A vehicle lamp auditing method and device

Abstract

The application discloses a vehicle lamp auditing method and device. The vehicle lamp auditing method comprises the following steps: acquiring a video frame sequence containing a target vehicle, wherein the target vehicle comprises a vehicle lamp to be audited, and a plurality of continuous target images are selected from the video frame sequence, wherein the time period corresponding to the plurality of continuous target images is related to the time period of executing the opening and closing control instruction of the vehicle lamp to be audited by the target vehicle; performing attribute detection on each target image for the vehicle lamp to be audited, obtaining the category attribute and state attribute of the vehicle lamp to be audited as the vehicle lamp attribute information corresponding to each target image; and determining whether the vehicle lamp to be audited is normal based on the vehicle lamp attribute information corresponding to each target image. Through the above method, the efficiency of the vehicle lamp auditing process and the accuracy of the auditing result can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to a vehicle light auditing method and device. BACKGROUND

[0002] With the development of society, the per capita vehicle ownership gradually increases, and the safe driving of vehicles in urban roads is becoming increasingly important. In order to ensure the driving safety of motor vehicles, the state stipulates that motor vehicles need to be regularly audited according to the regulations since the date of registration. Among them, the vehicle light as an important sign signal for communication with other vehicles in the driving process plays an important role in driving safety.

[0003] The traditional vehicle detection usually adopts manual auditing method, that is, the vehicle light is detected in turn by manual and the evidence is kept, which not only needs to invest a large amount of manpower to cause high auditing cost, but also the steps of the method are complicated and time-consuming, which seriously affects the efficiency of vehicle light auditing. On the other hand, it is inevitable to have mis-detection or missed detection in the manual auditing process, which cannot guarantee the accuracy of vehicle light auditing. SUMMARY

[0004] The technical problem solved by the present application is to provide a vehicle light auditing method and device, which can improve the efficiency of vehicle light auditing process and the accuracy of auditing results.

[0005] To solve the above technical problem, one technical solution adopted by the present application is to provide a vehicle light auditing method, comprising: acquiring a video frame sequence containing a target vehicle, wherein the target vehicle includes a vehicle light to be audited, and selecting a plurality of continuous target images from the video frame sequence, wherein the plurality of continuous target images correspond to a time period related to a time period in which the target vehicle executes a switching control instruction of the vehicle light to be audited; performing attribute detection on each target image for the vehicle light to be audited to obtain a category attribute and a state attribute of the vehicle light to be audited; and determining whether the vehicle light to be audited is normal based on the vehicle light attribute information corresponding to each target image.

[0006] Among them, the step of performing attribute detection on each target image for the vehicle light to be audited to obtain a category attribute and a state attribute of the vehicle light to be audited includes: detecting a plurality of vehicle light regions for each target image; wherein each vehicle light region includes a plurality of vehicle lights; acquiring a feature vector corresponding to each vehicle light region; and obtaining the category attribute and the state attribute corresponding to the vehicle light to be audited according to the feature vector as the vehicle light attribute information corresponding to each target image.

[0007] The step of obtaining the category attribute and the state attribute of the to-be-audited vehicle lamp according to the feature vector includes: dividing a plurality of the feature vectors into a left vehicle lamp vector set and a right vehicle lamp vector set according to vehicle lamp distribution characteristics; obtaining a conditional probability between each vector in the left vehicle lamp vector set and each vector in the right vehicle lamp vector set, and obtaining a paired vehicle lamp vector based on the conditional probability; performing vehicle lamp identification on the paired vehicle lamp vector, and outputting the category attribute and the state attribute of each vehicle lamp in the paired vehicle lamp vector; and obtaining the category attribute and the state attribute corresponding to the to-be-audited vehicle lamp from the category label and the state label of all vehicle lamps.

[0008] The step of performing vehicle lamp identification on each pair of the paired vehicle lamp vector and outputting the category attribute and the state attribute of each vehicle lamp in the paired vehicle lamp vector includes: splitting each vector in the paired vehicle lamp vector into a plurality of first feature sub-vectors and a plurality of second feature sub-vectors according to the number of multi-label categories; wherein the first feature sub-vector and the second feature sub-vector correspond to one vehicle lamp respectively; obtaining a conditional probability between each first feature sub-vector and each second feature sub-vector; and determining the category attribute and the state attribute of each vehicle lamp based on the conditional probability.

[0009] The state attribute of the to-be-audited vehicle lamp includes any one of an open state and a closed state; the step of determining whether the to-be-audited vehicle lamp is normal based on the category attribute and the state attribute of the to-be-audited vehicle lamp in all the target images includes: determining a first label number as a number of labels of the to-be-audited vehicle lamp in all the target images in the open state; determining a second label number as a number of labels of the to-be-audited vehicle lamp in all the target images in the closed state; and determining whether the to-be-audited vehicle lamp is normal based on the first label number and the second label number.

[0010] The step of determining whether the to-be-audited vehicle lamp is normal based on the first label number and the second label number includes: determining whether the first label number is greater than the second label number; if yes, determining that the to-be-audited vehicle lamp is in a normal state, and saving a plurality of continuous target images; otherwise, determining that the to-be-audited vehicle lamp is in a damaged state, and saving the video frame sequence.

[0011] The step of selecting continuous multiple target images from the video frame sequence comprises: detecting and tracking multiple vehicles in the video frame sequence, associating a same vehicle to a same identifier, querying the identifier corresponding to the target vehicle, and extracting continuous multiple target images corresponding to the identifier from the video frame sequence.

[0012] The category of the vehicle light to be audited comprises any one of a headlight, a brake light, a head left turn signal light, a tail left turn signal light, a head right turn signal light, and a tail right turn signal light.

[0013] To solve the above technical problems, another technical solution adopted by the present application is to provide a vehicle light auditing device comprising a memory and a processor coupled to each other, the memory storing program instructions, and the program instructions being used to be executed by the processor to implement the vehicle light auditing method mentioned in any of the above embodiments.

[0014] To solve the above technical problems, another technical solution adopted by the present application is to provide a computer readable storage medium storing a computer program, and the computer program is used for the vehicle light auditing method mentioned in any of the above embodiments.

[0015] Compared with the prior art, the present application has the following advantages: the present application provides a vehicle light auditing method, obtains target images about the opening and closing process of a vehicle light to be audited, and performs image multi-label detection on the target images, thereby obtaining category labels and state labels of the vehicle light to be audited, and then determining whether the vehicle light to be audited is normal according to the category labels and the state labels. The above scheme applies the image multi-label detection method to the process of vehicle light auditing, and a plurality of label information about the vehicle light to be audited can be obtained by detecting the target images once, which greatly improves the efficiency of vehicle light auditing, reduces the labor loss in vehicle light auditing, reduces the error probability caused by human factors, and improves the accuracy of the vehicle light auditing result. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0017] Figure 1 is a flowchart of an embodiment of the vehicle light auditing method of the present application;

[0018] Figure 2 is Figure 1 is a flowchart of an embodiment of step S101 in

[0019] Figure 3 is Figure 1 a flowchart of an embodiment of step S102 in

[0020] Figure 4 is Figure 3 a flowchart of an embodiment of step S303 in

[0021] Figure 5 is Figure 4 a network structure diagram of an embodiment of steps S401-S403 in

[0022] Figure 6 is Figure 4 a flowchart of an embodiment of step S403 in

[0023] Figure 7 is Figure 1 a flowchart of an embodiment of step S103 in

[0024] Figure 8 is Figure 7 a flowchart of an embodiment of step S603 in

[0025] Figure 9 is a frame diagram of an embodiment of the vehicle lamp auditing device of the present application.

[0026] Figure 10 is a structure diagram of an embodiment of the vehicle lamp auditing device of the present application.

[0027] Figure 11 is a frame diagram of an embodiment of the computer readable storage medium of the present application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0029] Please refer to Figure 1 , Figure 1 is a flowchart of an embodiment of the vehicle lamp auditing method of the present application. Specifically, it can include the following steps:

[0030] S101: acquire a video frame sequence containing a target vehicle, wherein the target vehicle comprises a to-be-audited vehicle light, and select a plurality of continuous target images from the video frame sequence, wherein the plurality of continuous target images correspond to a time period related to a time period in which the target vehicle executes a switching control instruction of the to-be-audited vehicle light.

[0031] Specifically, when a vehicle is subjected to annual inspection, the vehicle owner needs to drive the vehicle to a vehicle detection station to complete various detection tasks on the vehicle. The calibrated to-be-detected area in the vehicle detection station is an audit area of the vehicle, which can be an indoor scene or an outdoor scene, and is not specifically limited here. When the vehicle completely enters the calibrated to-be-detected area, two monitoring devices arranged in advance at both ends of the driving direction of the to-be-detected vehicle start to work, ensuring that one of the monitoring devices can completely capture the vehicle light area of the vehicle head, and the other monitoring device can completely capture the vehicle light area of the vehicle tail.

[0032] In this embodiment, the video frame sequence specifically refers to the whole process of the target vehicle from entering the to-be-detected area to leaving the to-be-detected area collected by the monitoring device. When the vehicle enters the to-be-detected area, a plurality of detection tasks including the vehicle light audit task need to be completed. The driver controls the to-be-audited vehicle light on the target vehicle according to external instructions, and the external instructions include a switching control instruction of the to-be-audited vehicle light. The plurality of continuous target images refer to part of the images selected from the video frame sequence, and the target images correspond to the switching process of the to-be-audited vehicle light, and the time period thereof corresponds to the time period in which the target vehicle executes the switching control instruction of the to-be-audited vehicle light.

[0033] In yet another embodiment, the category of the to-be-audited vehicle light includes any one of a headlight, a brake light, a vehicle head left turn signal light, a vehicle tail left turn signal light, a vehicle head right turn signal light, and a vehicle tail right turn signal light. For example, the to-be-audited vehicle light is a brake light, and the plurality of target images selected at this time correspond to the whole process of switching the brake light, and the time period of the target images is the same as the time period in which the switching control instruction of the brake light is executed.

[0034] In yet another embodiment, please refer to Figure 2 , Figure 2 is Figure 1 a flowchart of an embodiment of step S101 in Figure 1 .

[0035] S201: detect and track a plurality of vehicles in the video frame sequence, and associate the same vehicle to the same identifier.

[0036] Optionally, the technical solutions provided in the present application can simultaneously review the vehicle lights of multiple vehicles, and the monitoring device can completely capture the head or tail area of multiple vehicles in the captured picture. The detection and tracking method of deep learning is used to detect and track the video frame sequence to obtain multiple vehicle identification results, and the same vehicle in the video frame sequence is associated with the same ID identifier, thereby realizing the association of the same vehicle in different time dimensions.

[0037] Optionally, the license plate information and coordinate information of multiple vehicles can also be obtained through the detection and tracking algorithm. Specifically, (x j ,y j ,w j ,h j ) represents the coordinate position of the jth frame of the vehicle in the video frame sequence, wherein x j and y j represent the center point position of the vehicle, and w j and h j represent the width and height of the vehicle.

[0038] In the present embodiment, the coordinate information and license plate information obtained for each vehicle can be associated with the corresponding vehicle ID, and represented in the form of a set as Cars={C1,C2,…,C N}, wherein N is the total number of vehicles to be detected, and C i represents the vehicle information with ID identifier i, including the license plate information and coordinate information.

[0039] S202: Query the identifier corresponding to the target vehicle, and extract the continuous multiple frames of target images corresponding to the identifier from the video frame sequence.

[0040] Optionally, in the present embodiment, the ID identifier of the target vehicle is 1, and all images with the ID identifier 1 are extracted from the video frame sequence, and the effective detection picture including the target vehicle in each image is intercepted as the target image. The continuous multiple frames of target images are used as the to-be-identified sequence.

[0041] Through the above embodiments, the same monitoring device is used to simultaneously review the to-be-reviewed vehicle lights of multiple vehicles, and the detection and tracking algorithm is used to distinguish different vehicles, which greatly improves the vehicle review progress in the vehicle detection station and improves the vehicle light review efficiency.

[0042] S102: Perform attribute detection on each target image for the to-be-reviewed vehicle light to obtain the category attribute and state attribute of the to-be-reviewed vehicle light as the vehicle light attribute information corresponding to each target image.

[0043] Optionally, the image multi-label detection refers to outputting multiple attributes of a target in the same image. In the present application, the target is a vehicle lamp to be audited. The image multi-label detection is performed on each target image, and the final output attributes include two types of class attributes and state attributes, which together constitute the vehicle lamp attribute information of each target image.

[0044] Please refer to Figure 3 , Figure 3 is Figure 1 a flowchart of an embodiment of step S102 in

[0045] S301: For each target image, a plurality of vehicle lamp regions are detected and obtained, wherein each vehicle lamp region includes a plurality of vehicle lamps.

[0046] Optionally, since the positions of the turn signal, brake light and headlight of a vehicle are basically overlapped, and the positions of vehicle lamps of different brands or models are different, the vehicle lamp regions are combined and simplified when the image multi-label detection is used, so that the approximate region containing all vehicle lamps is used to replace the specific vehicle lamp regions such as the turn signal and brake light which are detected separately. Therefore, the vehicle lamp region in the technical solution of the present application includes a plurality of vehicle lamps. For example, in the front direction of the vehicle, only two vehicle lamp regions can be detected, which are used for identifying the vehicle lamps in the left and right directions, including the headlight, front turn signal, etc. For example, in the rear direction of the vehicle, two detection regions are also used to distinguish the vehicle brake light and rear turn signal. In the present embodiment, the vehicle lamp regions are detected by using a conventional detection head, and the plurality of vehicle lamp regions obtained are cut out separately from the target image.

[0047] S302: A feature vector corresponding to each vehicle lamp region is obtained.

[0048] Optionally, a conventional deep learning feature extraction network such as convolution and pooling is used to obtain a one-dimensional feature vector corresponding to each vehicle lamp region, wherein the length of the feature vector is N*C, N is the length of the feature representation required by each label, and C is the multi-label category.

[0049] S303: According to the feature vector, the class attribute and state attribute corresponding to the vehicle lamp to be audited are obtained as the vehicle lamp attribute information corresponding to each target image.

[0050] By applying the image multi-label detection method to the process of vehicle lamp auditing in the above-mentioned embodiments, the vehicle lamp detection region is simplified, a plurality of vehicle lamps can be included in one vehicle lamp region, and multiple label attributes can be output for the target vehicle lamp in the same vehicle lamp region, thereby improving the auditing efficiency.

[0051] Please refer to Figure 4 and Figure 5 , Figure 4 is Figure 3 a flowchart of an embodiment of step S303 inFigure 5 is Figure 4 A schematic diagram of a network structure of an embodiment of steps S401-S403 is shown in FIG. 4. The step S303 can specifically include:

[0052] S401: Divide the plurality of feature vectors into a left side vehicle light vector set 13 and a right side vehicle light vector set 15 according to the distribution characteristics of the vehicle lights.

[0053] Optionally, as shown in FIG. 3, the plurality of vehicle light regions 11 are preliminarily classified using the extracted feature vectors, and are divided into the left side vehicle light vector set 13 and the right side vehicle light vector set 15 according to the distribution characteristics of the left side and right side vehicle lights. Figure 5

[0054] S402: Obtain the conditional probability between each vector in the left side vehicle light vector set 13 and each vector in the right side vehicle light vector set 15, and obtain the paired vehicle light vector 17 based on the conditional probability.

[0055] Optionally, the left side vehicle light vector set is denoted as L, and the right side vehicle light vector set is denoted as R. The conditional probability P(L j |R i ) between all vectors in L and R is calculated, where i = 1, …, I, j = 1, …, J, I is the number of vectors in the left side vehicle light vector set, and J is the number of vectors in the right side vehicle light vector set. The vehicle light vectors are paired through the conditional probability values, where the higher the probability value, the more matched the two vehicle light vectors are. In this embodiment, the most suitable paired vehicle light vector 17 is L m and R n , indicating that the mth vehicle light vector in the left side set and the nth vehicle light vector in the right side set are the most suitable. L m and R n are merged to obtain the final paired vehicle light vector 17.

[0056] In yet another embodiment, a correlation coefficient matrix or between all vectors in L and R can also be calculated, where I is the number of vectors in the left side vehicle light vector set, and J is the number of vectors in the right side vehicle light vector set. The higher the correlation coefficient value, the more matched the two vehicle light vectors are. Finally, the most suitable paired vectors in the two sets are obtained according to the size of the correlation coefficient.

[0057] S403: Perform vehicle light recognition on the paired vehicle light vector 17, and output the class attribute and state attribute of each vehicle light in the paired vehicle light vector 17.

[0058] ​Specifically, the category attribute of the vehicle light refers to the position and type of the vehicle light, and the state attribute refers to the on-off state of the vehicle light. For example, the category attribute and the state attribute in the output paired vehicle light vector 17 are: the left brake light is on, or the right turn signal light is off, etc.

[0059] Referring to Figure 6 , Figure 6 is Figure 4 a flowchart of an embodiment of step S403. The above step S403 includes:

[0060] S501: According to the number of multi-label categories, each vector in the paired vehicle light vector is split into a plurality of first feature sub-vectors and a plurality of second feature sub-vectors; wherein the first feature sub-vector and the second feature sub-vector correspond to one vehicle light respectively.

[0061] Optionally, since the number of multi-label categories obtained in step S302 is C, two vectors L m and R n in the paired vehicle light vector are respectively split into C first feature sub-vectors C' i and C second feature sub-vectors C' j according to the number of multi-label categories.

[0062] S502: Obtain the conditional probability between each first feature sub-vector and each second feature sub-vector.

[0063] Optionally, the conditional probability P(C' j |C' i ) between each first feature sub-vector and each second feature sub-vector is also calculated, which represents the probability of the occurrence of state C' i under the condition that state C' j occurs.

[0064] S503: Determine the category attribute and the state attribute of each vehicle light based on the conditional probability.

[0065] Through the above embodiment, the feature sub-vector is convoluted by the vehicle light recognition network, and multiple vehicle light attribute information of each vehicle light is output, which effectively improves the vehicle light auditing efficiency.

[0066] S404: Obtain the category attribute and the state attribute corresponding to the vehicle light to be audited from the category attributes and the state attributes of all vehicle lights.

[0067] Optionally, in this embodiment, the category of the vehicle light to be audited is the left rear turn signal light, and the category attribute and the corresponding state attribute whose output content is the left rear turn signal light are filtered out from the category labels of all vehicle lights, which are taken as the category attribute and the state attribute corresponding to the vehicle light to be audited.

[0068] S103: Determine whether the to-be-audited vehicle light is normal based on the vehicle light attribute information corresponding to each target image.

[0069] In the embodiment, the state attribute of the to-be-audited vehicle light includes any one of an open state and a closed state. In a specific implementation scenario, refer to Figure 7 , Figure 7 is Figure 1 a flowchart of an embodiment of step S103 in FIG. 1. The step S103 includes:

[0070] S601: Determine the first label quantity as the number of labels whose state attribute of the to-be-audited vehicle light is in the open state in all target images.

[0071] S602: Determine the second label quantity as the number of labels whose state attribute of the to-be-audited vehicle light is in the closed state in all target images.

[0072] S603: Determine whether the to-be-audited vehicle light is normal based on the first label quantity and the second label quantity.

[0073] Through the above embodiment, the state data quantity of the to-be-audited vehicle light is counted for voting operation, and then the audit result of the to-be-audited vehicle light is determined, thereby effectively increasing the audit accuracy.

[0074] refer to Figure 8 , Figure 8 is Figure 7 a flowchart of an embodiment of step S603 in FIG. 1. The step S603 includes:

[0075] S701: Determine whether the first label quantity is greater than the second label quantity.

[0076] S702: If yes, determine that the to-be-audited vehicle light is in a normal state, and save the continuous multiple target images.

[0077] Optionally, for the vehicle light determined to be usable, only the target image used for making the determination needs to be retained and associated with the ID of the vehicle, so as to facilitate subsequent review of the audit record.

[0078] S703: Otherwise, determine that the to-be-audited vehicle light is in a damaged state, and save the video frame sequence.

[0079] Optionally, for the vehicle light determined to be in the damaged state, the complete video frames of the vehicle from entering the detection area to leaving the detection area need to be retained, so as to improve the reliability of subsequent manual re-inspection, and enable the audit personnel to adjust the detection result.

[0080] By the above-mentioned embodiments, the image multi-label detection method is applied to the process of the vehicle lamp auditing, and a plurality of label information about the vehicle lamp to be audited can be obtained by detecting the target image once, which greatly improves the efficiency of the vehicle lamp auditing, reduces the labor loss in the vehicle lamp auditing, reduces the error probability caused by human factors, and further improves the accuracy of the vehicle lamp auditing result.

[0081] Please refer to Figure 9 , Figure 9 is a schematic diagram of an embodiment of a vehicle lamp auditing device of the present application. The vehicle lamp auditing device 100 comprises an acquisition module 10, a detection module 12 and a judgment module 14. Specifically, the acquisition module 10 is configured to acquire a video frame sequence containing a target vehicle, wherein the target vehicle comprises a vehicle lamp to be audited, and a plurality of continuous target images are selected from the video frame sequence, wherein the time period corresponding to the plurality of continuous target images is related to the time period of the switching control instruction of the target vehicle for the vehicle lamp to be audited. The detection module 12 is configured to detect the attribute of the vehicle lamp to be audited for each target image, and obtain the class attribute and the state attribute of the vehicle lamp to be audited as the vehicle lamp attribute information corresponding to each target image. The judgment module 14 is configured to determine whether the vehicle lamp to be audited is normal based on the vehicle lamp attribute information corresponding to all target images. By the above-mentioned design, the image multi-label detection method is applied to the process of the vehicle lamp auditing, and a plurality of label information about the vehicle lamp to be audited can be obtained by detecting the target image once, which greatly improves the efficiency of the vehicle lamp auditing, reduces the labor loss in the vehicle lamp auditing, reduces the error probability caused by human factors, and further improves the accuracy of the vehicle lamp auditing result. Please refer to Figure 10 , Figure 10 is a structural schematic diagram of an embodiment of a vehicle lamp auditing device of the present application. The device 20 comprises a memory 200 and a processor 202 coupled with each other, the memory 200 stores program instructions, and the processor 202 is configured to execute the program instructions to implement the quality evaluation method of the face image mentioned in any of the above-mentioned embodiments.

[0082] Specifically, the processor 202 can also be referred to as a CPU (Central Processing Unit). The processor 202 can be an integrated circuit chip having a processing capability of signals. The processor 202 can also be a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor or the like. In addition, the processor 202 can be implemented by multiple integrated circuit chips together.

[0083] Referring to Figure 11 , Figure 11 is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of the present application. The computer-readable storage medium 30 stores a computer program 300 which can be read by a computer, and the computer program 300 can be executed by a processor to implement the quality evaluation method mentioned in any of the above embodiments. The computer program 300 can be stored in the computer-readable storage medium 30 in the form of a software product, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in the various embodiments of the present application. The computer-readable storage medium 30 with storage function can be a U disk, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk or an optical disk, etc. Various media that can store program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, etc.

[0084] In summary, unlike the prior art, the car light auditing method provided in the present application obtains a target image about the opening and closing process of the car light to be audited, and performs image multi-label detection on the target image, thereby obtaining the category attribute and state attribute of the car light to be audited, and then judging whether the car light to be audited is normal according to the category attribute and state attribute. The above scheme applies the image multi-label detection method to the process of car light auditing. Once the target image is detected, multiple label information about the car light to be audited can be obtained, which greatly improves the efficiency of car light auditing, reduces the labor loss in car light auditing, reduces the error probability caused by human factors, and further improves the accuracy of the car light auditing result.

[0085] The above merely provides the implementation of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which is made according to the content of the present application and the accompanying drawings, is also included in the patent protection scope of the present application.

Claims

1. A vehicle light auditing method, characterized by, The method comprises the following steps: obtaining a video frame sequence containing a target vehicle, wherein the target vehicle comprises a vehicle lamp to be audited, and a plurality of continuous target images are selected from the video frame sequence, wherein the time period corresponding to the plurality of continuous target images is related to the time period in which the target vehicle executes the opening and closing control instruction of the vehicle lamp to be audited; performing attribute detection on each target image for the vehicle lamp to be audited to obtain the category attribute and state attribute of the vehicle lamp to be audited as the vehicle lamp attribute information corresponding to each target image; determining whether the vehicle lamp to be audited is normal based on the vehicle lamp attribute information corresponding to each target image; wherein the step of performing attribute detection on each target image for the vehicle lamp to be audited to obtain the category attribute and state attribute of the vehicle lamp to be audited as the vehicle lamp attribute information corresponding to each target image comprises: for each target image, detecting and obtaining a plurality of vehicle lamp regions; wherein each vehicle lamp region comprises a plurality of vehicle lamps; obtaining a feature vector corresponding to each vehicle lamp region; dividing a plurality of feature vectors into a left vehicle lamp vector set and a right vehicle lamp vector set according to vehicle lamp distribution characteristics; obtaining the conditional probability between each vector in the left vehicle lamp vector set and each vector in the right vehicle lamp vector set, and obtaining a paired vehicle lamp vector based on the conditional probability; according to the number of multi-label categories, splitting each vector in the paired vehicle lamp vector into a plurality of first feature sub-vectors and a plurality of second feature sub-vectors; wherein the first feature sub-vector and the second feature sub-vector correspond to one vehicle lamp respectively; obtaining the conditional probability between each first feature sub-vector and each second feature sub-vector; determining the category attribute and the state attribute of each vehicle lamp based on the conditional probability; obtaining the category attribute and the state attribute of the vehicle lamp to be audited from the category attribute and the state attribute of all vehicle lamps.

2. The vehicle lamp auditing method according to claim 1, wherein: the state attribute of the vehicle lamp to be audited comprises any one of an open state and a closed state; the step of determining whether the vehicle lamp to be audited is normal based on the category attribute and the state attribute of the vehicle lamp to be audited in all target images comprises: determining the number of labels in which the state attribute of the vehicle lamp to be audited in all target images is an open state as a first label number; determining the number of labels in which the state attribute of the vehicle lamp to be audited in all target images is a closed state as a second label number; determining whether the vehicle lamp to be audited is normal based on the first label number and the second label number.

3. The vehicle light auditing method of claim 2, wherein, The step of determining whether the vehicle lamp to be audited is normal based on the first label number and the second label number comprises: determining whether the first label number is greater than the second label number; if yes, determining that the vehicle lamp to be audited is in a normal state, and saving the plurality of continuous target images; otherwise, determining that the vehicle lamp to be audited is in a damaged state, and saving the video frame sequence.

4. The vehicle light auditing method of claim 1, wherein The step of selecting continuous multiple target images from the sequence of video frames comprises: detecting and tracking multiple vehicles in the sequence of video frames, associating the same vehicle to the same identification; querying the identification corresponding to the target vehicle, and extracting continuous multiple target images corresponding to the identification from the sequence of video frames.

5. The vehicle light auditing method of claim 1, wherein The category of the vehicle light to be audited includes any one of a headlight, a brake light, a head left turn signal light, a tail left turn signal light, a head right turn signal light, and a tail right turn signal light.

6. A vehicle light auditing apparatus characterized by comprising: The computer readable storage medium stores a computer program for implementing the vehicle light auditing method according to any one of claims 1-5.

7. A computer readable storage medium characterized by The computer readable storage medium stores a computer program for implementing the vehicle light auditing method according to any one of claims 1-5.

Images