CN114359839A - Method and system for identifying entrance of electric vehicle into elevator - Google Patents
Method and system for identifying entrance of electric vehicle into elevator Download PDFInfo
- Publication number
- CN114359839A CN114359839A CN202210084354.3A CN202210084354A CN114359839A CN 114359839 A CN114359839 A CN 114359839A CN 202210084354 A CN202210084354 A CN 202210084354A CN 114359839 A CN114359839 A CN 114359839A
- Authority
- CN
- China
- Prior art keywords
- information
- electric vehicle
- analysis
- elevator
- analysis result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Abstract
The invention discloses a method and a system for identifying an electric vehicle entering an elevator, belonging to the field of image identification.A plurality of pictures are compiled along a time sequence by obtaining video data of a target object within preset time; calculating the possibility that the electric vehicle exists in the picture, and calculating the possibility that the electric vehicle enters the elevator through a weighted average; the automatic identification can be realized in the process that the electric vehicle enters the elevator, and the manpower is saved and the identification effect is high. For the condition that a user modifies the appearance of the electric vehicle or carries large-volume objects, image information in different images is spliced and combined into a new combined image through plane projection transformation, the possibility of the electric vehicle existing in the combined image is calculated, and the accuracy of electric vehicle identification is further improved.
Description
Technical Field
The invention belongs to the field of image recognition, and particularly relates to a method and a system for recognizing an electric vehicle entering an elevator.
Background
With the popularization of electric bicycles, the problem of community management of electric bicycles is increasingly prominent. Mainly embody in aspects such as the security of charging of electric motor car, a lot of electric motor car owners can directly ride the electric motor car home in for the electric motor car charges. Under the conditions of the existing battery and charging technology, the electric vehicle has certain probability of fire when being charged. Once the electric vehicle is on fire, the electric vehicle can quickly spread within a few seconds, and particularly, when a plurality of people are used to sleep at night, safety accidents are more easily caused, and the life and property safety of residents are threatened.
In the prior art, a target object in an image can be detected by analyzing image data acquired by video monitoring equipment, but due to various reasons, the novel type of the target object cannot be accurately acquired. For example, in the use, the user can decorate the outward appearance of electric motor car or be used for carrying on bulky article, and specifically speaking, the user can install the dress that keeps out the wind and cover or keep out the wind on the electric motor car, or place bulky on electric motor car carrier basket, seat, leads to electric motor car outside profile to be sheltered from by other article, leads to the image data that current video monitoring equipment gathered to form the leak hunting to the electric motor car very easily.
Disclosure of Invention
The present invention is directed to a method, an apparatus, a server and a readable storage medium for identifying an electric vehicle, so as to solve the problems associated with the background art.
Based on the technical problem, the invention provides an identification method, an identification device, a server and a readable storage medium for an electric vehicle, which comprise the following four aspects.
In a first aspect, the present invention provides a method for identifying an electric vehicle, the method comprising: establishing a first identification model, wherein the first identification model is used for identifying the battery car; acquiring first information, wherein the first information is a plurality of pieces of picture information of a target object distributed along a time sequence acquired by monitoring equipment; performing first analysis on the first information through the first recognition model to obtain a first analysis result; performing second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information; judging whether to perform third analysis according to whether the second analysis result is between a second threshold value and a third threshold value; the third analysis comprises combining the first information of the same target object at different times into second information through plane projection transformation; performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result; judging whether the target object is an electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executing a first instruction; the first command is a command for prohibiting riding.
Preferably, before performing the first analysis, the method further comprises: acquiring a first interval, wherein the first interval is the time required by a target object to pass through an elevator door; and judging whether to execute the first analysis or not according to the first interval.
Preferably, the method further comprises: the first identification model adopts a VoVNet network structure as a training backbone network, a CenterNet algorithm is used as a core algorithm for identification and detection, and then a Pythroch model capable of detecting and identifying the electric vehicle and the human shape is trained on a Pythroch deep learning frame at a PC end to obtain pre-stored characteristic information; the first analysis adopts a multitask loss function to predict and optimize the position of an extraction frame of the electric vehicle in the first information; and comparing the characteristic information in the extraction frame with the characteristic information prestored in the first recognition model to obtain a first analysis result of the first information.
Preferably, the method further comprises: classifying the picture information into first image information, second image information, and third image information according to a timing; and giving a predetermined weight to the first image information, the second image information and the third image information, and calculating by a weighted average to obtain a second analysis result.
Preferably, the method further comprises: obtaining third information, wherein the third information is a plurality of pieces of picture information with optimal first analysis results in the same target object; analyzing and processing to obtain fourth information, wherein the fourth information is the feature information of the target object in the optimal candidate frame of the third information; splicing and merging the fourth information of the same target object at different times through planar projection transformation to obtain second information; and comparing the second information with the pre-stored characteristic information to obtain the characteristic similarity and the fourth analysis result of the second information.
Preferably, the method further comprises: judging whether the fourth analysis result is greater than a fourth threshold value; if yes, executing a first instruction; the first instruction comprises an alarm prompt; the fourth threshold is 80% x combined image area/real area; the real area is data pre-stored in an identification model of the electric vehicle.
Preferably, the method further comprises: determining the shooting angle and the picture light field information of the third information; optimizing the fourth information through the color information and the pixel field; judging whether the fourth information is the same characteristic region of the target object, and if so, forming a first characteristic region set; otherwise, accumulating the first characteristic region set through plane projection transformation to obtain a second characteristic region set; and fusing the overlapped parts in the second characteristic region set, eliminating splicing gaps and obtaining second information.
In a second aspect, the present invention also provides an identification apparatus for an electric vehicle, the apparatus comprising:
the first preset unit is used for establishing a first identification model, and the first identification model is used for identifying the battery car;
the monitoring device comprises a first acquisition unit, a second acquisition unit and a control unit, wherein the first acquisition unit is used for acquiring first information, and the first information is a plurality of pieces of picture information of a target object distributed along a time sequence, which is acquired by the monitoring device;
the first processing unit is used for carrying out first analysis on the first information through the first recognition model to obtain a first analysis result;
the second processing unit is used for carrying out second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
the first judging unit is used for judging whether to perform third analysis according to whether the second analysis result is between a second threshold value and a third threshold value;
the third processing unit is used for performing third analysis, and the third analysis comprises the step of combining the first information of the same target object at different times into second information through plane projection transformation;
the fourth processing unit is used for performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result;
the first execution unit judges whether the target object is the electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executes a first instruction; the first command is a command for prohibiting riding.
In a third aspect, the present invention further provides a server for identifying an electric vehicle, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for identifying an electric vehicle when executing the program.
In a fourth aspect, a computer-readable storage medium, wherein the program when executed by a processor implements the steps of the method for identifying an electric vehicle.
Has the advantages that: the invention relates to a method, a device, a server and a readable storage medium for identifying an electric vehicle, wherein a plurality of pictures are compiled along a time sequence by obtaining video data of a target object in a preset time; calculating the possibility that the electric vehicle exists in the picture, and calculating the possibility that the electric vehicle enters the elevator through a weighted average; the automatic identification can be realized in the process that the electric vehicle enters the elevator, and the manpower is saved and the identification effect is high. For the condition that a user modifies the appearance of the electric vehicle or carries large-volume objects, image information in different images is spliced and combined into a new combined image through plane projection transformation, the possibility of the electric vehicle existing in the combined image is calculated, and the accuracy of electric vehicle identification is further improved.
Drawings
Fig. 1 is a schematic flow chart of an identification method of an electric vehicle in embodiment 1 of the present invention.
Fig. 2 is a schematic flow chart of an identification method of an electric vehicle in embodiment 2 of the present invention.
Fig. 3 is an identification device for an electric vehicle according to embodiment 3 of the present invention.
Fig. 4 is a schematic structural diagram of an exemplary electronic device in embodiment 4 of the present invention.
Description of reference numerals: the device comprises a first preset unit 11, a first acquisition unit 12, a first processing unit 13, a second processing unit 14, a first judgment unit 15, a third processing unit 16, a fourth processing unit 17, a first execution unit 18, a bus 300, a receiver 301, a processor 302, a transmitter 303, a memory 304 and a bus interface 305.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be apparent, however, to one skilled in the art, that the present invention may be practiced without one or more of these specific details. In other instances, well-known features have not been described in order to avoid obscuring the invention.
Example 1
As shown in fig. 1, fig. 1 is a schematic flow chart of an identification method of an electric vehicle in embodiment 1 of the present invention, and the identification method of the electric vehicle includes the following steps:
s101, establishing a first identification model, wherein the first identification model is used for identifying the battery car.
The first identification model calculates the existence of the electric vehicle in the picture information through an image identification technology; in this embodiment, the preset identification model of the electric vehicle adopts a VoVNet network structure as a training backbone network, uses a centret algorithm as a core algorithm for identification detection, and then trains a pitorch model capable of detecting the identified electric vehicle and human shape on a pitorch deep learning frame at a PC end; the Pytrch model is converted into a Caffe model supported by Haisi 3516dv300 chip and is deployed on a Haisi board.
S102, acquiring first information, wherein the first information is a plurality of pieces of picture information of a target object distributed along a time sequence acquired by monitoring equipment;
specifically, a monitoring device is adopted to detect a target object in an elevator in real time, video information in a period from the start of the elevator to the complete entrance of the target object into the elevator is collected, video data in the period is obtained, the video data is transmitted to an image processing module, and a plurality of pictures are compiled along a time sequence through an encoder. In this embodiment, the visible light sensor is a dome camera and is installed at the entrance of a corridor or at one side of an elevator door based on an elevator principle; the preset time is several seconds after the elevator door is opened until the target object completely enters the elevator, and preferably the preset time is 2 seconds after the elevator door is opened until the target object completely enters the elevator; generally speaking, the preset time is 8-12 s, and 3s of video data after the elevator door is opened are compiled according to the interval of 1s to obtain 3 pictures, namely first image information; compiling the video data of the target object in the motion process (generally 3-7 s) from the beginning of entering the elevator to the time when the target object completely enters the elevator according to the interval of 0.5s to obtain N pictures, namely second image information; and (4) compiling video data of a plurality of seconds after the target object completely enters the elevator according to the interval of 1s to obtain 2 pictures, namely third image information.
S103, performing first analysis on the first information through the first recognition model to obtain a first analysis result;
said first analysis comprises calculating a likelihood of an electric vehicle being present in said first information by image recognition techniques; the first analysis result indicates that the possibility of the electric vehicle exists in the first information; in this embodiment, an identification model of the electric vehicle is preset, a position of an extraction frame of the electric vehicle in each picture of the first image information, the second image information and the third image information is predicted, and then feature information in the extraction frame is compared with feature information stored in the identification model of the electric vehicle in advance to obtain feature similarity of each picture, namely, the possibility of the electric vehicle existing in each picture. Then, the possibility that the electric vehicle exists in each picture is classified according to the first image information, the second image information and the third image information, and the possibility that the electric vehicle enters the elevator is obtained through weighted average calculation.
The first analysis adopts a multitask loss function to predict and optimize the position of an extraction frame of the electric vehicle in the first information; and comparing the characteristic information in the extraction frame with the characteristic information prestored in the first recognition model to obtain a first analysis result of the first information.
The prediction method for extracting the position of the frame is based on a convolutional neural network model, and the convolutional neural network model comprises three cascaded networks of a first neural network P-NeT for generating the candidate frame of the electric vehicle, a second neural network R-NeT for further optimizing the candidate frame and a third neural network O-NeT for finally producing the candidate frame. The operation method specifically comprises the following steps: smoothing the image information by using a low-pass filter, sampling the smoothed image, and constructing an image pyramid; then inputting the image data into the first neural network, inputting the result of the first neural network into the second neural network, and finally inputting the result in the second neural network into the third neural network. The first neural network, the second neural network and the third neural network are used for progressively stripping the noise signals layer by layer; and then comparing the characteristic information prestored in the identification model of the electric vehicle to obtain the characteristic similarity in each picture, namely the possibility of the electric vehicle existing in the image information of the picture. Wherein, the stripping method adopts a multitask loss function.
The loss function of the electric vehicle is as follows:
Li det＝-(yi detlog(pi)+(1-yi det)(1-log(pi)))
wherein L isi detAs a function of cross-over loss of the electric vehicle, piPredicting the classification probability of the candidate frame sample as the electric vehicle by the network; y isi detA flag that the candidate box sample is true; y isi det∈{0,1}；
The bounding box regression training function is:
wherein L isi boxCalculating Euclidean distance between the regression frame coordinate predicted by the network and the actual regression frame coordinate for predicting the frame regression loss function,
regression box coordinates predicted for the network; y isi boxIs the actual regression box coordinates.
The training function of the features in the electric vehicle is as follows:
wherein L isi landmarkLocating a loss function for the key points, calculating a regression loss function, calculating Euclidean distances between the corner coordinates predicted by the network and the actual corner coordinates,
corner coordinates predicted for the network;
are the actual corner coordinates.
The candidate frame is obtained through the multi-task loss function optimization, noise signals are reduced, and accuracy of image recognition is improved. Meanwhile, preprocessing is performed on the second judgment, and interference factors in the picture splicing process are reduced.
S104, performing second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
specifically, a specific calculation algorithm for calculating the possibility that the target object is the electric vehicle in the second analysis conforms to the following model:
wherein, P0The possibility that the electric vehicle enters the elevator is the second analysis result; n is1、n2、n3The number of pictures in the first image information, the second image information and the third image information are respectively. P1、P2、P3The possibility that the electric vehicle exists in each picture in the first image information, the second image information and the third image information is respectively set; A. b, C are weights in the first image information, the second image information and the third image information respectively, and are related to the installation positions of the image recognition model and the monitoring equipment, and the sum A, B, C is 1; in this embodiment, a value range of a is 0.219 to 0.320, a value range of B is 0.357 to 0.563, and a value range of B is 0.207 to 0.308.
Because a picture possibly represents that the electric vehicle is not comprehensively and accurately predicted, pictures of multiple different visual angles in the whole process of intercepting a target object entering the elevator are captured, the information characteristics of the target object can be comprehensively reflected by the image information of the multiple different visual angles, and the possibility that the electric vehicle enters the elevator can be obtained more accurately.
S105, judging whether to perform third analysis according to whether the second analysis result is between a second threshold and a third threshold;
specifically, the possibility P of the electric vehicle entering the elevator is compared0With the size of the second threshold, P0If the value is larger than the second threshold value, the target object is considered to be the electric vehicle, and an instruction for prohibiting taking is sent out through the elevator; the elevator can be stopped to execute the next action, and a user is warned through sound prompt and light prompt, so that the situation that the electric vehicle cannot take the elevator is guaranteed. P0If the value is less than the second threshold value, the elevator is considered to have possible electric vehicles, and step S106 is executed to further judge and detect the image information. The size of the second threshold is related to the installation positions of the map recognition model and the monitoring device, and in this embodiment, the second threshold is 80%.
Possibility P of further entering elevator for electric vehicle0Make a judgment if P0If the value is larger than the third threshold value, the elevator is considered to have the possibility of having the electric vehicle, and if the value is larger than the third threshold value, the step S105 is executed, and further judgment is made on the image information; if P0If the value is smaller than the third threshold value, the elevator is determined to be free of the electric vehicle, and the elevator can run normally. In this embodiment, the third threshold is 40%.
And S106, the third analysis comprises the step of combining the first information of the same target object at different times into second information through plane projection transformation.
In particular, since the deformation amount of the same object is relatively small in the moving process, the superposition of multiple pictures of the same object at different moments is obtained. Thus, the third analysis comprises the steps of: obtaining third information, wherein the third information is a plurality of pieces of picture information with optimal first analysis results in the same target object; analyzing and processing to obtain fourth information, wherein the fourth information is the feature information of the target object in the optimal candidate frame of the third information; splicing and merging the fourth information of the same target object at different times through planar projection transformation to obtain second information; and comparing the second information with the characteristic information prestored in the first recognition model to obtain the characteristic similarity of the second information, namely a fourth analysis result. In other words, the second information is image information in different images spliced and combined into combined image information through plane projection transformation; and the fourth analysis result is the feature similarity of the combined image information by comparing the combined image information with the pre-stored feature information.
In this embodiment, the second judgment is based on the combination of the external contours, and the specific combination method includes the steps of:
selecting a plurality of pictures with the highest possibility of electric vehicles in the step S102, and determining the shooting angle of the visible light detector and the picture light field information according to the background picture, and recording as F (S, t, u, v); where s and t are the angular resolution of the light field; u and v are the spatial resolution of the light field;
intercepting a plurality of pictures in the optimal candidate frame, and optimizing image information through the color information and pixel field, wherein an optimization model of the process is as follows:
wherein, I is a characteristic point set on the image information; n is a feature point field set on the image information; d (l)P) Is a data item; s (l)P,lq) Is a smoothing term; p is a feature point on the image information; q is a point in the four neighbourhood of p.
Comparing color information and pixel field similarity between feature regions of different images, and when the difference between the color information and the pixel field similarity is smaller than a threshold value, determining that the feature regions of different images are in the same feature region of the target object, and forming a first feature region set;
and then, accumulating the residual characteristic areas in different images into the first characteristic area set through plane projection transformation by using a shooting angle perspective principle to obtain a second characteristic area set.
The planar projective transformation is:
wherein, Fwa(s, t, u, v) is light field information of the picture after plane projection transformation; fin(s, T, u, v) is light field information of a picture with the highest possibility of existence of the electric vehicle, TdA spatial transformation matrix for the d picture; d is the number of the plane projection transformation pictures.
And fusing the overlapped partial images by using a minimum suture line algorithm and a Poisson image fusion algorithm, eliminating splicing gaps to obtain a combined image of the target object, comparing the combined image with characteristic information prestored in an identification model of the electric vehicle, performing fourth analysis, and obtaining the possibility that the target object is the electric vehicle.
S107, performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result;
the fourth analysis is the same as the first analysis, and is used for calculating the possibility of the electric vehicle in the first information by an image recognition technology; the fourth analysis result indicates that the possibility of the electric vehicle exists in the second information; the specific process may refer to step S103, which is not described herein.
S108, judging whether the target object is an electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executing a first instruction; the first instruction includes an alert prompt.
Specifically, the second analysis results P are compared0And a second threshold value, if P0If the value is larger than the second threshold value, the target object is considered to be the electric vehicle, and an instruction for prohibiting taking is sent out through the elevator; judging whether the fourth analysis result is greater than a fourth threshold value; if so, the target object is regarded as the electric vehicle, and the command of prohibiting riding is executed; otherwise, the elevator can run normally. Wherein, because of the image segment of the target object, the similarity limit is the combined image area/real area. Thus, in the present embodiment, the fourth threshold is 80% × combined image area/real area; the real area is approximate to data stored in the identification model of the electric vehicle in advance and is a fixed value. The command for prohibiting the elevator from taking can specifically be that the elevator stops executing the next action, and a user is warned through sound prompt and light prompt, so that the electric vehicle cannot take the elevator.
Compared with the prior art, the embodiment has the following advantages: through the above steps S101 to S104, most electric vehicles have been identified. However, during use, users inevitably modify the appearance of the electric vehicle or mount bulky items. For example, in winter, a user can install a wind screen mantle or a wind screen cover on the electric vehicle; or place bulky on electric motor car carrier basket, seat, lead to electric motor car external profile to be sheltered from by other article, lead to step S103 to form the electric motor car easily and leak the detection, consequently through step S105 to S108, further improve the precision nature of electric motor car discernment.
Example 2
As shown in fig. 2, fig. 2 is a schematic flow chart of an identification method of an electric vehicle according to embodiment 2 of the present invention, and the identification method of the electric vehicle includes the following steps:
s201, acquiring a first interval, wherein the first interval is the time required by a target object to pass through an elevator door; and judging whether to execute the first analysis or not according to the first interval.
In other words, it is determined whether the first interval is smaller than the first threshold, if the elevator is operating normally, otherwise step S203 is executed. Specifically, the time required for the target object to completely enter the elevator from entering the elevator is recorded by the photoelectric sensor. Because all install photoelectric sensing ware among the current elevator for whether detect the target object body and pass through the lift-cabin door, this step can directly improve on this basis, need not to increase other hardware equipment newly. When the time that the target object passes through the elevator door is less than a first threshold value, the situation that no electric vehicle enters the elevator is considered; otherwise, there is a possibility that the electric vehicle enters the elevator, and thus the step S203 is started. In the present embodiment, the first threshold is preferably 1 to 2 seconds.
In the prior art, the opening of the electric vehicle identification system is judged by the opening state of an elevator door. However, the occurrence rate of the event that the electric vehicle enters the elevator is far lower than the normal situation of the elevator, and the judgment condition causes a large amount of calculation resources to be wasted, and the loss of the controller is increased. The inventor finds that the length of the electric vehicle is far larger than the width of the human body, so the time of the electric vehicle passing through the elevator door is far longer than the time of the human body passing through the elevator door. The invention takes the time of the target object passing through the elevator door as the triggering basis of the identification method. And the existing elevator is generally provided with a photoelectric sensor for detecting whether a target object passes through the elevator door or not, and other hardware equipment does not need to be newly added when the method is adopted for judging.
S202, establishing a first identification model, wherein the first identification model is used for identifying the battery car.
The first identification model calculates the existence of the electric vehicle in the picture information through an image identification technology; in this embodiment, the preset identification model of the electric vehicle adopts a VoVNet network structure as a training backbone network, uses a centret algorithm as a core algorithm for identification detection, and then trains a pitorch model capable of detecting the identified electric vehicle and human shape on a pitorch deep learning frame at a PC end; the Pytrch model is converted into a Caffe model supported by Haisi 3516dv300 chip and is deployed on a Haisi board.
S203, acquiring first information, wherein the first information is a plurality of pieces of picture information of a target object distributed along a time sequence acquired by monitoring equipment;
specifically, a monitoring device is adopted to detect a target object in an elevator in real time, video information in a period from the start of the elevator to the complete entrance of the target object into the elevator is collected, video data in the period is obtained, the video data is transmitted to an image processing module, and a plurality of pictures are compiled along a time sequence through an encoder. In this embodiment, the visible light sensor is a dome camera and is installed at the entrance of a corridor or at one side of an elevator door based on an elevator principle; the preset time is several seconds after the elevator door is opened until the target object completely enters the elevator, and preferably the preset time is 2 seconds after the elevator door is opened until the target object completely enters the elevator; generally speaking, the preset time is 8-12 s, and 3s of video data after the elevator door is opened are compiled according to the interval of 1s to obtain 3 pictures, namely first image information; compiling the video data of the target object in the motion process (generally 3-7 s) from the beginning of entering the elevator to the time when the target object completely enters the elevator according to the interval of 0.5s to obtain N pictures, namely second image information; and (4) compiling video data of a plurality of seconds after the target object completely enters the elevator according to the interval of 1s to obtain 2 pictures, namely third image information.
S204, performing first analysis on the first information through the first recognition model to obtain a first analysis result;
said first analysis comprises calculating a likelihood of an electric vehicle being present in said first information by image recognition techniques; the first analysis result indicates that the possibility of the electric vehicle exists in the first information; in this embodiment, an identification model of the electric vehicle is preset, a position of an extraction frame of the electric vehicle in each picture of the first image information, the second image information and the third image information is predicted, and then feature information in the extraction frame is compared with feature information stored in the identification model of the electric vehicle in advance to obtain feature similarity of each picture, namely, the possibility of the electric vehicle existing in each picture. Then, the possibility that the electric vehicle exists in each picture is classified according to the first image information, the second image information and the third image information, and the possibility that the electric vehicle enters the elevator is obtained through weighted average calculation.
The first analysis adopts a multitask loss function to predict and optimize the position of an extraction frame of the electric vehicle in the first information; and comparing the characteristic information in the extraction frame with the characteristic information prestored in the first recognition model to obtain a first analysis result of the first information.
The prediction method for extracting the position of the frame is based on a convolutional neural network model, and the convolutional neural network model comprises three cascaded networks of a first neural network P-NeT for generating the candidate frame of the electric vehicle, a second neural network R-NeT for further optimizing the candidate frame and a third neural network O-NeT for finally producing the candidate frame. The operation method specifically comprises the following steps: smoothing the image information by using a low-pass filter, sampling the smoothed image, and constructing an image pyramid; then inputting the image data into the first neural network, inputting the result of the first neural network into the second neural network, and finally inputting the result in the second neural network into the third neural network. The first neural network, the second neural network and the third neural network are used for progressively stripping the noise signals layer by layer; and then comparing the characteristic information prestored in the identification model of the electric vehicle to obtain the characteristic similarity in each picture, namely the possibility of the electric vehicle existing in the image information of the picture. Wherein, the stripping method adopts a multitask loss function.
The loss function of the electric vehicle is as follows:
Li det＝-(yi detlog(pi)+(1-yi det)(1-log(pi)))
wherein L isi detAs a function of cross-over loss of the electric vehicle, piPredicting the classification probability of the candidate frame sample as the electric vehicle by the network; y isi detA flag that the candidate box sample is true; y isi det∈{0,1}；
The bounding box regression training function is:
wherein L isi boxCalculating Euclidean distance between the regression frame coordinate predicted by the network and the actual regression frame coordinate for predicting the frame regression loss function,
regression box coordinates predicted for the network; y isi boxIs the actual regression box coordinates.
The training function of the features in the electric vehicle is as follows:
wherein L isi landmarkLocating a loss function for the key points, calculating a regression loss function, calculating Euclidean distances between the corner coordinates predicted by the network and the actual corner coordinates,
corner coordinates predicted for the network;
are the actual corner coordinates.
The candidate frame is obtained through the multi-task loss function optimization, noise signals are reduced, and accuracy of image recognition is improved. Meanwhile, preprocessing is performed on the second judgment, and interference factors in the picture splicing process are reduced.
S205, performing second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
specifically, a specific calculation algorithm for calculating the possibility that the target object is the electric vehicle in the second analysis conforms to the following model:
wherein, P0The possibility that the electric vehicle enters the elevator is the second analysis result; n is1、n2、n3The number of pictures in the first image information, the second image information and the third image information are respectively. P1、P2、P3The possibility that the electric vehicle exists in each picture in the first image information, the second image information and the third image information is respectively set; A. b, C are weights in the first image information, the second image information and the third image information respectively, and are related to the installation positions of the image recognition model and the monitoring equipment, and the sum A, B, C is 1; in this embodiment, a value range of a is 0.219 to 0.320, a value range of B is 0.357 to 0.563, and a value range of B is 0.207 to 0.308.
Because a picture possibly represents that the electric vehicle is not comprehensively and accurately predicted, pictures of multiple different visual angles in the whole process of intercepting a target object entering the elevator are captured, the information characteristics of the target object can be comprehensively reflected by the image information of the multiple different visual angles, and the possibility that the electric vehicle enters the elevator can be obtained more accurately.
S206, judging whether to perform third analysis according to whether the second analysis result is between a second threshold value and a third threshold value;
specifically, the possibility P of the electric vehicle entering the elevator is compared0And the size of the second threshold，P0If the value is larger than the second threshold value, the target object is considered to be the electric vehicle, and an instruction for prohibiting taking is sent out through the elevator; the elevator can be stopped to execute the next action, and a user is warned through sound prompt and light prompt, so that the situation that the electric vehicle cannot take the elevator is guaranteed. P0If the value is less than the second threshold value, the elevator is considered to have possible electric vehicles, and step S207 is executed to further judge and detect the image information. The size of the second threshold is related to the installation positions of the map recognition model and the monitoring device, and in this embodiment, the second threshold is 80%.
Possibility P of further entering elevator for electric vehicle0Make a judgment if P0If the value is larger than the third threshold value, the elevator is considered to have the possibility of having the electric vehicle, and if the value is larger than the third threshold value, the step S207 is executed, and further judgment is made on the image information; if P0If the value is smaller than the third threshold value, the elevator is determined to be free of the electric vehicle, and the elevator can run normally. In this embodiment, the third threshold is 40%.
And S207, the third analysis comprises the step of combining the first information of the same target object at different times into second information through plane projection transformation.
In particular, since the deformation amount of the same object is relatively small in the moving process, the superposition of multiple pictures of the same object at different moments is obtained. Thus, the third analysis comprises the steps of: obtaining third information, wherein the third information is a plurality of pieces of picture information with optimal first analysis results in the same target object; analyzing and processing to obtain fourth information, wherein the fourth information is the feature information of the target object in the optimal candidate frame of the third information; splicing and merging the fourth information of the same target object at different times through planar projection transformation to obtain second information; and comparing the second information with the characteristic information prestored in the first recognition model to obtain the characteristic similarity of the second information, namely a fourth analysis result. In other words, the second information is image information in different images spliced and combined into combined image information through plane projection transformation; and the fourth analysis result is the feature similarity of the combined image information by comparing the combined image information with the pre-stored feature information.
In this embodiment, the second judgment is based on the combination of the external contours, and the specific combination method includes the steps of:
selecting a plurality of pictures with the highest possibility of electric vehicles in the step S203, and determining the shooting angle of the visible light detector and the picture light field information according to the background picture, and recording as F (S, t, u, v); where s and t are the angular resolution of the light field; u and v are the spatial resolution of the light field;
intercepting a plurality of pictures in the optimal candidate frame, and optimizing image information through the color information and pixel field, wherein an optimization model of the process is as follows:
wherein, I is a characteristic point set on the image information; n is a feature point field set on the image information; d (l)P) Is a data item; s (l)P,lq) Is a smoothing term; p is a feature point on the image information; q is a point in the four neighbourhood of p.
Comparing color information and pixel field similarity between feature regions of different images, and when the difference between the color information and the pixel field similarity is smaller than a threshold value, determining that the feature regions of different images are in the same feature region of the target object, and forming a first feature region set;
and then, accumulating the residual characteristic areas in different images into the first characteristic area set through plane projection transformation by using a shooting angle perspective principle to obtain a second characteristic area set.
The planar projective transformation is:
wherein, Fwa(s, t, u, v) is light field information of the picture after plane projection transformation; fin(s, T, u, v) is light field information of a picture with the highest possibility of existence of the electric vehicle, TdSpatial transformation moment for the d-th pictureArraying; d is the number of the plane projection transformation pictures.
And fusing the overlapped partial images by using a minimum suture line algorithm and a Poisson image fusion algorithm, eliminating splicing gaps to obtain a combined image of the target object, comparing the combined image with characteristic information prestored in an identification model of the electric vehicle, performing fourth analysis, and obtaining the possibility that the target object is the electric vehicle.
S208, performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result;
the fourth analysis is the same as the first analysis, and is used for calculating the possibility of the electric vehicle in the first information by an image recognition technology; the fourth analysis result indicates that the possibility of the electric vehicle exists in the second information; the specific process may refer to step S103, which is not described herein.
S209, judging whether the target object is an electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executing a first instruction; the first instruction includes an alert prompt.
Specifically, the second analysis results P are compared0And a second threshold value, if P0If the value is larger than the second threshold value, the target object is considered to be the electric vehicle, and an instruction for prohibiting taking is sent out through the elevator; judging whether the fourth analysis result is greater than a fourth threshold value; if so, the target object is regarded as the electric vehicle, and the command of prohibiting riding is executed; otherwise, the elevator can run normally. Wherein, because of the image segment of the target object, the similarity limit is the combined image area/real area. Thus, in the present embodiment, the fourth threshold is 80% × combined image area/real area; the real area is approximate to data stored in the identification model of the electric vehicle in advance and is a fixed value. The command for prohibiting the elevator from taking can specifically be that the elevator stops executing the next action, and a user is warned through sound prompt and light prompt, so that the electric vehicle cannot take the elevator.
Compared with the prior art, the embodiment has the following advantages: through the above steps S201 to S204, most electric vehicles have been identified. However, during use, users inevitably modify the appearance of the electric vehicle or mount bulky items. For example, in winter, a user can install a wind screen mantle or a wind screen cover on the electric vehicle; or place bulky on electric motor car carrier basket, seat, lead to electric motor car external profile to be sheltered from by other article, lead to step S204 to form the electric motor car easily and leak the detection, consequently through step S205 to S209, further improve the accuracy nature of electric motor car discernment.
Example 3
Based on the same inventive concept as the identification method of the electric vehicle in the foregoing embodiment 1 or 2, the present invention also provides an identification apparatus of an electric vehicle, as shown in fig. 3, the apparatus including:
the first preset unit 11 is used for establishing a first identification model, and the first identification model is used for identifying the battery car;
the first acquiring unit 12 is configured to acquire first information, where the first information is information of a plurality of pictures of a target object distributed along a time sequence acquired by a monitoring device;
the first processing unit 13 is configured to perform a first analysis on the first information through the first recognition model to obtain a first analysis result;
the second processing unit 14 is configured to perform a second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
a first judging unit 15, configured to judge whether to perform a third analysis according to whether the second analysis result is between a second threshold and a third threshold;
a third processing unit 16, configured to perform a third analysis, where the third analysis includes combining the first information of the same target object at different times into second information through planar projection transformation;
a fourth processing unit 17, configured to perform a fourth analysis on the second information through the first identification model, so as to obtain a fourth analysis result;
the first execution unit 18 is configured to determine whether the target object is an electric vehicle according to the second analysis result and the fourth analysis result, and if so, execute a first instruction; the first command is a command for prohibiting riding.
Further, the apparatus further comprises:
the second acquisition unit is used for acquiring a first interval, wherein the first interval is the time required by the target object to pass through the elevator door;
and the second judging unit is used for judging whether to execute the first analysis according to the first interval.
Further, the apparatus further comprises:
the fifth processing unit is used for predicting and optimizing the position of an extraction frame of the electric vehicle in the image information by adopting a multitask loss function;
and the sixth processing unit is used for comparing the characteristic information in the extraction frame with the characteristic information prestored in the identification model of the electric vehicle to obtain the characteristic similarity of each picture, namely the possibility that the electric vehicle enters the elevator.
Further, the apparatus further comprises:
a seventh processing unit configured to classify the picture information into first image information, second image information, and third image information according to a timing;
and the eighth processing unit is used for giving a preset weight to the first image information, the second image information and the third image information and obtaining a second analysis result through weighted average calculation.
Further, the apparatus further comprises:
the ninth processing unit is used for obtaining third information, wherein the third information is a plurality of pieces of picture information with optimal first analysis results in the same target object;
the tenth processing unit is used for analyzing and processing to obtain fourth information, wherein the fourth information is the feature information of the target object in the optimal candidate frame of the third information;
the eleventh processing unit is used for splicing and merging the fourth information of the same target object at different times through planar projection transformation to obtain second information;
and the third judging unit is used for comparing the second information with the pre-stored characteristic information to obtain the characteristic similarity and the fourth analysis result of the second information.
Further, the apparatus further comprises:
a fourth judging unit, configured to judge whether the fourth analysis result is greater than a fourth threshold; if yes, executing a first instruction; the first instruction comprises an alarm prompt;
the fourth threshold is 80% x combined image area/real area;
the real area is data pre-stored in an identification model of the electric vehicle.
Further, the apparatus further comprises:
a twelfth processing unit, configured to determine a shooting angle and picture light field information of the third information;
a thirteenth processing unit, configured to optimize the fourth information according to the color information and the pixel area;
a fourteenth processing unit, configured to determine whether the fourth information is the same feature region of the target object, and if so, form a first feature region set; otherwise, accumulating the first characteristic region set through plane projection transformation to obtain a second characteristic region set;
and the fifteenth processing unit is used for fusing the overlapped parts in the second characteristic region set, eliminating the splicing gap and obtaining second information.
Various changes and specific examples of the identification method of the electric vehicle in the foregoing embodiment 1 or embodiment 2 are also applicable to the identification device of the electric vehicle in this embodiment, and the implementation method of the identification device of the electric vehicle in this embodiment is clear to those skilled in the art from the foregoing detailed description of the identification method of the electric vehicle, so for the sake of brevity of the description, detailed description is not repeated here.
Example 4
Based on the same inventive concept as the method for identifying an electric vehicle in the foregoing embodiment, the present invention further provides a server for identifying an electric vehicle, as shown in fig. 4, fig. 4 is an exemplary electronic device in embodiment 4, and includes a memory 304, a processor 302, and a computer program stored on the memory 304 and executable on the processor 302, and when the processor 302 executes the program, the processor 302 implements the steps of any one of the methods for remote vital sign monitoring described above.
Where in fig. 4 a bus architecture (represented by bus 300), bus 300 may include any number of interconnected buses and bridges, bus 300 linking together various circuits including one or more processors, represented by processor 302, and memory, represented by memory 304. The bus 300 may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. A bus interface 305 provides an interface between the bus 300 and the receiver 301 and transmitter 303. The receiver 301 and the transmitter 303 may be the same element, i.e., a transceiver, providing a means for communicating with various other apparatus over a transmission medium.
The processor 302 is responsible for managing the bus 300 and general processing, and the memory 304 may be used for storing data used by the processor 302 in performing operations.
Example 5
Based on the same inventive concept as the identification method of the electric vehicle in the foregoing embodiment, the present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of: obtaining video data of a target object in preset time, and compiling a plurality of pictures along a time sequence; calculating the possibility that the electric vehicle exists in the picture, and calculating the possibility that the electric vehicle enters the elevator through a weighted average; judging whether the possibility that the electric vehicle enters the elevator is greater than a second threshold value or not; if the value is larger than the second threshold value, executing an instruction for prohibiting riding; otherwise, executing the next step; judging whether the possibility that the electric vehicle enters the elevator is greater than a third threshold value or not; if the threshold value is larger than the third threshold value, executing the next step; otherwise, the elevator can run normally; splicing and merging image information in different images into a new combined image through plane projection transformation, and calculating the possibility of electric vehicles existing in the combined image; judging whether the possibility of the electric vehicle in the combined image is greater than a fourth threshold value or not; if yes, executing an instruction for prohibiting riding; otherwise, the elevator can run normally.
One or more technical solutions in the embodiments of the present invention at least have one or more of the following technical effects: compiling a plurality of pictures along a time sequence by obtaining video data of a target object within a preset time; calculating the possibility that the electric vehicle exists in the picture, and calculating the possibility that the electric vehicle enters the elevator through a weighted average; the automatic identification can be realized in the process that the electric vehicle enters the elevator, and the manpower is saved and the identification effect is high. For the condition that a user modifies the appearance of the electric vehicle or carries large-volume objects, image information in different images is spliced and combined into a new combined image through plane projection transformation, the possibility of the electric vehicle existing in the combined image is calculated, and the accuracy of electric vehicle identification is further improved.
It should be noted that the various features described in the above embodiments may be combined in any suitable manner without departing from the scope of the invention. The invention is not described in detail in order to avoid unnecessary repetition.
Claims (10)
1. Method for identifying the entry of an electric vehicle into an elevator, characterized in that it comprises:
establishing a first identification model, wherein the first identification model is used for identifying the battery car;
acquiring first information, wherein the first information is a plurality of pieces of picture information of a target object distributed along a time sequence acquired by monitoring equipment;
performing first analysis on the first information through the first recognition model to obtain a first analysis result;
performing second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
judging whether to perform third analysis according to whether the second analysis result is between a second threshold value and a third threshold value;
the third analysis comprises combining the first information of the same target object at different times into second information through plane projection transformation;
performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result;
judging whether the target object is an electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executing a first instruction; the first command is a command for prohibiting riding.
2. The method of identifying an electric vehicle entering an elevator of claim 1, wherein prior to performing the first analysis, the method further comprises:
acquiring a first interval, wherein the first interval is the time required by a target object to pass through an elevator door;
and judging whether to execute the first analysis or not according to the first interval.
3. The method of identifying an electric vehicle entering an elevator according to claim 1, further comprising:
the first identification model adopts a VoVNet network structure as a training backbone network, a CenterNet algorithm is used as a core algorithm for identification and detection, and then a Pythroch model capable of detecting and identifying the electric vehicle and the human shape is trained on a Pythroch deep learning frame at a PC end to obtain pre-stored characteristic information;
the first analysis adopts a multitask loss function to predict and optimize the position of an extraction frame of the electric vehicle in the first information;
and comparing the characteristic information in the extraction frame with the characteristic information prestored in the first recognition model to obtain a first analysis result.
4. The method of identifying an electric vehicle entering an elevator according to claim 1, further comprising:
classifying the picture information into first image information, second image information, and third image information according to a timing;
and giving a predetermined weight to the first image information, the second image information and the third image information, and calculating by a weighted average to obtain a second analysis result.
5. The method of identifying an electric vehicle entering an elevator according to claim 3, further comprising:
obtaining third information, wherein the third information is a plurality of pieces of picture information with optimal first analysis results in the same target object;
analyzing and processing to obtain fourth information, wherein the fourth information is the feature information of the target object in the optimal candidate frame of the third information;
splicing and merging the fourth information of the same target object at different times through planar projection transformation to obtain second information;
and comparing the second information with the characteristic information prestored in the first identification model to obtain the characteristic similarity and the fourth analysis result of the second information.
6. The method of identifying an electric vehicle entering an elevator according to claim 5, further comprising:
judging whether the fourth analysis result is greater than a fourth threshold value; if yes, executing a first instruction; the first instruction comprises an alarm prompt;
the fourth threshold is 80% x combined image area/real area;
the real area is data pre-stored in an identification model of the electric vehicle.
7. The method of identifying an electric vehicle entering an elevator according to claim 5, further comprising:
determining the shooting angle and the picture light field information of the third information;
optimizing the fourth information through the color information and the pixel field;
judging whether the fourth information is the same characteristic region of the target object, and if so, forming a first characteristic region set; otherwise, accumulating the first characteristic region set through plane projection transformation to obtain a second characteristic region set;
and fusing the overlapped parts in the second characteristic region set, eliminating splicing gaps and obtaining second information.
8. Device for identifying the entry of an electric vehicle into an elevator, characterized in that it comprises:
the first preset unit is used for establishing a first identification model, and the first identification model is used for identifying the battery car;
the monitoring device comprises a first acquisition unit, a second acquisition unit and a control unit, wherein the first acquisition unit is used for acquiring first information, and the first information is a plurality of pieces of picture information of a target object distributed along a time sequence, which is acquired by the monitoring device;
the first processing unit is used for carrying out first analysis on the first information through the first recognition model to obtain a first analysis result;
the second processing unit is used for carrying out second analysis on the first analysis result to obtain a second analysis result; the second analysis comprises a weighted average summary of the first information;
the first judging unit is used for judging whether to perform third analysis according to whether the second analysis result is between a second threshold value and a third threshold value;
the third processing unit is used for performing third analysis, and the third analysis comprises the step of combining the first information of the same target object at different times into second information through plane projection transformation;
the fourth processing unit is used for performing fourth analysis on the second information through the first recognition model to obtain a fourth analysis result;
the first execution unit is used for judging whether the target object is the electric vehicle or not according to the second analysis result and the fourth analysis result, and if so, executing a first instruction; the first command is a command for prohibiting riding.
9. A server for electric vehicle identification, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 7 when executing the program.
10. A computer-readable storage medium, characterized in that the program is adapted to carry out the steps of the method of any one of claims 1-7 when executed by a processor.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2021101490648 | 2021-02-03 | ||
| CN202110149064 | 2021-02-03 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114359839A true CN114359839A (en) | 2022-04-15 |
Family
ID=81093565
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210084354.3A Withdrawn CN114359839A (en) | 2021-02-03 | 2022-01-23 | Method and system for identifying entrance of electric vehicle into elevator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114359839A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115909321A (en) * | 2023-03-08 | 2023-04-04 | 成都睿瞳科技有限责任公司 | Identification method and system for elevator car and storage medium |
-
2022
- 2022-01-23 CN CN202210084354.3A patent/CN114359839A/en not_active Withdrawn
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115909321A (en) * | 2023-03-08 | 2023-04-04 | 成都睿瞳科技有限责任公司 | Identification method and system for elevator car and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109686109B (en) | Parking lot safety monitoring management system and method based on artificial intelligence | |
| KR101995107B1 (en) | Method and system for artificial intelligence based video surveillance using deep learning | |
| CN103839373B (en) | A kind of unexpected abnormality event Intelligent Recognition alarm device and warning system | |
| US9363483B2 (en) | Method for available parking distance estimation via vehicle side detection | |
| US20190082149A1 (en) | System and method for gate monitoring during departure or arrival of an autonomous vehicle | |
| US11709282B2 (en) | Asset tracking systems | |
| KR101387628B1 (en) | Entrance control integrated video recorder | |
| KR102195706B1 (en) | Method and Apparatus for Detecting Intruder | |
| JP2007334623A (en) | Face authentication device, face authentication method, and access control device | |
| CN112991585B (en) | Access personnel management method and computer readable storage medium | |
| KR20130097868A (en) | Intelligent parking management method and system based on camera | |
| CN113158752A (en) | Intelligent safety management and control system for electric power staff approach operation | |
| CN110189355A (en) | Safe escape channel occupies detection method, device, electronic equipment and storage medium | |
| CN109446926A (en) | A kind of traffic monitoring method and device, electronic equipment and storage medium | |
| CN111814510A (en) | Detection method and device for remnant body | |
| CN114202742A (en) | Method and device for judging whether battery car enters elevator | |
| CN110780356A (en) | Subway platform clearance foreign matter detecting system | |
| KR101454644B1 (en) | Loitering Detection Using a Pedestrian Tracker | |
| CN108805184B (en) | Image recognition method and system for fixed space and vehicle | |
| CN112132048A (en) | Community patrol analysis method and system based on computer vision | |
| CN114359839A (en) | Method and system for identifying entrance of electric vehicle into elevator | |
| CN116403162B (en) | Airport scene target behavior recognition method and system and electronic equipment | |
| CN116246402A (en) | Monitoring method and device | |
| KR101924643B1 (en) | Recognition system for vehicle | |
| CN115331162A (en) | Cross-scale infrared pedestrian detection method, system, medium, equipment and terminal |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220415 |
|
| WW01 | Invention patent application withdrawn after publication |