantity of the idle parking stall between two neighboring vehicle.

Second aspect provides a kind of method of determining idle parking stall, which comprises obtain comprising described at least two The spatial perception data of a vehicle；It is each that two neighboring vehicle at least two vehicle is obtained according to the spatial perception data From the external frame in bottom at least one feature locations point location information, the external frame in bottom is corresponding vehicle on ground On orthographic projection boundary rectangle frame, the feature locations point is used to indicate the point on the predeterminated position of the external frame in bottom；According to The location information of at least one feature locations point on the external frame in bottom of the two neighboring vehicle, determines described two neighboring Idle parking stall between vehicle.

In this scenario, pass through at least one on the external frame in bottom of the two neighboring vehicle in identification spatial perception data The location information of a feature locations point, and determined according to the location information of at least one feature locations point on the external frame in bottom Idle parking stall between two neighboring vehicle, it is only necessary to which existing in spatial perception data can recognize that the adjacent of the external frame in bottom Two vehicles, it can the idle parking stall between the two vehicles is determined, to improve the recognition effect of idle parking stall.

The third aspect provides a kind of model training method, which comprises obtains sample image, the sample graph It seem the image for being labeled with three key points on two dimension encirclement frame in advance, the two dimension surrounds the boundary rectangle that frame is vehicle Frame, three key points are three of three vertex correspondences in the external frame in bottom of the vehicle in the sample image Characteristic point, three vertex are used to indicate three vertex of the visible external frame in bottom in the sample image, and described three A key point is respectively at the two dimension and surrounds on the left side bearing, lower sideline and right side bearing of frame, and the external frame in bottom is pair Answer the boundary rectangle frame of the orthographic projection of vehicle on the ground；Model training is carried out according to the sample image, obtains key point knowledge Other model, the key point identification model three tops in the external frame in bottom of the vehicle to be identified in input picture for identification The relative position of corresponding three key points in the input picture of point, the input picture is the two of the vehicle to be identified Dimension surrounds the image in frame.

In this scenario, it first passes through in advance and surrounds the sample image progress engineering that frame subscript is poured in three key points in two dimension Training is practised, to obtain key point identification model, to be wrapped to two dimension to be identified by key point identification model in the application stage Image in peripheral frame is identified, with determination three key points therein, three passes on frame are surrounded to two dimension to realize The automatic identification of key point.

Fourth aspect, provides a kind of device of determining idle parking stall, the device have realize above-mentioned first aspect and The function of the method for free time parking stall is determined provided by the optional implementation of first aspect.The function can pass through hardware It realizes, corresponding software realization can also be executed by hardware.The hardware or software include one or more than one with it is above-mentioned The corresponding unit of function.

5th aspect provides a kind of device of determining idle parking stall, which, which has, realizes that above-mentioned second aspect is mentioned The function of the method for determining idle parking stall of confession.The function can also be executed corresponding by hardware realization by hardware Software realization.The hardware or software include one or more than one unit corresponding with above-mentioned function.

6th aspect, provides a kind of model training apparatus, which, which has, realizes mould provided by the above-mentioned third aspect The function of type training method.The function can also execute corresponding software realization by hardware realization by hardware.Institute It states hardware or software includes one or more than one unit corresponding with above-mentioned function.

7th aspect, provides a kind of parking stall and determines equipment, which determines that equipment includes: processor and memory, institute Memory is stated for storing programmable instructions, the processor is for calling the programmable instructions stored on the reservoir to hold The method that determines idle parking stall of the row as described in the optional implementation of above-mentioned first aspect and first aspect.

Eighth aspect provides a kind of parking stall and determines equipment, which determines that equipment includes: processor and memory, institute Memory is stated for storing programmable instructions, the processor is for calling the programmable instructions stored on the reservoir to hold The method that determines idle parking stall of the row as described in above-mentioned second aspect.

9th aspect, provides a kind of parking stall and determines equipment, which determines that equipment includes: processor and memory, institute Memory is stated for storing programmable instructions, the processor is for calling the programmable instructions stored on the reservoir to hold Model training method of the row as described in the above-mentioned third aspect.

Tenth aspect provides a kind of computer readable storage medium, and including in the computer readable storage medium can Programming instruction, computer execute the programmable instructions to realize the optional realization such as above-mentioned first aspect and first aspect The method of free time parking stall is determined described in scheme.

On the one hand tenth, provides a kind of computer readable storage medium, includes in the computer readable storage medium Programmable instructions, computer execute the programmable instructions to realize the side for determining idle parking stall as described in above-mentioned second aspect Method.

12nd aspect, provides a kind of computer readable storage medium, includes in the computer readable storage medium Programmable instructions, computer execute the programmable instructions to realize that the model as described in the above-mentioned third aspect determines method.

13rd aspect provides a kind of computer program, includes at least one programmable finger in the computer program Enable, when computer program operation in a computer when, computer can by execute at least one described programmable instructions with Realize the method for determining idle parking stall as described in the optional implementation of above-mentioned first aspect and first aspect.

Fourteenth aspect provides a kind of computer program, includes at least one programmable finger in the computer program Enable, when computer program operation in a computer when, computer can by execute at least one described programmable instructions with Realize the method for determining idle parking stall as described in above-mentioned second aspect.

15th aspect provides a kind of computer program, includes at least one programmable finger in the computer program Enable, when computer program operation in a computer when, computer can by execute at least one described programmable instructions with Realize that the model as described in the above-mentioned third aspect determines method.

Detailed description of the invention

Fig. 1 is the architecture diagram that the system of free time parking stall is determined involved in the embodiment of the present application；

Fig. 2 is the flow chart for the method for determining idle parking stall that one exemplary embodiment of the application provides；

Fig. 3 is a kind of image detection schematic diagram that embodiment illustrated in fig. 2 is related to；

Fig. 4 is a kind of two-dimentional frame detection schematic diagram that embodiment illustrated in fig. 2 is related to；

Fig. 5 is a kind of visible key point schematic diagram that embodiment illustrated in fig. 2 is related to；

Fig. 6 is a kind of method flow diagram for model training method that embodiment illustrated in fig. 2 is related to；

Fig. 7 is the original that the second deep neural network of one kind that embodiment illustrated in fig. 2 is related to returns key point normalization numerical value Reason figure；

Fig. 8 is the convolution that embodiment illustrated in fig. 2 is related to and prediction principle figure；

Fig. 9 is a kind of camera imaging schematic diagram that embodiment illustrated in fig. 2 is related to；

Figure 10 is the external frame schematic diagram of a kind of vehicle bottom that embodiment illustrated in fig. 2 is related to；

Figure 11 is the flow chart for the idle parking stall of identification that embodiment illustrated in fig. 2 is related to；

Figure 12 is the idle parking stall type identification flow chart that embodiment illustrated in fig. 2 is related to；

Figure 13 is a kind of parking stall type schematic diagram that embodiment illustrated in fig. 2 is related to；

Figure 14 is the idle parking stall position identification process figure that embodiment illustrated in fig. 2 is related to；

Figure 15 is that there is no the schematic diagrames of idle parking stall between two neighboring vehicle that embodiment illustrated in fig. 2 is related to；

Figure 16 is the schematic diagram that there is idle parking stall between two neighboring vehicle that embodiment illustrated in fig. 2 is related to；

Figure 17 is a kind of schematic diagram for determining fixed point coordinate that embodiment illustrated in fig. 2 is related to；

Figure 18 is the flow chart for the method for determining idle parking stall that the exemplary embodiment of the application provides；

Figure 19 is the structural schematic diagram that the parking stall that one exemplary embodiment of the application provides determines equipment；

Figure 20 is the structural schematic diagram for the model training equipment that one exemplary embodiment of the application provides；

Figure 21 is a kind of structure chart of the device for determining idle parking stall that the exemplary embodiment of the application provides；

Figure 22 is a kind of structure chart for model training apparatus that the exemplary embodiment of the application provides.

Specific embodiment

To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with attached drawing to the application embodiment party Formula is described in further detail.

Fig. 1 is the architecture diagram that the system of free time parking stall is determined involved in the embodiment of the present application.As shown in Figure 1, the determination The system of idle parking stall may include that data acquisition equipment 110 and parking stall determine equipment 120.

Wherein, data acquisition equipment 110 can be the camera device of acquisition two dimensional image；Alternatively, data acquisition equipment 110 are also possible to the equipment for acquiring spatial perception data, for example, the data acquisition equipment 110 can be acquisition radar points cloud number According to radar detecting equipment, alternatively, the data acquisition equipment 110 be also possible to acquire laser point cloud data laser detection equipment Deng.

Parking stall determines that equipment 120 can be the computer equipment with data identifying processing ability.

Above-mentioned data acquisition equipment 110 can be set in the car, and acquire the vehicle image of vehicle periphery, and will acquisition To vehicle data information be sent to parking stall and determine equipment 120, parking stall determines equipment 120 by knowing to vehicle data information Not, to determine in vehicle data information with the presence or absence of idle parking stall, and attitude information (such as the parking stall of idle parking stall The information such as coordinate of the vertex in bodywork reference frame), so as to the application such as subsequent progress automatic parking.

The system of the determination free time parking stall can be independently equipped in vehicle, for example, above-mentioned data acquisition equipment 110 and Parking stall determines that equipment 120 is all disposed in vehicle, and data acquisition equipment 110 and parking stall determine and pass through wired network between equipment 120 Network is communicated.

Alternatively, the system of the determination free time parking stall can also be deployed in vehicle and network server respectively, for example, above-mentioned Data acquisition equipment 110 can be set in the car, and above-mentioned parking stall determines that equipment 120 can be set in network server, Data acquisition equipment 110 and parking stall, which determine, to be communicated between equipment 120 by wireless network.

In actual parking lot or parking position, for the ease of management, parking stall is typically all according to certain rule Restrain concentrated setting, for example, in parking lot usual opposite lane it is vertical or it is oblique several parking stalls are continuously set, and this is several It is parallel to each other between a parking stall, and when parking stall is arranged in roadside, usually relative to parallel continuously setting several vehicles in lane Position, and it is end to end between several parking stalls.Therefore, for the vehicle of parked state, when between two neighboring vehicle Distance it is enough when, have that very big there may be idle parking stalls between the two neighboring vehicle.

Based on the above principles, scheme shown in the embodiment of the present application can be closed by the distance between two neighboring vehicle System, the position determined the parking stall between two vehicles with the presence or absence of the free time, there are how many idle parking stalls and each idle parking stall Confidence breath.It is specific for example, parking stall determines that equipment 120, can be with when being identified vehicle data information to determine idle parking stall The vehicle data information for containing at least two vehicle is obtained, and according to adjacent two at least two vehicle of vehicle data information acquisition The location information of at least one feature locations point on the external frame in a respective bottom of vehicle, and according to the bottom of two neighboring vehicle The location information of at least one feature locations point on the external frame in portion, determines the idle parking stall between the two neighboring vehicle.It is logical This method is crossed, as long as in collected vehicle data information including at least two vehicles that can identify the external frame in bottom, i.e., It can identify the idle parking stall between the two vehicles, not need the parking stall line on identification ground, in this embodiment it is not even necessary on ground There are parking stall lines, higher to the recognition accuracy of idle parking stall；In addition, method shown in the embodiment of the present application, it is only necessary to acquire To the two dimensional image or spatial perception data for containing at least two Adjacent vehicles, it can the identification for realizing idle parking stall, from And can identify the idle parking stall between remote vehicle, remote to the identification distance of idle parking stall, recognition effect is more preferable.

In the embodiment of the present application, above-mentioned parking stall determines that equipment 120 can carry out the knowledge of idle parking stall according to two dimensional image Not, the identification of idle parking stall can also be carried out by spatial perception data.Embodiment shown in following FIG. 2 will be according to X-Y scheme As being introduced for carrying out the identification of idle parking stall.

Referring to FIG. 2, the stream of the method for determining idle parking stall provided it illustrates one exemplary embodiment of the application Cheng Tu.The parking stall that this method can be used for system shown in Figure 1 determines in equipment 120.As shown in Fig. 2, the determination free time parking stall Method may include:

Step 201, the vehicle image for containing at least two vehicle is obtained.

In the embodiment of the present application, data acquisition equipment can acquire the image of vehicle periphery, and by the figure of vehicle periphery As being sent to after parking stall determines that equipment, parking stall determine that equipment receives image capture device acquired image, to receiving Image tentatively identified, in the image whether include two or more vehicle to determine, if so, determination obtains The vehicle image is got, otherwise it is assumed that vehicle image has not been obtained.

Step 202, determine that the two neighboring respective two dimension of vehicle is surrounded at least two vehicle according to the vehicle image Frame, it is the boundary rectangle frame that vehicle is corresponded in the vehicle image which, which surrounds frame,.

The boundary rectangle frame of object in image refer to four edges respectively with object in the picture most upper pixel, most play picture The rectangle frame that vegetarian refreshments, most left pixel and most right pixel connect, boundary rectangle frame can usually be used to indicate that object is being schemed Position as in.In the embodiment of the present application, it is indicated by the boundary rectangle frame of the vehicle in image (i.e. two dimension surrounds frame) The position of vehicle in the picture.

It in the embodiment of the present application, can be by surrounding frame identification model (for example, the encirclement frame identification model can be the One deep neural network) image recognition is carried out to the vehicle image, it is adjacent to identify at least two vehicles in vehicle image The two dimension of two vehicles surrounds frame, which can be that carry out machine previously according to first sample image trained The identification model arrived, the sample image can be the image for being labeled with the two dimension encirclement frame of each vehicle in advance.

Above-mentioned encirclement frame identification model can carry out machine training by model training equipment and obtain, wherein model instruction Practice equipment and the parking stall determines that equipment can be different two computer equipments, for example, the model training equipment is network side Server, and parking stall determines that equipment is mobile unit, alternatively, the model training equipment determines that equipment is also possible to the parking stall The same equipment, for example, the model training equipment and parking stall determine that equipment is realized in the same server of network side.

For example, developer can collect several before parking stall determines that equipment carries out image recognition to vehicle image First sample image comprising vehicle, and frame is surrounded in the two dimension that the acceptance of the bid of first sample image is poured in each vehicle, it later will mark It is poured in the first sample image input model training equipment that two dimension surrounds frame, is surrounded by model training equipment according to the good two dimension of mark The first sample image training of frame obtains above-mentioned encirclement frame identification model, and subsequent parking stall determines that equipment carries out image to vehicle image When identification, image recognition is carried out according to the encirclement frame identification model that the training of model training equipment obtains, to obtain vehicle image packet In at least two vehicles contained, the two neighboring respective two dimension of vehicle surrounds frame.

For example, referring to FIG. 3, it illustrates the invention relates to a kind of image detection schematic diagram, such as Fig. 3 institute Show, after the first deep neural network receives input picture, feature extraction is carried out to input picture first, obtains the spy of small size Sign figure (such as characteristic pattern that small size can be extracted by way of multilayer convolution) is carrying out two according to the characteristic pattern of small size Dimension surrounds frame detection.

Wherein, the first deep neural network can be detected by reference frame two dimension surround frame, for example, referring to FIG. 4, its Show the invention relates to a kind of two-dimentional frame detection schematic diagram, as shown in figure 4, the first deep neural network is in small ruler The default k benchmark of each location of pixels of very little characteristic pattern surrounds frame, and surrounds the position of frame relative to each benchmark by prediction Deviation is set to determine the position of true two dimension encirclement frame in image.Wherein, the 4k positional number exported in Fig. 4 via middle layer According to the position deviation for indicating relatively a k reference frame, corresponding benchmark in addition is indicated via 2k categorical data of middle layer output The classification (such as background or contain object) of frame region is surrounded, the first deep neural network is selected by preset threshold value The reference frame containing object is selected, the recognition result that final two dimension surrounds frame is then obtained according to corresponding deviation.

Optionally, the first deep neural network that the two dimension of above-mentioned vehicle for identification surrounds frame can be fast area volume Product neural network (Faster Region Convolutional Neural Networks, Faster RCNN) network, single case Detector (Single Shot MultiBox Detector, SSD) network or uniformly real-time objects detection (Unified, Real-Time Object Detection) network etc., wherein unified real-time objects detection network is also referred to as YOLO (full name You Only Look Once) network.The embodiment of the present application for the first deep neural network type without limitation.

Parking stall identification equipment execution step 202 obtains the respective two dimension of two neighboring vehicle and surrounds after frame, it can root Frame is surrounded according to the respective two dimension of two neighboring vehicle, determines at least one spy on the external frame in the respective bottom of two neighboring vehicle The location information of location point is levied, the external frame in the bottom is the boundary rectangle frame of the orthographic projection of corresponding vehicle on the ground, the spy Sign location point is point on the predeterminated position of the external frame in the bottom, according on the external frame in the bottom of the two neighboring vehicle at least The location information of one feature locations point, it can determine the idle parking stall between the two neighboring vehicle.

It is specific for example, including three vertex of the corresponding external frame in bottom, this three tops at least one feature locations point Point is used to indicate three vertex of the visible external frame in bottom in vehicle image, and the location information is used to indicate bodywork reference frame In coordinate for, according to two neighboring vehicle it is respective two dimension surround frame, determine outside the respective bottom of two neighboring vehicle When connecing the location information of at least one feature locations point on frame, parking stall identifies that equipment can be respective according to two neighboring vehicle Two dimension surrounds frame, determines coordinate of three vertex of the external frame in the respective bottom of two neighboring vehicle in bodywork reference frame.It should Determine the process of coordinate of three vertex of the external frame in the respective bottom of two neighboring vehicle in bodywork reference frame with reference to subsequent Step 203 is to step 204.

Step 203, for any vehicle i in two neighboring vehicle, determine that the two dimension of vehicle i surrounds three on frame Two-dimensional coordinate of the key point in the coordinate system of the vehicle image, three key points are the external frames in bottom with vehicle i Three characteristic points of three vertex correspondences.

The external frame in the bottom of vehicle is rectangle frame, and four vertex of the rectangle frame are 4 spies of the external frame of vehicle bottom Levy location point.It can detect that two dimension surrounds the vehicle of frame for one, it is corresponding in 4 vertex of the external frame in the bottom of the vehicle Visible characteristic point usually has 3 in vehicle image, and 3 vertex correspondences in the external frame in the bottom of the vehicle are in vehicle image In characteristic point be properly termed as corresponding 3 key points of the vehicle, and this 3 visible key points are respectively at two dimension and surround frame Left, down, right edge on.For example, referring to FIG. 5, it illustrates the invention relates to a kind of visible key point signal Figure, as shown in figure 5, the corresponding three visible key points in vehicle image of the external frame 51 in the bottom of vehicle are respectively P1, P2 And P3, wherein key point P1 is located at the left edge that vehicle two dimension surrounds frame 52, and key point P2 is located at vehicle two dimension and surrounds frame 52 Lower edge, key point P3 are located at the right hand edge that vehicle two dimension surrounds frame 52, the 4th vertex in the external frame 51 in the bottom of vehicle Corresponding key point P4 is in the state that is blocked, and surrounds in the two dimension of vehicle invisible in the image-region that frame is surrounded.

Optionally, for any vehicle i in two neighboring vehicle, three on frame are surrounded in the two dimension for determining vehicle i When two-dimensional coordinate of the key point in the coordinate system of the vehicle image, parking stall identifies that equipment can be true by key point identification model The two dimension of fixed vehicle i surrounds the relative position of three key points on frame, and according to three on the two dimension encirclement frame of vehicle i The relative position of a key point, the two dimension for calculating vehicle i surround three key points on frame in the coordinate system of the vehicle image In two-dimensional coordinate.

Wherein, which is to carry out the identification mould that machine training obtains according to the second sample image in advance Type, second sample image are the images for being labeled with three key points on two dimension encirclement frame in advance.For example, implementing in the present invention In example, which can be the second deep neural network.

Please refer to Fig. 6 it illustrates the present embodiments relate to a kind of model training method method flow diagram, such as Fig. 6 Shown, which may include steps of:

Step 61, the second sample image is obtained.

Wherein, when obtaining the second sample image, model training equipment can receive second sample of developer's input This image.

Step 62, model training is carried out according to second sample image, obtains key point identification model.

Wherein, the key point identification model for identification in the external frame in bottom of the vehicle to be identified in input picture three The relative position of three key points of a vertex correspondence in the input image, the input picture are that the two dimension of vehicle to be identified is surrounded Image in frame.

For example, developer can before parking stall determines that the image that equipment surrounds in frame the two dimension of vehicle identifies To collect the sample image that the two dimension of several vehicles surrounds frame, and surrounded in the sample image of frame in the two dimension of several vehicles The good two dimension of mark surrounds three key points on frame respectively, obtains the second sample image of three key points of mark, later will The the second sample image input model training equipment for having marked key point, by model training equipment according to the training of the second sample image Above-mentioned key point identification model is obtained, when the image that subsequent parking stall determines that equipment surrounds in frame the two dimension of vehicle identifies, Image recognition is carried out according to the key point identification model that the training of model training equipment obtains, i.e. parking stall determines equipment by the two of vehicle Dimension surrounds the image input key point identification model that frame is surrounded, and surrounds frame by the two dimension that key point identification model exports the vehicle On three key points relative position.

In embodiments of the present invention, above-mentioned second sample image can be the two dimension packet marked out in first sample image Image in peripheral frame, i.e. developer collect the first sample image that several include vehicle, and get the bid in first sample image The two dimension for being poured in each vehicle is surrounded after frame, is also further surrounded in the two dimension of mark and is marked three key points on frame, will mark The two dimension for being poured in three key points surrounds the image that is surrounded of frame as the second sample image.Alternatively, above-mentioned second sample image It is also possible to intercept out from other images except first sample image, the two dimension of the vehicle in other images surrounds frame In image.

After parking stall determines the relative position that equipment identification obtains three key points on the two dimension encirclement frame of vehicle, it can Two-dimensional coordinate of four vertex of frame in the coordinate system of vehicle image is surrounded in conjunction with the two dimension of vehicle, calculates the two dimension of vehicle Surround two-dimensional coordinate of three key points on frame in the coordinate system of vehicle image.

For example, the relative position for three key points surrounded on frame with the two dimension of above-mentioned vehicle is three key points and two Dimension is surrounded for the relative position between four vertex of frame, and parking stall determines that equipment obtains this three by key point identification model After key point and two dimension surround the relative position between four vertex of frame, four vertex of frame are surrounded in vehicle in conjunction with the two dimension Two-dimensional coordinate in the coordinate system of image can calculate three key points in the coordinate of vehicle image according to geometrical principle Two-dimensional coordinate in system.

Optionally, for any vehicle i in two neighboring vehicle, determining vehicle i's by key point identification model When two dimension surrounds the relative position of three key points on frame, it can be wrapped by two dimension of the key point identification model to vehicle i Image in peripheral frame carries out image recognition, and the two dimension for obtaining the first key point relative to vehicle i surrounds the inclined of frame bottom left vertex The first percentage, the second key point between shifting amount and the two dimension encirclement frame left side line length of vehicle i is relative to vehicle i's Two dimension surround frame bottom left vertex offset and vehicle i two dimension surround the following line length of frame between the second percentage and Third key point is surrounded relative to the two dimension of vehicle i on the right of the offset of frame bottom right vertex and the two dimension encirclement frame of vehicle i Third percentage between line length.The relative position of three key points on frame is surrounded in the two dimension according to vehicle i, calculates vehicle When the two dimension of i surrounds two-dimensional coordinate of three key points on frame in the coordinate system of vehicle image, parking stall identifies equipment root Four vertex of frame are surrounded at this according to the two dimension of first percentage, second percentage, the third percentage and vehicle i Coordinate in the coordinate system of vehicle image calculates first key point, second key point and each leisure of third key point Two-dimensional coordinate in the coordinate system of the vehicle image.Wherein, the first key point is that the two dimension of vehicle i surrounds the spy on frame left side bearing Point is levied, the second key point is that the two dimension of the vehicle i surrounds the characteristic point on frame lower sideline, and third key point is the vehicle i Two dimension surround frame right side bearing on characteristic point.

Optionally, in embodiments of the present invention, above by the first percentage of key point identification model acquisition, the 200th Divide ratio and third percentage, the two dimension of the as vehicle surrounds three key points on frame in the coordinate system of the vehicle image Relative position.

In the embodiment of the present application, three passes on the two dimension encirclement frame to determine vehicle by the second deep neural network For key point, the input of second deep neural network is that the two dimension of vehicle surrounds the image in frame, as shown in figure 5, second is deep It is respectively py1, px2 and py3 that degree neural network, which needs the value predicted, wherein py1 is to surround frame left side bearing with the two dimension of vehicle Length be 1, to P1 point to two dimension encirclement the distance between frame bottom left vertex be normalized as a result, px2 is with vehicle The length that two dimension surrounds frame lower sideline is 1, surrounds the knot that the distance between frame bottom left vertex is normalized to two dimension to P2 point Fruit, py3 be with vehicle two dimension surround frame right side bearing length be 1, to P3 point to two dimension encirclement frame bottom right vertex between away from From be normalized as a result, to above three numerical value carry out percentageization can obtain above-mentioned first percentage, second Percentage and third percentage.

Referring to FIG. 7, it illustrates the invention relates to the second deep neural network of one kind return key point return One changes the schematic diagram of numerical value.As shown in fig. 7, the feature extraction sub-network in the second deep neural network surrounds the two dimension of vehicle Image in frame carries out feature extraction, and the second deep neural network is respectively py1 according to the value for the feature output prediction extracted, Px2, py3, represent each key point and two dimension surrounds the normalized cumulant of frame corresponding vertex.

By taking second deep neural network is convolutional neural networks as an example, above-mentioned detection process can refer to shown in Fig. 8 Convolution and prediction principle figure carry out characteristic pattern extraction by the image that the two dimension to vehicle is surrounded in frame, and to mentioning in fig. 8 The characteristic pattern got carries out convolution operation several times and is directly exported with constantly reducing characteristic pattern size finally by full articulamentum Above-mentioned 3 predicted values.

After determining above three predicted value, it can surround coordinate of four vertex in vehicle image of frame according to two dimension Coordinate in system, to determine that the two dimension of vehicle surrounds two dimension seat of three key points on frame in the coordinate system of vehicle image Mark.

Above-mentioned py1, px2, py3 value range is between (0,1).By taking P2 key point as an example, if px2=0, represent P2 with The bottom left vertex that the two dimension of vehicle surrounds frame is overlapped, if px2=1, the two dimension for representing P2 and vehicle surrounds frame bottom right vertex weight It closes.It is different with the two dimension encirclement pixel distance on four vertex of frame from directly three key points of prediction, it is pre- in scheme shown in the application Survey normalized as a result, it is possible to cope with the dimensional variation in different scenes well, i.e. the vehicle for different distances, thirdly A corresponding parameter value range of key point will be consistent.

Step 204, three key points on frame are surrounded in the coordinate system of the vehicle image according to the two dimension of vehicle i Two-dimensional coordinate calculates coordinate of three vertex of the external frame in bottom of vehicle i in bodywork reference frame.

It in the embodiment of the present application, can be in conjunction with the nominal data of the image capture device of acquisition vehicle image, Yi Jishang Two-dimensional coordinate of three key points in the coordinate system of vehicle image is stated, is determined outside the corresponding bottom of above three key point Connect coordinate of three vertex of frame in bodywork reference frame.

Optionally, above-mentioned nominal data can be the internal reference matrix and outer ginseng matrix of image capture device, for adjacent two Any vehicle i in a vehicle, parking stall identify the internal reference matrix and outer ginseng matrix of the available image capture device of equipment, and root Two-dimensional coordinate and the internal reference of three key points in the coordinate system of the vehicle image on frame are surrounded according to the two dimension of vehicle i Matrix and the outer ginseng matrix, calculate coordinate of three vertex of the external frame in bottom of vehicle i in bodywork reference frame.

Wherein, the coordinate in above-mentioned bodywork reference frame can be two-dimensional coordinate, be also possible to three-dimensional coordinate.For example, working as vehicle When on the level ground, Z axis of the external frame in the bottom of vehicle in bodywork reference frame is (i.e. perpendicular to the coordinate of horizontal plane Axis) coordinate is always 0, at this point, in coordinate of three vertex of the external frame in bottom for determining vehicle i in bodywork reference frame, Only need to consider the X axis coordinate and Y axis coordinate in horizontal plane.And when vehicle is on slope or rugged ground When, Z axis coordinate of the external frame in the bottom of vehicle in bodywork reference frame may not be 0, at this point, outside the bottom for determining vehicle i When connecing coordinate of three vertex of frame in bodywork reference frame, in addition to need to consider X axis coordinate in horizontal plane and Y axis coordinate it Outside, it is also necessary to consider the coordinate of Z axis.

Wherein, the internal reference matrix of above-mentioned image capture device and outer ginseng matrix can be demarcated in advance and be stored in parking stall identification In equipment, when obtaining the internal reference matrix and outer ginseng matrix of image capture device, parking stall identifies that equipment can be directly from local reading Take the internal reference matrix and outer ginseng matrix；Alternatively, the internal reference matrix of the image capture device and outer ginseng matrix can also be demarcated in advance And be stored in image capture device, parking stall identifies that equipment can be to the image capture device request internal reference matrix and outer ginseng Matrix.

Referring to FIG. 9, it illustrates the invention relates to a kind of camera imaging schematic diagram.Shadow surface represents in figure Plane where the coordinate system of vehicle image, x, y-coordinate axis are the reference axis in the coordinate system of vehicle image, the pass identified The coordinate of key point is the coordinate (u, v) in the coordinate system of vehicle image.The corresponding point in bodywork reference frame of the key point is Fig. 9 In P point (i.e. above-mentioned P1, P2 or P3), coordinate in bodywork reference frame is (X, Y, Z).Pass through above-mentioned second depth mind Go out after the two-dimensional coordinate (u, v) of three key points on the external frame in bottom through Network Recognition, can use image capture device Imaging formula calculate coordinate (X, Y, Z) of the key point under bodywork reference frame.

Wherein, imaging formula are as follows:

In above-mentioned imaging formula, M1, M2 are respectively the internal reference matrix and outer ginseng matrix of image capture device, the two matrixes It can be obtained by the preparatory calibration to image capture device.Since the external frame in bottom is the packet of the view field of vehicle on the ground Peripheral frame, when vehicle is on level ground, key point on frame external for all bottoms has Z=0.Therefore it is above-mentioned at As u, v, M in formula₁, M₂And Z is known quantity, only s, X, tri- unknown quantitys of Y, and above-mentioned imaging formula can 3 sides of simultaneous Journey, i.e. above three unknown variable s, X, Y have unique solution.

Step 205, the coordinate according to three vertex of the external frame in bottom of two neighboring vehicle in bodywork reference frame, really Coordinate of the idle parking stall in bodywork reference frame between fixed two neighboring vehicle.

Wherein, parking stall identification equipment can be according to three vertex of the external frame in the two neighboring respective bottom of vehicle at this Coordinate in bodywork reference frame calculates the 4th vertex of the external frame in the two neighboring respective bottom of vehicle in the car body coordinate Coordinate in system；According to seat of four vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame Mark, calculates coordinate of the central point of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame；According to parking stall Angle between longitudinal direction and central point line determines parking stall type, which is according to the two neighboring vehicle In at least one vehicle the external frame in bottom longitudinal direction determined by direction, the central point line be the two neighboring vehicle The external frame in bottom central point between line, which includes vertical parking stall, parallel space or oblique parking stall；Root According to the parking stall, type determines parking stall quantity n, and parking stall quantity n is the parking stall quantity between the two neighboring vehicle；When the n is big In or equal to 2 integer when, determine coordinate of the n-1 Along ent on the central point line in the bodywork reference frame, should The central point line is divided into n equal portions by n-1 Along ent；According to the parking stall longitudinal direction and the n-1 Along ent at this Coordinate in bodywork reference frame determines coordinate of the corresponding idle parking stall of the n-1 Along ent in the bodywork reference frame.

By taking any vehicle i in two neighboring vehicle as an example, three vertex of the external frame in bottom of vehicle i are being obtained in vehicle After coordinate in body coordinate system, parking stall identifies that equipment can be sat according to three vertex of the external frame in bottom of vehicle i in car body Coordinate in mark system, calculates the 4th coordinate of the vertex in bodywork reference frame in the external frame in bottom of vehicle i.

Figure 10 show the invention relates to a kind of external frame schematic diagram of vehicle bottom, wherein the bottom of vehicle External frame is a rectangle, and the position distribution on corresponding four vertex is as shown in Figure 10, and wherein the coordinate of P1, P2, P3 are it is known that C point Coordinate for the central point of the external frame in bottom, P4 is unknown, if coordinate of the P1 in bodywork reference frame is (X₁, Y₁, Z₁), P2 is in vehicle Coordinate in body coordinate system is (X₂, Y₂, Z₂), coordinate of the P3 in bodywork reference frame is (X₃, Y₃, Z₃), P4 is in bodywork reference frame In coordinate be (X₄, Y₄, Z₄), coordinate of the C point in bodywork reference frame is (X_C, Y_C, Z_C)。

The central point C coordinate of rectangle can be calculated by P1 coordinate and P3 coordinate:

Similarly, central point C coordinate can also be calculated by P2 and P4:

The above-mentioned two formula of simultaneous can calculate coordinate of the P4 point in bodywork reference frame based on following formula.

So far, coordinate of 4 vertex of the external frame of vehicle bottom in bodywork reference frame all successfully calculates in image Come.

Wherein, above-mentioned parking stall longitudinal direction refers to the direction on two longer sides in parking stall, and parking stall central point refers to one The geometric center point of parking stall.

Wherein, above-mentioned vertical parking stall refers to parking stall longitudinal direction perpendicular to the parking stall in lane, and parallel space refers to that parking stall is long Edge direction is parallel to the parking stall in lane, and oblique parking stall refer to parking stall longitudinal direction and track direction between it is parallel and it is vertical it Between, it is generally the case that the parking stall longitudinal direction of oblique parking stall and the angle of track direction be in (30 °, 60 °) or (120 °, 150 °) in section.

In practical applications, for different types of parking stall, the distance between central point of two neighboring parking stall also not phase Together, for example, when two neighboring parking stall is vertical parking stall, a long side is shared between the two vertical parking stalls, at this point, this is adjacent The distance between central point of two parking stalls is a parking stall width；For another example, when two neighboring parking stall is parallel space, this A short side is shared between two parallel spaces, at this point, the distance between central point of the two neighboring parking stall is a parking stall Length；For another example, when two neighboring parking stall is oblique parking stall, the distance between central point of the two oblique parking stalls is between one Between a parking stall length and a parking stall width, specific value depends on the size of the angle between oblique parking stall and lane.Cause This is being calculated after coordinate of four vertex of the external frame in bottom for determining two neighboring vehicle in bodywork reference frame When the distance between the central point of the external frame in the bottom of two vehicles, need to determine the two neighboring vehicle in conjunction with parking stall type The external frame in bottom central point line on can mark off several Along ents, when the quantity for the Along ent that can be marked off is greater than Or when being equal to 1 (i.e. above-mentioned n be more than or equal to 2), it can determine between the two neighboring vehicle there may be idle parking stall, The location information of idle parking stall can be just determined according to Along ent at this time.

Optionally, this determines parking stall type according to the angle between parking stall longitudinal direction and central point line, comprising: when this When angle is 90 °, determine that the parking stall type is vertical parking stall；When the angle is 0 °, determine that the parking stall type is parallel space； When the angle is in (30 °, 60 °) or (120 °, 150 °) section, determine that the parking stall type is oblique parking stall.

Optionally, this determines parking stall quantity n according to the parking stall type, comprising: when the parking stall type is vertical parking stall, n For the maximum integer less than or equal to d/w；When the parking stall type is parallel space, n is the maximum less than or equal to d/h Integer；When the parking stall type is oblique parking stall, n is the maximum integer less than or equal to d/ (w/sin α)；Wherein, d is to be somebody's turn to do The length of central point line, w are default parking stall width, and h is default parking stall length, and α is the angle.

Wherein, the value of above-mentioned w and h can be set in advance in parking stall identification equipment, specifically, the w and h can be by developing Personnel or administrative staff are configured according to the length and width of preset standard parking stall, i.e. w is the width of preset standard parking stall, H is the length of preset standard parking stall.

Figure 11 be the invention relates to a kind of parking stall of identification free time flow chart.As shown in figure 11, the free time vehicle Position identification process is carried out based on the recognition result of the external frame of the vehicle bottom of input, and the identification process is broadly divided into two steps It is rapid: idle parking stall type identification and idle parking stall position identification.Specific embodiment are as follows:

Step 1 (idle parking stall type identification): according to the external frame central point line in the bottom of two neighboring vehicle and bottom The angle of external frame longitudinal direction determines parking stall classification；

Step 2 (idle parking stall position identification): using four vertex of the external frame in the bottom of two Adjacent vehicles in car body The coordinate on four vertex of the coordinate computation-free parking stall in coordinate system.

The embodiment of the present application is described respectively for above-mentioned two key step.

Step 1, idle parking stall type identification.

The type of idle parking stall is mainly according in the external frame in bottom of the external frame longitudinal direction of vehicle and two neighboring vehicle Heart point line angle judges.Specifically, please referring to Figure 12, it illustrates a kind of idle parking stall classes shown in the embodiment of the present application Type identification process figure.It is described in detail below with reference to flow chart shown in Figure 12:

(a) using the recognition result (i.e. the result of the acquisition of above-mentioned steps 204) of the external frame detection in the bottom of vehicle, phase is calculated The key message of the spatial position of adjacent vehicle, i.e., the sideline direction of the external frame in bottom of two neighboring vehicle and center point coordinate. Specifically, the external frame in the bottom of vehicle is a rectangle frame, description is indicated by the coordinate on four vertex.For example, in Figure 10 P1, P2, P3, P4 are 4 vertex of external frame, and sideline direction is divided into longitudinal direction and short side direction, by corresponding two sidelines Direction is averaged to obtain, by taking Figure 10 as an example, longitudinal direction can with the corresponding direction cut-off line P1P4 (can by straight line and lane it Between angle indicate) and straight line P2P3 corresponding direction average value, and short side direction can be with the corresponding side of cut-off line P1P2 To the average value of (can be indicated by the angle between straight line and lane) and the corresponding direction straight line P3P4.

(b) using the recognition confidence of the external frame in the bottom of two neighboring vehicle as weight, the bottom of two vehicles is calculated The weighted average of the respective longitudinal direction of the external frame in portion obtains the parking stall longitudinal direction of idle parking stall that may be present.

Wherein, the recognition confidence of the external frame in the bottom of vehicle can be through above-mentioned key point identification model to vehicle Two dimension surround frame in image carry out identification acquisition recognition result (i.e. the vehicle two dimension surround frame on three key points Corresponding first percentage, the second percentage and third percentage) be accurate result probability.

Specifically, second deep neural network identifies and exports a vehicle in the embodiment of the present application above-mentioned steps 203 Two dimension when surrounding the numerical value of three key points corresponding py1, px2 and py3 on frame, while also exporting these three numbers Value is the probability of exact value, i.e., above-mentioned recognition confidence.In the step (b), the side of idle parking stall that may be present is calculated When line direction, with the number of three key points corresponding py1, px2 and py3 on the external frame in the bottom of two neighboring vehicle The recognition confidence of value is weighted and averaged the bottom of the two neighboring vehicle respective longitudinal direction of external frame as weight, The parking stall longitudinal direction of the idle parking stall that may be present obtained.

In alternatively possible implementation, parking stall identifies that equipment can also be by the external frame in the bottom of two neighboring vehicle Longitudinal direction be determined as above-mentioned parking stall longitudinal direction.

(c) according to the parking stall longitudinal direction of idle parking stall that may be present, the parking stall of idle parking stall that may be present is calculated Angle α between the central point line of the external frame in the bottom of longitudinal direction and two neighboring vehicle, and it is true according to the size of angle α The parking stall type of fixed idle parking stall that may be present.Please refer to Figure 13, it illustrates the invention relates to a kind of parking stall Type schematic diagram.

Wherein, as shown in figure 13, when angle α is close to 90 °, i.e., the difference between angle α and 90 ° within a predetermined range when, The type that can determine the idle parking stall between adjacent two vehicle is vertical parking stall.

When angle α is close to 0 °, i.e., the difference between angle α and 0 ° within a predetermined range when, can determine adjacent two vehicle it Between idle parking stall type be parallel space.

When angle α is in section (30 °, 60 °) or interior (120 °, 150 °), the sky between adjacent two vehicle can be determined The type of not busy parking stall is oblique parking stall.

Optionally, if angle α is not belonging to any one of the above situation, it may be considered that the operation of this identification parking stall type Failure, the parking stall type for exporting idle parking stall that may be present is sky.

Step 2, idle parking stall position identification.

Please refer to Figure 14, it illustrates the invention relates to a kind of idle parking stall position identification process figure, such as scheme Shown in 14, the detailed step of idle parking stall position identification is described below:

(a) central point line etc. point interval t is calculated according to the parking stall type identified in above-mentioned steps 1；

Wherein, for vertical parking stall, t=w；For parallel space, t=h, for oblique parking stall, t=(w/sin α).

Optionally, if the recognition result of above-mentioned parking stall type is sky, it may be considered that the bottom of above-mentioned two neighboring vehicle There is no idle parking stalls between external frame.

(b) the length d of the central point line of the external frame in the bottom of two neighboring vehicle obtains n divided by equal part interval t, wherein N is the maximum integer less than or equal to d/t.Whether carried out containing available free parking stall according to transfers between divisions between two neighboring vehicle Judgement:

As n≤1, there is no idle parking stalls between two neighboring vehicle；

As n > 1, there are n-1 idle parking stall between two neighboring vehicle, corresponding Along ent is idle parking stall Central point.

Please refer to Figure 15 and Figure 16, wherein Figure 15 shows between a kind of two neighboring vehicle that there is no idle parking stalls Schematic diagram, Figure 16 show that it illustrates the schematic diagrames that there is idle parking stall between two neighboring vehicle.In Figure 15, for hanging down Straight turning position, n=d/w=1, therefore judge that there is no idle parking stalls, in Figure 16, for vertical parking stall, n=d/w=3, therefore There is no two idle parking stalls for judgement.

(c) for there is idle parking stall situation, please refer to Figure 17, it illustrates the invention relates to a kind of determination Pinpoint the schematic diagram of coordinate, as shown in figure 17, from the central point P0 of idle parking stall, along idle parking stall longitudinal direction and Short side direction deviates the distance of h/2, w/2 respectively, obtains the coordinate on idle tetra- vertex parking stall P1, P2, P3, P4.

Optionally, the above-mentioned coordinate according to parking stall longitudinal direction and n-1 Along ent in bodywork reference frame, determines n- When coordinate of the corresponding idle parking stall of 1 Along ent in bodywork reference frame, parking stall identifies that equipment can be for n-1 etc. Any Along ent i, corresponding positive and negative along parking stall longitudinal direction since coordinate of the Along ent i in bodywork reference frame in branch Both direction moves h/2 respectively, obtains the midpoint of two short sides of the corresponding idle parking stall Along ent i in bodywork reference frame Coordinate；Since coordinate of the midpoint of two short sides of the corresponding idle parking stall Along ent i in bodywork reference frame, along Along ent The corresponding positive and negative both direction of short side direction of the corresponding idle parking stall i moves w/2 respectively, obtains the Along ent i corresponding free time Coordinate of four vertex of parking stall in bodywork reference frame, the short side direction and parking stall long side of the corresponding idle parking stall Along ent i Direction is vertical.Alternatively, parking stall identification equipment can also be for Along ent i any in n-1 Along ent, from Along ent i in car body Coordinate in coordinate system starts, and the corresponding positive and negative both direction of short side direction along the corresponding idle parking stall Along ent i is moved respectively Dynamic w/2 obtains coordinate of the midpoint of two long sides of the corresponding idle parking stall Along ent i in bodywork reference frame；From Along ent i Coordinate of the midpoint of two long sides of corresponding free time parking stall in bodywork reference frame starts, corresponding just along parking stall longitudinal direction Anti- both direction moves h/2 respectively, obtains seat of four vertex of the corresponding idle parking stall Along ent i in bodywork reference frame Mark.

By taking Figure 17 as an example, in practical applications, parking stall identifies that equipment can be first vertically square respectively from central point P0 H/2 is deviated to (i.e. the positive and negative both direction of parking stall longitudinal direction), reaches the midpoint of two short sides of idle parking stall, then from two (i.e. the positive and negative both direction of parking stall short side direction) deviates w/2 in left-right direction respectively at the midpoint of short side, reaches idle parking stall Four vertex, and obtain the coordinate in bodywork reference frame on four vertex.

Alternatively, identification equipment in parking stall can also first deviate w/2 from central point P0 in left-right direction respectively, reach empty The midpoint of two long sides of not busy parking stall, then h/2 is deviated along the vertical direction respectively from the midpoint of two long sides, reach idle parking stall Four vertex, and obtain coordinate of four vertex in bodywork reference frame.

(d) the coordinate recognition result on four vertex of final output can be sent to vehicle control module, so as to vehicle control Module carries out motion control, such as automatic parking etc. to vehicle.The output format of the coordinate recognition result on four vertex can be with For 8 dimensional vectors, each element is respectively the X-axis and the Y axis coordinate (Z on four vertex on four vertex in 8 dimensional vector 0) axial coordinate is.

In conclusion method shown in the embodiment of the present application, by identifying the two neighboring vehicle in two-dimensional vehicle image Two dimension surround frame, and according to two neighboring vehicle two dimension surround frame determine two neighboring vehicle the external frame in bottom three Coordinate of a vertex in bodywork reference frame, and determined according to coordinate of three vertex of the external frame in bottom in bodywork reference frame Idle parking stall between two neighboring vehicle, it is only necessary to there are adjacent two that can recognize that two dimension surrounds frame in vehicle image A vehicle, it can it determines free time parking stall that may be present between the two vehicles, does not need the parking stall line on identification ground, Even without on ground, there are parking stall lines, higher to the recognition accuracy of idle parking stall；In addition, shown in the embodiment of the present application Method, it is only necessary to collect the vehicle image for containing at least two Adjacent vehicles, it can realize the identification of idle parking stall, thus It can identify the idle parking stall between remote vehicle, remote to the identification distance of idle parking stall, recognition effect is more preferable.

Step 203 in above-mentioned embodiment illustrated in fig. 2 is only external with the respective bottom of two neighboring vehicle to step 205 At least one feature locations point on frame is the external frame in the respective bottom of two neighboring vehicle visible three in vehicle image It is illustrated for vertex, in practical applications, at least one on the external frame in the above-mentioned two neighboring respective bottom of vehicle is special Sign location point is not restricted to the external frame in above-mentioned bottom visible three vertex in vehicle image.

For example, at least one above-mentioned feature locations point may include vehicle the external frame in bottom it is visible in vehicle image At least one other feature locations point except three vertex and three vertex, with the location information of the idle parking stall of raising Identify accuracy.For example, at least one other feature locations point except three vertex may include the license plate center of vehicle Point, vehicle the external frame in bottom the external frame in the bottom at upper midpoint, vehicle while on predetermined ratio Along ent or vehicle The location point of the easy discrimination such as four angle points of body.

Alternatively, at least one above-mentioned feature locations point can also only include one or two location points, for example, it is above-mentioned extremely A few feature locations point may include the external frame in bottom of vehicle one or two visible vertex in vehicle image, or Person, at least one above-mentioned feature locations point also may include the license plate central point of vehicle, vehicle the external frame in bottom side on Midpoint, vehicle the external frame in bottom side on predetermined ratio Along ent or vehicle body the easy discrimination such as four angle points position One or two location points in point.Wherein, two neighboring vehicle can be substantially determined by one or two location points Between whether there is parking stall.For example, when at least one feature locations point only includes the external frame in bottom of vehicle in vehicle image When a visible vertex, parking stall identification equipment can calculate the external frame in the respective bottom of two neighboring vehicle in vehicle image Distance d1 of the visible vertex in bodywork reference frame, and it is big between d1 and preset distance threshold d0 according to this distance Small relationship is determined with the presence or absence of idle parking stall between the two neighboring vehicle, for example, parking stall is known when d1 is more than or equal to d0 Other equipment, which can determine between the two neighboring vehicle, has idle parking stall, wherein the value of d0 can be set to 3h.Alternatively, working as In vehicle image when visible two vertex, parking stall identification is set the external frame in bottom of at least one feature locations point comprising vehicle It is standby to calculate the midpoint on the external frame in the respective bottom of two neighboring vehicle visible two vertex in vehicle image in car body Distance d2 in coordinate system, and the size relation between d2 and preset distance threshold d0 determines that this is two neighboring according to this distance With the presence or absence of idle parking stall between vehicle.

In alternatively possible implementation, parking stall identifies that equipment can also be according to the corresponding sky in the space comprising vehicle Between perception data identify parking stall, i.e., parking stall identification equipment can be according to vehicle each in spatial perception data in bodywork reference frame In coordinate, determine coordinate of the external frame in the bottom of each vehicle on the ground in bodywork reference frame, and according to two neighboring The external frame in the bottom of vehicle on the ground determines the idle parking stall between two neighboring vehicle in the coordinate in bodywork reference frame.

Figure 18 is please referred to, it illustrates the methods for determining idle parking stall of the exemplary embodiment of the application offer Flow chart.The parking stall that this method can be used for system shown in Figure 1 determines in equipment 120.As shown in figure 18, the determination free time parking stall Method may include:

Step 1801, the spatial perception data for containing at least two vehicle are obtained.

In the embodiment of the present application, above-mentioned spatial perception data can be radar point cloud data or laser point cloud data. It, can basis after parking stall identification equipment receives the radar point cloud data or laser point cloud data of data acquisition equipment transmission Radar point cloud data or laser point cloud data judge with the presence or absence of at least two vehicles in corresponding space, if so, determining Get the spatial perception data.

For example, the spatial perceptions such as radar point cloud data or laser point cloud data data usually by radar detecting equipment or Each space coordinate point composition that laser detection equipment detection obtains, these space coordinate points constitute the wheel of the object in space The profile of exterior feature, each object that identification equipment in parking stall can be constituted according to spatial perception data determines whether corresponding object is vehicle ?.

Step 1802, according to the spatial perception data, the two neighboring respective bottom of vehicle at least two vehicle is obtained The location information of at least one feature locations point on the external frame in portion.

In the embodiment of the present application, after identification equipment in parking stall gets spatial perception data, spatial perception data pair are determined In the space answered after the profile of each vehicle, the orthographic projection of each vehicle on the ground is determined according to the profile of each vehicle, And at least one feature locations point on the external frame in bottom of each vehicle is determined according to the orthographic projection of each vehicle on the ground Location information, the external frame in the bottom of a vehicle be the orthographic projection of the profile comprising the vehicle on the ground rectangle frame.

Step 1803, believed according to the position of at least one feature locations point on the external frame in bottom of the two neighboring vehicle Breath, determines the idle parking stall between the two neighboring vehicle.

The implementation procedure of the step 1803 is similar with the step 205 in above-mentioned embodiment illustrated in fig. 2, and details are not described herein again.

In the embodiment of the present application, data acquisition equipment can acquire the spatial perception data of vehicle periphery, and by vehicle The spatial perception data of surrounding are sent to parking stall and determine equipment, and parking stall determines that equipment receives the collected sky of image capture device Between after perception data, at least one passed through on the external frame in bottom of the two neighboring vehicle in identification spatial perception data is special The location information of location point is levied, and determines two according to the location information of at least one feature locations point on the external frame in bottom Idle parking stall between vehicle, it is only necessary to there is adjacent two vehicle that can recognize that the external frame in bottom in spatial perception data , it can the idle parking stall between the two vehicles is determined, to improve the recognition effect of idle parking stall.

Figure 19 is the structural schematic diagram that the parking stall that one exemplary embodiment of the application provides determines equipment 1900, the parking stall Determine that equipment 1900 can be implemented as the parking stall in system shown in FIG. 1 and determine equipment 120.As shown in figure 19, which determines Equipment 1900 may include: processor 1901 and memory 1903.

Processor 1901 may include one or more than one processing unit, which can be central processing list Member (English: central processing unit, CPU) or network processing unit (English: network processor, NP) Deng.

Memory 1903 can be used for programmable instructions, which can be called by processor 1901.In addition, this is deposited All kinds of business datums or user data can also be stored in reservoir 1903.The programmable instructions may include obtaining module, packet Peripheral frame obtains module, position determination module and parking stall determining module；

Wherein, it obtains module to be called by processor 1901, obtains vehicle as related in Fig. 2 or embodiment illustrated in fig. 18 to execute The function of data information (i.e. vehicle image or spatial perception data).

It surrounds frame determining module to be called by processor 1901, to execute related determining vehicle in embodiment as shown in Figure 2 Two dimension surrounds the function of frame.

Position determination module is called by processor 1901, two neighboring to execute related determination in embodiment as shown in Figure 2 The function of the location information of at least one feature locations point on the external frame in the respective bottom of vehicle.

Parking stall determining module is called by processor 1901, determines phase as related in Fig. 2 or embodiment illustrated in fig. 18 to execute The function of idle parking stall between adjacent two vehicles.

Optionally, which determines that equipment 1900 can also include communication interface 1904, which can wrap Include network interface.Wherein, the network interface is for connecting image capture device, and optionally, which also can connect it Its equipment.Specifically, the network interface may include wired network interface, such as Ethernet interface or optical fiber interface, alternatively, Network interface also may include radio network interface, such as wireless lan interfaces or cellular mobile network interface.Parking stall is true Locking equipment 1900 is communicated by the network interface 1904 with image capture device or other equipment.

Optionally, processor 1901 can be connected with bus with memory 1903 and communication interface 1904.

Optionally, which determines that equipment 1900 can also include output equipment 1905 and input equipment 1907.Output Equipment 1905 and input equipment 1907 are connected with processor 1901.Output equipment 1905 can be the display for showing information Device, the power amplifying device for playing sound etc., output equipment 1905 can also include o controller, to provide output to display Screen, power amplifying device etc..Input equipment 1907 can be for user input such as mouse of information, keyboard, electronic touch pen or The equipment of person's touch panel etc, input equipment 1907 can also include o controller for receiving and handling from mouse The input of the equipment such as mark, keyboard, electronic touch pen or touch panel.

Figure 20 is the structural schematic diagram for the model training equipment 2000 that one exemplary embodiment of the application provides.Such as Figure 20 Shown, which may include: processor 2001 and memory 2003.

Processor 2001 may include one or more than one processing unit, which can be central processing list Member (English: central processing unit, CPU) or network processing unit (English: network processor, NP) Deng.

Memory 2003 can be used for programmable instructions, which can be called by processor 2001.In addition, this is deposited All kinds of business datums or user data can also be stored in reservoir 2003.The programmable instructions may include obtain module and Training module；

Wherein, it obtains module to be called by processor 2001, to execute second sample image of related acquisition as shown in FIG. 6 Function.

Training module is called by processor 2001, to execute the function as shown in FIG. 6 in relation to training key point identification model Energy.

Optionally, which can also include communication interface 2004, which can wrap Include network interface.The network interface can connect other equipment.Specifically, the network interface may include wired network interface, Such as Ethernet interface or optical fiber interface, alternatively, network interface also may include radio network interface, such as WLAN Interface or cellular mobile network interface.

Optionally, processor 2001 can be connected with bus with memory 2003 and communication interface 2004.

Optionally, which can also include output equipment 2005 and input equipment 2007.Output Equipment 2005 and input equipment 2007 are connected with processor 2001.Output equipment 2005 can be the display for showing information Device, the power amplifying device for playing sound etc., output equipment 2005 can also include o controller, to provide output to display Screen, power amplifying device etc..Input equipment 2007 can be for user input such as mouse of information, keyboard, electronic touch pen or The equipment of person's touch panel etc, input equipment 2007 can also include o controller for receiving and handling from mouse The input of the equipment such as mark, keyboard, electronic touch pen or touch panel.

Figure 21 is a kind of structure chart of the device for determining idle parking stall that the exemplary embodiment of the application provides, should Determine that the device of free time parking stall can determine the portion of equipment as parking stall by being implemented in combination with for hardware circuit or hardware and software Divide or all, the parking stall which determines that equipment can be in above-mentioned embodiment shown in FIG. 1 determines equipment 120.The determination The device of idle parking stall may include: acquiring unit 2101, surround frame determination unit 2102, position determination unit 2103 and vehicle Position determination unit 2104；

The acquiring unit 2101 can be used to implement such as the determining idle parking stall of Fig. 2 or embodiment illustrated in fig. 18 offer In method, the function in relation to obtaining vehicle data information (i.e. vehicle image or spatial perception data).

The encirclement frame determination unit 2102 can be used to implement the determining idle parking stall that embodiment as shown in Figure 2 provides In method, in relation to determining that the two dimension of vehicle surrounds the function of frame.

The position determination unit 2103 can be used to implement the side for determining idle parking stall that embodiment as shown in Figure 2 provides In method, the function of the related location information for determining at least one feature locations point on the external frame in the two neighboring respective bottom of vehicle Energy.

The parking stall determination unit 2104 can be used to implement the determining idle vehicle such as Fig. 2 or embodiment illustrated in fig. 18 offer In the method for position, the function in relation to determining the idle parking stall between two neighboring vehicle.

Figure 22 is a kind of structure chart for model training apparatus that the exemplary embodiment of the application provides, model instruction Practicing device being implemented in combination with as some or all of of model training equipment by hardware circuit or hardware and software.It should Model training apparatus may include: acquiring unit 2201 and training unit 2202；

The acquiring unit 2201, can be used to implement in model training method as shown in FIG. 6, related to obtain the second sample The function of image.

The training unit 2202, can be used to implement in model training method as shown in FIG. 6, and related training key point is known The function of other model.

It should be understood that the device of determining idle parking stall provided by the above embodiment is when identifying idle parking stall, and Model training apparatus is in training pattern, can be with only with the division progress of above-mentioned each functional unit for example, in practical application It is completed as needed and by above-mentioned function distribution by different functional units, i.e., the internal structure of device is divided into different function Energy unit, to complete all or part of the functions described above.In addition, the dress of determining idle parking stall provided by the above embodiment It sets and belongs to same design, and model training apparatus and model training method embodiment category with the embodiment of the method for idle parking stall is determined In same design, specific implementation process is detailed in embodiment of the method, and which is not described herein again.

The embodiment of the present application also provides a kind of computer program, and it is programmable to may include at least one in the computer program Instruction, when computer program is run in a computer, computer can be by executing at least one programmable instructions with reality The now all or part of step of the method as shown in above-mentioned Fig. 2, Fig. 6 or Figure 18.

Above-mentioned embodiments herein serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.

Those of ordinary skill in the art will appreciate that realizing all or part of the steps that above-described embodiment is executed by processor It may be implemented by hardware, relevant hardware can also be controlled by programmable instructions and is completed, the programmable instructions It can store in a kind of computer readable storage medium, i.e., processor calls what is stored in computer readable storage medium to compile Cheng Zhiling, so that the processor is executed as determined idle parking stall method in above-mentioned Fig. 2 or embodiment illustrated in fig. 18, alternatively, making It obtains the processor and executes model training method as shown in Figure 6 above, computer readable storage medium mentioned above can be Read-only memory, disk or CD etc..

A kind of specific embodiment that the above, only the application can be realized, but the protection scope of the application is not Be confined to this, anyone skilled in the art within the technical scope of the present application, using claim as base Plinth is changed or replaces obtained scheme, should all cover within the scope of protection of this application.Therefore, the protection of the application Range should be subject to the protection scope in claims.

Claims (20)

1. a kind of method of determining idle parking stall, which is characterized in that the described method includes:

Obtain the vehicle image for containing at least two vehicle；

Determine at least two vehicle that the respective two dimension of two neighboring vehicle surrounds frame according to the vehicle image, described two It is the boundary rectangle frame that vehicle is corresponded in the vehicle image that dimension, which surrounds frame,；

Frame is surrounded according to the two neighboring respective two dimension of vehicle, determines the external frame in the respective bottom of two neighboring vehicle On at least one feature locations point location information, the external frame in bottom is the orthographic projection of corresponding vehicle on the ground Boundary rectangle frame, the feature locations point are used to indicate the point on the predeterminated position of the corresponding external frame in bottom；

According to the location information of at least one feature locations point on the external frame in bottom of the two neighboring vehicle, determine described in Idle parking stall between two neighboring vehicle.

2. the method according to claim 1, wherein at least one described feature locations point includes corresponding bottom Three vertex of external frame, three vertex are used to indicate three tops of the visible external frame in bottom in the vehicle image Point, the location information are used to indicate the coordinate in bodywork reference frame；

Wherein, described that frame is surrounded according to the two neighboring respective two dimension of vehicle, determine that the two neighboring vehicle is respective The location information of at least one feature locations point on the external frame in bottom, comprising:

Frame is surrounded according to the two neighboring respective two dimension of vehicle, determines the external frame in the respective bottom of two neighboring vehicle Coordinate of three vertex in the bodywork reference frame；

The location information of at least one feature locations point on the external frame in bottom according to the two neighboring vehicle determines Idle parking stall between the two neighboring vehicle, comprising:

According to coordinate of three vertex of the external frame in bottom of the two neighboring vehicle in the bodywork reference frame, institute is determined State coordinate of the idle parking stall in the bodywork reference frame between two neighboring vehicle.

3. according to the method described in claim 2, it is characterized in that, described wrap according to the two neighboring respective two dimension of vehicle Peripheral frame determines coordinate of three vertex of the external frame in the respective bottom of two neighboring vehicle in the bodywork reference frame, Include:

For any vehicle i in the two neighboring vehicle, determine that the two dimension of the vehicle i is surrounded by key point identification model The relative position of three key points on frame, three key points are three vertex with the external frame in bottom of the vehicle i Corresponding three characteristic points；The key point identification model is the identification model obtained previously according to sample image training, described Sample image is the image for being labeled with three key points on two dimension encirclement frame in advance；

The relative position that three key points on frame are surrounded according to the two dimension of the vehicle i, the two dimension for calculating the vehicle i are surrounded Two-dimensional coordinate of three key points in the coordinate system of the vehicle image on frame；

Two dimension of three key points surrounded on frame according to the two dimension of the vehicle i in the coordinate system of the vehicle image is sat Mark, calculates coordinate of three vertex of the external frame in bottom of the vehicle i in the bodywork reference frame.

4. according to the method described in claim 3, it is characterized in that, the two dimension of the vehicle i surrounds three key points on frame Relative position include: the first key point relative to the vehicle i two dimension surround frame bottom left vertex offset and the vehicle The first percentage that the two dimension of i is surrounded between the line length of the frame left side, the second key point are surrounded relative to the two dimension of the vehicle i The second percentage and third that the offset of frame bottom left vertex and the two dimension of the vehicle i are surrounded between the following line length of frame close Key point surrounds the offset of frame bottom right vertex and the two dimension encirclement frame right side bearing of the vehicle i relative to the two dimension of the vehicle i Third percentage between length, wherein first key point is that the two dimension of the vehicle i surrounds the feature on frame left side bearing Point, second key point are that the two dimension of the vehicle i surrounds the characteristic point on frame lower sideline, and the third key point is described The two dimension of vehicle i surrounds the characteristic point on frame right side bearing；

Wherein, the relative position that three key points on frame are surrounded according to the two dimension of the vehicle i, calculates the vehicle i Two dimension surround two-dimensional coordinate of three key points in the coordinate system of the vehicle image on frame, comprising:

Frame is surrounded according to the two dimension of first percentage, second percentage, the third percentage and the vehicle i Coordinate of four vertex in the coordinate system of the vehicle image, calculate first key point, second key point with And the two-dimensional coordinate in the coordinate system of each leisure vehicle image of third key point.

5. the method according to claim 3 or 4, which is characterized in that described to be surrounded on frame according to the two dimension of the vehicle i Two-dimensional coordinate of three key points in the coordinate system of the vehicle image calculates three of the external frame in bottom of the vehicle i Coordinate of the vertex in the bodywork reference frame, comprising:

The internal reference matrix and outer ginseng matrix of image capture device are obtained, described image acquisition equipment is the acquisition vehicle image Equipment；

Two dimension of three key points surrounded on frame according to the two dimension of the vehicle i in the coordinate system of the vehicle image is sat Mark and the internal reference matrix and the outer ginseng matrix, calculate three vertex of the external frame in bottom of the vehicle i in the vehicle Coordinate in body coordinate system.

6. according to any method of claim 2 to 5, which is characterized in that the bottom according to the two neighboring vehicle Coordinate of three vertex of the external frame in portion in the bodywork reference frame, determines the idle parking stall between the two neighboring vehicle Coordinate in the bodywork reference frame, comprising:

According to coordinate of three vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame, meter Calculate the 4th coordinate of the vertex in the bodywork reference frame of the external frame in the respective bottom of two neighboring vehicle；

According to coordinate of four vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame, meter Calculate coordinate of the central point of the external frame in the respective bottom of two neighboring vehicle in the bodywork reference frame；

Parking stall type is determined according to the angle between parking stall longitudinal direction and central point line, and the parking stall longitudinal direction is basis Direction determined by the longitudinal direction of the external frame in bottom of at least one vehicle in the two neighboring vehicle, the central point connect Line be the two neighboring vehicle the external frame in bottom central point between line, the parking stall type include vertical parking stall, Parallel space or oblique parking stall；

Determine that parking stall quantity n, the parking stall quantity n are the parking stall numbers between the two neighboring vehicle according to the parking stall type Amount；

When the n is the integer more than or equal to 2, determine n-1 Along ent on the central point line in the vehicle The central point line is divided into n equal portions by the coordinate in body coordinate system, the n-1 Along ent；

According to the coordinate of the parking stall longitudinal direction and the n-1 Along ent in the bodywork reference frame, determine described in Coordinate of the corresponding idle parking stall of n-1 Along ent in the bodywork reference frame.

7. according to the method described in claim 6, it is characterized in that, described according to the parking stall longitudinal direction and the n-1 Coordinate of a Along ent in the bodywork reference frame determines the corresponding idle parking stall of the n-1 Along ent described Coordinate in bodywork reference frame, comprising:

For Along ent i any in the n-1 Along ent, opened from coordinate of the Along ent i in the bodywork reference frame Begin, moves h/2 respectively along the corresponding positive and negative both direction of the parking stall longitudinal direction, obtain the Along ent i corresponding free time Coordinate of the midpoint of two short sides of parking stall in the bodywork reference frame；From the two of the corresponding idle parking stall the Along ent i Coordinate of the midpoint of short side in the bodywork reference frame starts, along the short side side of the corresponding idle parking stall the Along ent i W/2 is moved respectively to corresponding positive and negative both direction, obtains four vertex of the corresponding idle parking stall the Along ent i described Coordinate in bodywork reference frame；Wherein, h is default parking stall length, and w is default parking stall width, the Along ent i corresponding free time The short side direction of parking stall is vertical with the parking stall longitudinal direction；

Alternatively,

For Along ent i any in the n-1 Along ent, opened from coordinate of the Along ent i in the bodywork reference frame Begin, the corresponding positive and negative both direction of short side direction along the corresponding idle parking stall the Along ent i moves w/2 respectively, obtains institute State coordinate of the midpoint of two long sides of the corresponding idle parking stall Along ent i in the bodywork reference frame；From the Along ent i Coordinate of the midpoint of two long sides of corresponding free time parking stall in the bodywork reference frame starts, along the parking stall longitudinal direction Corresponding positive and negative both direction moves h/2 respectively, obtains four vertex of the corresponding idle parking stall the Along ent i in the vehicle Coordinate in body coordinate system；Wherein, h is default parking stall length, and w is default parking stall width, the corresponding idle vehicle of the Along ent i The short side direction of position is vertical with the parking stall longitudinal direction.

8. method according to claim 6 or 7, which is characterized in that described according to parking stall longitudinal direction and the central point Angle between line determines parking stall type, comprising:

When the angle is 90 °, determine that the parking stall type is vertical parking stall；

When the angle is 0 °, determine that the parking stall type is parallel space；

When the angle is in (30 °, 60 °) or (120 °, 150 °) section, determine that the parking stall type is oblique vehicle Position.

9. according to any method of claim 6 to 8, which is characterized in that described to determine parking stall according to the parking stall type Quantity n, comprising:

When the parking stall type is vertical parking stall, determine that n is the maximum integer less than or equal to d/w；

When the parking stall type is parallel space, determine that n is the maximum integer less than or equal to d/h；

When the parking stall type is oblique parking stall, determine that n is the maximum integer less than or equal to d/ (w/sin α)；

Wherein, d is the length of the central point line, and w is default parking stall width, and h is default parking stall length, and α is the angle.

10. a kind of device of determining idle parking stall, which is characterized in that described device includes:

Acquiring unit, for obtaining the vehicle image for containing at least two vehicle；

Frame determination unit is surrounded, for determining that two neighboring vehicle is respectively at least two vehicle according to the vehicle image Two dimension surround frame, it is described two dimension surround frame be the boundary rectangle frame that vehicle is corresponded in the vehicle image；

Position determination unit determines the two neighboring vehicle for surrounding frame according to the two neighboring respective two dimension of vehicle The location information of at least one feature locations point on the respective external frame in bottom, the external frame in bottom is corresponding vehicle The boundary rectangle frame of orthographic projection on the ground, the feature locations point are used to indicate on the predeterminated position of the external frame in bottom Point；

Parking stall determination unit, at least one feature locations point on the external frame in bottom according to the two neighboring vehicle Location information determines the idle parking stall between the two neighboring vehicle.

11. device according to claim 10, which is characterized in that at least one described feature locations point includes corresponding bottom Three vertex of the external frame in portion, three vertex are used to indicate three tops of the visible external frame in bottom in the vehicle image Point, the location information are used to indicate the coordinate in bodywork reference frame；

The position determination unit determines described adjacent two specifically for surrounding frame according to the two dimension of the two neighboring vehicle Coordinate of three vertex of the external frame in the bottom of a vehicle in the bodywork reference frame；

The parking stall determination unit, specifically for three vertex according to the external frame in bottom of the two neighboring vehicle described Coordinate in bodywork reference frame determines seat of the idle parking stall between the two neighboring vehicle in the bodywork reference frame Mark.

12. device according to claim 11, which is characterized in that the position determination unit is specifically used for,

For any vehicle i in the two neighboring vehicle, determine that the two dimension of the vehicle i is wrapped by key point identification model The relative position of three key points in peripheral frame；Three key points are three tops with the external frame in bottom of the vehicle i Corresponding three characteristic points of point；The key point identification model is the identification for carrying out machine training previously according to sample image and obtaining Model, the sample image are the images for being labeled with three key points on two dimension encirclement frame in advance；

The relative position that three key points on frame are surrounded according to the two dimension of the vehicle i, the two dimension for calculating the vehicle i are surrounded Two-dimensional coordinate of three key points in the coordinate system of the vehicle image on frame；

Two dimension of three key points surrounded on frame according to the two dimension of the vehicle i in the coordinate system of the vehicle image is sat Mark, calculates coordinate of three vertex of the external frame in bottom of the vehicle i in the bodywork reference frame.

13. device according to claim 12, which is characterized in that the two dimension of the vehicle i surrounds three keys on frame The relative position of point include: the first key point relative to the vehicle i two dimension surround the offset of frame bottom left vertex with it is described The two dimension of vehicle i surrounds the two dimension packet of the first percentage, the second key point relative to the vehicle i between the line length of the frame left side The offset of peripheral frame bottom left vertex and the two dimension of the vehicle i surround the second percentage and third between the following line length of frame Key point is surrounded relative to the two dimension of the vehicle i on the right of the offset of frame bottom right vertex and the two dimension encirclement frame of the vehicle i Third percentage between line length, wherein first key point is that the two dimension of the vehicle i surrounds the spy on frame left side bearing Point is levied, second key point is that the two dimension of the vehicle i surrounds the characteristic point on frame lower sideline, and the third key point is institute State the characteristic point on the two dimension encirclement frame right side bearing of vehicle i；

The relative position of three key points on frame is surrounded in the two dimension according to the vehicle i, calculates the two dimension packet of the vehicle i When two-dimensional coordinate in the coordinate system of the vehicle image of three key points in peripheral frame, the position determination unit, specifically For,

Frame is surrounded according to the two dimension of first percentage, second percentage, the third percentage and the vehicle i Coordinate of four vertex in the coordinate system of the vehicle image, calculate first key point, second key point with And the two-dimensional coordinate in the coordinate system of each leisure vehicle image of third key point.

14. device according to claim 12 or 13, which is characterized in that surrounded on frame according to the two dimension of the vehicle i Two-dimensional coordinate of three key points in the coordinate system of the vehicle image, calculate the three of the external frame in bottom of the vehicle i When coordinate of a vertex in the bodywork reference frame, the position determination unit is specifically used for,

The internal reference matrix and outer ginseng matrix of image capture device are obtained, described image acquisition equipment is the acquisition vehicle image Equipment；

Two dimension of three key points surrounded on frame according to the two dimension of the vehicle i in the coordinate system of the vehicle image is sat Mark and the internal reference matrix and the outer ginseng matrix, calculate three vertex of the external frame in bottom of the vehicle i in the vehicle Coordinate in body coordinate system.

15. 1 to 14 any device according to claim 1, which is characterized in that the parking stall determination unit is specifically used for,

According to coordinate of three vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame, meter Calculate the 4th coordinate of the vertex in the bodywork reference frame of the external frame in the respective bottom of two neighboring vehicle；

According to coordinate of four vertex of the external frame in the two neighboring respective bottom of vehicle in the bodywork reference frame, meter Calculate coordinate of the central point of the external frame in the respective bottom of two neighboring vehicle in the bodywork reference frame；

Parking stall type is determined according to the angle between parking stall longitudinal direction and central point line, and the parking stall longitudinal direction is basis Direction determined by the longitudinal direction of the external frame in bottom of at least one vehicle in the two neighboring vehicle, the central point connect Line be the two neighboring vehicle the external frame in bottom central point between line, the parking stall type include vertical parking stall, Parallel space or oblique parking stall；

Determine that parking stall quantity n, the parking stall quantity n are the parking stall numbers between the two neighboring vehicle according to the parking stall type Amount；

When the n is the integer more than or equal to 2, determine n-1 Along ent on the central point line in the vehicle The central point line is divided into n equal portions by the coordinate in body coordinate system, the n-1 Along ent；

According to the coordinate of the parking stall longitudinal direction and the n-1 Along ent in the bodywork reference frame, determine described in Coordinate of the corresponding idle parking stall of n-1 Along ent in the bodywork reference frame.

16. device according to claim 15, which is characterized in that according to the parking stall longitudinal direction and the n-1 Coordinate of a Along ent in the bodywork reference frame determines the corresponding idle parking stall of the n-1 Along ent described When coordinate in bodywork reference frame, the parking stall determination unit is specifically used for

For Along ent i any in the n-1 Along ent, opened from coordinate of the Along ent i in the bodywork reference frame Begin, moves h/2 respectively along the corresponding positive and negative both direction of the parking stall longitudinal direction, obtain the Along ent i corresponding free time Coordinate of the midpoint of two short sides of parking stall in the bodywork reference frame；From the two of the corresponding idle parking stall the Along ent i Coordinate of the midpoint of short side in the bodywork reference frame starts, along the short side side of the corresponding idle parking stall the Along ent i W/2 is moved respectively to corresponding positive and negative both direction, obtains four vertex of the corresponding idle parking stall the Along ent i described Coordinate in bodywork reference frame；Wherein, h is default parking stall length, and w is default parking stall width, the Along ent i corresponding free time The short side direction of parking stall is vertical with the parking stall longitudinal direction；

Alternatively,

For Along ent i any in the n-1 Along ent, opened from coordinate of the Along ent i in the bodywork reference frame Begin, the corresponding positive and negative both direction of short side direction along the corresponding idle parking stall the Along ent i moves w/2 respectively, obtains institute State coordinate of the midpoint of two long sides of the corresponding idle parking stall Along ent i in the bodywork reference frame；From the Along ent i Coordinate of the midpoint of two long sides of corresponding free time parking stall in the bodywork reference frame starts, along the parking stall longitudinal direction Corresponding positive and negative both direction moves h/2 respectively, obtains four vertex of the corresponding idle parking stall the Along ent i in the vehicle Coordinate in body coordinate system；Wherein, h is default parking stall length, and w is default parking stall width, the corresponding idle vehicle of the Along ent i The short side direction of position is vertical with the parking stall longitudinal direction.

17. device according to claim 15 or 16, which is characterized in that according to parking stall longitudinal direction and the central point When angle between line determines parking stall type, the parking stall determination unit is specifically used for,

When the angle is 90 °, determine that the parking stall type is vertical parking stall；

When the angle is 0 °, determine that the parking stall type is parallel space；

When the angle is in (30 °, 60 °) or (120 °, 150 °) section, determine that the parking stall type is oblique vehicle Position.

18. 5 to 17 any device according to claim 1, which is characterized in that determining parking stall according to the parking stall type When quantity n, the parking stall determination unit is specifically used for,

When the parking stall type is vertical parking stall, determine that n is the maximum integer less than or equal to d/w；

When the parking stall type is parallel space, determine that n is the maximum integer less than or equal to d/h；

When the parking stall type is oblique parking stall, determine that n is the maximum integer less than or equal to d/ (w/sin α)；

Wherein, d is the length of the central point line, and w is default parking stall width, and h is default parking stall length, and α is the angle.

19. a kind of parking stall determines equipment, which is characterized in that the parking stall determines that equipment includes: processor and memory, described to deposit Reservoir is for storing programmable instructions；

The processor, for calling the programmable instructions stored on the reservoir to execute such as any institute of claim 1 to 9 The method for determining idle parking stall stated.

20. a kind of computer readable storage medium, which is characterized in that refer in the computer readable storage medium comprising programmable It enables, computer executes the programmable instructions and the computer is made to realize the determining free time as described in any one of claim 1 to 9 The method of parking stall.