WO2023092982A1 - 状态检测方法、装置、计算机设备、存储介质和程序产品 - Google Patents

状态检测方法、装置、计算机设备、存储介质和程序产品 Download PDF

Info

Publication number
WO2023092982A1
WO2023092982A1 PCT/CN2022/096575 CN2022096575W WO2023092982A1 WO 2023092982 A1 WO2023092982 A1 WO 2023092982A1 CN 2022096575 W CN2022096575 W CN 2022096575W WO 2023092982 A1 WO2023092982 A1 WO 2023092982A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
state
target
state information
image
Prior art date
Application number
PCT/CN2022/096575
Other languages
English (en)
French (fr)
Inventor
申影影
Original Assignee
上海商汤智能科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 上海商汤智能科技有限公司 filed Critical 上海商汤智能科技有限公司
Publication of WO2023092982A1 publication Critical patent/WO2023092982A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present disclosure relates to the technical field of image processing, and in particular to a state detection method, device, computer equipment, storage medium and program product.
  • the detection of the state of the car light has great application value in judging whether the driver is driving legally.
  • the status of the lights can be determined by detecting the images including the lights.
  • the whole vehicle image is usually detected by a vehicle light state classifier, and the state of the vehicle light is obtained through supervised training. Since the entire vehicle image contains more vehicle information, using the entire vehicle image for training results in low detection accuracy of the state of the vehicle lights.
  • Embodiments of the present disclosure at least provide a state detection method, device, computer equipment, storage medium, and program product, which further optimizes the initial state information corresponding to the detected target object, and can obtain higher-precision target state information.
  • an embodiment of the present disclosure provides a state detection method, including:
  • the state sequence information includes information used to characterize the target object at multiple historical moments historical status information
  • target state information corresponding to the target object is determined.
  • a plurality of sub-object images are spliced into a target image, so that the target image only includes the sub-object images of the preset parts of the target object to be detected, and since the images of the non-detected parts are removed, the images of the preset parts to be detected are focused (i.e. sub-object image), therefore, the object state detection on the target image can obtain accurate initial state information corresponding to the target object.
  • the initial state information is further optimized by using the historical state information that characterizes the target object at multiple historical moments. Since the historical state information at multiple historical moments is associated with the current initial state information, combining multiple historical The historical state information at each moment optimizes the initial state information, which can improve the accuracy of the optimized target state information.
  • the state sequence information further includes target validity information corresponding to each historical state information
  • the determining the target state information corresponding to the target object based on the initial state information and the state sequence information includes:
  • the state information of the target object in a certain historical period can be reflected more accurately through multiple historical state information, that is, the above-mentioned effective state information.
  • the target validity information corresponding to the information for example, the effective historical state information can be used to determine the above effective state information more accurately.
  • using the effective state information to optimize the initial state information can improve the accuracy of the target state information corresponding to the final target object.
  • it also includes:
  • the initial validity information corresponding to the initial state information is used as the target validity information corresponding to the initial state information , and adding the initial state information and target validity information to the sequence corresponding to the state sequence information to obtain updated state sequence information.
  • the amount of historical state information accumulated in the state sequence information is not enough, it cannot accurately reflect the state information of the target object in a certain historical period. At this time, it is necessary to add historical state information to the state sequence information.
  • the step of determining the initial validity information is further included:
  • the reliability is greater than a first preset value, it is determined that the initial validity information corresponding to the initial state information is valid.
  • the first preset value can be determined based on empirical values, it can more accurately evaluate whether the initial validity information corresponding to the initial state information is valid. Therefore, by comparing the credibility of the initial state information Whether the initial validity information is valid can be accurately determined by the distance between the first preset value and the initial validity information. If the reliability is greater than the first preset value, it is determined that the initial validity information is valid.
  • the method further includes:
  • the target state information and the target validity information are added to a state sequence to obtain updated state sequence information.
  • the target state information is the state information of the optimized target object, and the optimized target state information and the target validity information corresponding to the target state information are added to the state sequence information, which can improve the historical state information in the state sequence information.
  • the determining the target validity information corresponding to the target state information includes:
  • the target state information in response to the reliability being less than or equal to the first preset value, the valid state information being the same as the initial state information, and the reliability being greater than a second preset value
  • the corresponding target validity information is valid; wherein, the first preset value is greater than the second preset value.
  • the first preset value and the second preset value can be determined based on empirical values, which can more accurately evaluate whether the target validity information corresponding to the target state information is valid; combined with the first preset value, the second The two preset values, the valid state information and the initial state information can determine the target validity information more accurately.
  • the determining the target validity information corresponding to the target status information further includes:
  • the target validity information corresponding to the target state information is invalid.
  • the degree of reliability is small (less than or equal to the second preset value), or the degree of reliability is relatively large (greater than the second preset value, but less than or equal to the first preset value), and the valid state information is related to When the initial state information is different, the target state information is likely to be invalid. Therefore, the target validity information can be determined more accurately by combining the first preset value, the second preset value, the valid state information, and the initial state information.
  • the state sequence information further includes timing information corresponding to each of the historical state information
  • the target is determined based on the timing information corresponding to each of the historical state information in the state sequence information and the standard state sequence of each preset state corresponding to the target object.
  • the predicted state information corresponding to the object including:
  • the standard state sequence of each preset state corresponding to the target object Based on the timing information corresponding to each of the historical state information in the state sequence information, the standard state sequence of each preset state corresponding to the target object, and the image sampling frequency, determine the prediction corresponding to the target object State information; wherein, the standard state sequence includes a state start subsequence and a state end subsequence of corresponding preset states.
  • the standard state sequence of each preset state corresponding to the target object can be accurately determined according to the image sampling frequency, and a more accurate Forecast status information.
  • the acquisition of sub-object images corresponding to multiple preset parts contained in the target object in the image to be detected includes:
  • the splicing of multiple sub-object images into a target image includes:
  • the sub-object images corresponding to each preset part are spliced to obtain the target image.
  • the sub-object images corresponding to the preset parts are spliced according to the position information, and the obtained target image not only focuses on the preset parts, but also retains the relative position information of the preset parts on the target object, that is, retains the preset position Corresponding structural information, therefore, detection based on the target image can improve the efficiency and accuracy of state detection.
  • the embodiment of the present disclosure also provides a state detection device, including:
  • the information acquisition part is configured to acquire sub-object images corresponding to multiple preset parts contained in the target object in the image to be detected, and state sequence information corresponding to the target object; the state sequence information includes information used to characterize the Historical state information of the target object at multiple historical moments;
  • the initial state detection part is configured to stitch a plurality of sub-object images into a target image, and perform object state detection on the target image to obtain initial state information corresponding to the target object;
  • the state optimization part is configured to determine target state information corresponding to the target object based on the initial state information and the state sequence information.
  • the state sequence information further includes target validity information corresponding to each historical state information
  • the state optimization part is further configured to, when the number of historical state information stored in the state sequence information is greater than a preset number, based on a plurality of the historical state information and each of the historical state information corresponding target validity information, determining the valid state information corresponding to the target object in the state sequence information;
  • the device includes a sequence updating part
  • the sequence update part is configured to use the initial validity information corresponding to the initial state information as the Target validity information corresponding to the initial state information, and adding the initial state information and target validity information to the sequence corresponding to the state sequence information to obtain updated state sequence information.
  • sequence updating part is further configured to obtain the initial validity information corresponding to the initial state information as the target validity information corresponding to the initial state information.
  • the reliability is greater than a first preset value, it is determined that the initial validity information corresponding to the initial state information is valid.
  • sequence updating part is further configured to determine the target state after determining the target state information corresponding to the target object based on the initial state information and the effective state information.
  • Target validity information corresponding to the information
  • the target state information and the target validity information are added to a state sequence to obtain updated state sequence information.
  • sequence updating part is further configured to obtain the credibility corresponding to the initial state information
  • the target state information in response to the reliability being less than or equal to the first preset value, the valid state information being the same as the initial state information, and the reliability being greater than a second preset value
  • the corresponding target validity information is valid; wherein, the first preset value is greater than the second preset value.
  • sequence updating part is further configured to respond to the reliability being less than or equal to the first preset value, and the valid state information is different from the initial state information Same, determine that the target validity information corresponding to the target state information is invalid;
  • the target validity information corresponding to the target state information is invalid.
  • the state sequence information further includes timing information corresponding to each of the historical state information
  • the state optimization part is further configured to determine the target object based on the timing information corresponding to each of the historical state information in the state sequence information and the standard state sequence of each preset state corresponding to the target object Corresponding forecast status information;
  • the state optimization part is further configured to obtain a preset image sampling frequency
  • the standard state sequence of each preset state corresponding to the target object Based on the timing information corresponding to each of the historical state information in the state sequence information, the standard state sequence of each preset state corresponding to the target object, and the image sampling frequency, determine the prediction corresponding to the target object State information; wherein, the standard state sequence includes a state start subsequence and a state end subsequence of corresponding preset states.
  • the information acquisition part is further configured to identify the image to be detected, and obtain the position of each preset position in the target object among the plurality of preset positions included in the target object.
  • the initial state detection part is further configured to splice the sub-object images corresponding to each preset part according to the position information of each preset part in the image to be detected to obtain the target image.
  • an embodiment of the present disclosure further provides a computer device, including: a processor, a memory, and a bus, the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processing The processor communicates with the memory through a bus, and when the machine-readable instructions are executed by the processor, the above-mentioned first aspect, or the steps of any possible state detection method in the first aspect are executed.
  • embodiments of the present disclosure further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned first aspect, or any of the first aspects of the first aspect, may be executed. Steps of a possible state detection method.
  • the embodiments of the present disclosure further provide a computer program product, including computer readable codes, where the computer program product includes a computer program or an instruction, and when the computer program or instruction is run on a computer device, the The computer device executes the above first aspect, or the steps of any possible state detection method in the first aspect.
  • FIG. 1 shows a flowchart of a state detection method provided by an embodiment of the present disclosure
  • Fig. 2 shows a schematic flow chart of the vehicle lamp state detection process provided by the embodiment of the present disclosure
  • FIG. 3 shows a flow chart of determining target state information provided by an embodiment of the present disclosure
  • Fig. 4 shows a flow chart of determining target state information provided by an embodiment of the present disclosure
  • Fig. 5 shows a schematic diagram of a state detection device provided by an embodiment of the present disclosure
  • FIG. 6 shows a schematic structural diagram of a computer device provided by an embodiment of the present disclosure.
  • the state detection of the vehicle light can determine the state of the vehicle light by detecting the image including the vehicle light in practice.
  • the related technology is to detect the entire vehicle image through the vehicle light state classifier, and supervise the training to obtain the state of the vehicle light. Since the entire vehicle image contains more vehicle information, using the entire vehicle image for training results in low detection accuracy of the state of the vehicle lights.
  • the present disclosure provides a state detection method, which stitches multiple sub-object images into a target image, so that the target image only includes the sub-object images of the preset parts of the target object to be detected. Since the images of the non-detected parts are removed, the focus The image of the detected preset part (ie, the sub-object image), therefore, the object state detection is performed on the target image, and the accurate initial state information corresponding to the target object can be obtained. Afterwards, the initial state information is further optimized by using the historical state information that characterizes the target object at multiple historical moments. Since the historical state information at multiple historical moments is associated with the current initial state information, combining multiple historical The historical state information at each moment optimizes the initial state information, which can improve the accuracy of the optimized target state information.
  • the execution subject of the state detection method provided in the embodiments of the present disclosure is generally a computer device with certain computing capabilities.
  • computer equipment refers to servers, notebook computers, tablet computers, desktop computers, smart TVs, set-top boxes, mobile devices (such as mobile phones, portable video players, personal digital assistants, dedicated messaging devices, portable game devices), etc.
  • Devices with status detection capabilities may be implemented by a processor invoking computer-readable instructions stored in a memory.
  • the state detection method provided by the embodiment of the present disclosure will be described below by taking the execution subject as a computer device as an example.
  • the state detection method provided in the embodiment of the present disclosure can be applied to detect the state of the light of the motor vehicle light.
  • the basic state of the light state includes turning left, Right turn, double flashing, braking, no turning (both left and right lights are off, it may be moving forward during the day, this state requires further optimization to determine the final state of the lights).
  • the state of car light detection includes four basic states of turning left, turning right, double flashing, and not turning, and braking, braking/forward (both left and right lights are on, which may be braking during the day or at night) Forward, this state needs to be further optimized), the arrangement of the three states of non-braking, determine 12 different detection states of the car lights, namely left turn + brake, right turn + brake, double flash + brake, no Turn + brake (that is, brake), turn left + brake / move forward, turn right + brake / move forward, double flash + brake / move forward, do not turn + brake / move forward (that is, brake to move forward), turn left + no brake (that is, turn left), turn right + no brake (that is, turn right), double flash + no brake (that is, double flash), no turn + no brake (that is, no turn).
  • FIG. 1 is a flowchart of a state detection method provided by an embodiment of the present disclosure, the method includes steps S101 to S103, wherein:
  • S101 Obtain the sub-object images corresponding to multiple preset parts contained in the target object in the image to be detected, and the state sequence information corresponding to the target object; the state sequence information includes historical state information used to characterize the target object at multiple historical moments .
  • the image to be detected includes a target object, and the target object is an object waiting for state detection, such as a vehicle, a specific pedestrian in pedestrian re-identification (Person re-identification, reid), 3D monocular target detection (monocular 3d object detection, components in mono3D), etc.
  • state detection such as a vehicle, a specific pedestrian in pedestrian re-identification (Person re-identification, reid), 3D monocular target detection (monocular 3d object detection, components in mono3D), etc.
  • the preset parts include car light components, such as left car light component, right car light component, brake light component (ie dome light component) and so on.
  • the left headlight assembly includes the left turn signal, fog lamp and position marker lamp, etc.
  • the left headlight assembly includes the right turn signal, fog lights and position lights, etc.
  • the brake light assembly includes brake lights and the like. It should be noted that different types of vehicles have different layouts of lights. For example, a certain type of vehicle does not have a brake light assembly.
  • the sub-object images corresponding to the multiple preset parts included in the target object may include multiple sub-object images corresponding to vehicle lights.
  • the car light components are installed in different positions in the vehicle, and the position of the car light components and each car light component can be detected by using the car light detection model, and the position information is obtained; after that, according to the position information, the car light is extracted from the image to be detected
  • the image corresponding to the component is cut out to obtain multiple sub-object images, for example, the left car light component image, the right car light component image, and the brake light component image.
  • the state sequence information 30 includes historical state information 31 used to characterize the target object at multiple historical moments.
  • the state sequence corresponding to the state sequence information includes a plurality of historical states corresponding to the historical state information arranged in sequence according to the sequence information. As shown in FIG. 2 , a plurality of historical status information 31 are arranged according to time sequence information t, and the time sequence from top to bottom is t-1, t-2, t-3, . . . , t-n.
  • the historical state may include the state after the initial state corresponding to the initial state information obtained at the historical moment has been optimized through the following optimization process (S102-S103).
  • the process of determining the historical state information may refer to the initial state in the following S102-S103
  • the information is optimized to the target state information and added to the description of the state sequence information, and the repeated parts will not be repeated here.
  • the target state information becomes historical state information.
  • S102 Stitch multiple sub-object images into a target image, and perform object state detection on the target image to obtain initial state information corresponding to the target object.
  • the vehicle light detection model 21 (such as Faster RCNN and other neural network models, and use the eye light sample image to perform Training) detect the car light assembly in the image 20 to be detected, and output the sub-object image comprising the car light assembly, wherein the sub-object image includes the left car light assembly image 22, the right car light assembly image 23, the brake light assembly image 24, the left turn signal assembly image 25, right turn signal light image 26; afterward, the above-mentioned multiple sub-object images are spliced according to the position order of each light assembly on the vehicle, and the target image 27 is determined; after that, the target image 27 is input into the vehicle light
  • the state detection model (such as Faster RCNN and other neural network models, using sample images of different states of the lights for training) 28 detects the state of the lights and determines the corresponding initial state information 29 of the vehicle.
  • Stitching the above-mentioned plurality of sub-object images according to the position order of each lamp assembly on the vehicle may include: splicing the left lamp assembly image 22 on the left side of the middle part of the target image 27, and splicing the right lamp assembly image 23 on the target image
  • the brake light assembly image 24 is spliced at the top of the target image
  • the left turn signal assembly image 25 is spliced at the bottom left side of the target image
  • the right turn signal image 26 is spliced at the bottom right side of the target image 27
  • the left headlight assembly of the vehicle includes a left turn signal
  • the right headlight assembly includes a right turn signal without a brake light. Therefore, the brake light assembly image 24, the left turn signal assembly image 25 and the right turn signal image 26 have no substantial content. display, which is a black frame.
  • the efficiency of vehicle light state detection can be improved.
  • the detection of the state of the light is only performed on the light assembly, and the influence of interference factors other than the light assembly is reduced, so the error of the detection of the state of the light can be reduced, and the accuracy of the state detection of the light can be improved.
  • the vehicle light state detection model 28 outputs the initial state information corresponding to the vehicle, and the credibility of the initial state information, wherein the lower the reliability, the less credible the initial state information output by the vehicle light state detection model, that is, the vehicle light state detection model The detection error is large.
  • the initial state information includes the state information corresponding to the 12 different vehicle light detection states listed above, such as label information, that is, each vehicle detection state corresponds to a label, and the vehicle light state detection model outputs a label, which is used to indicate the currently detected vehicle.
  • label information that is, each vehicle detection state corresponds to a label
  • the vehicle light state detection model outputs a label, which is used to indicate the currently detected vehicle.
  • the state of the light Exemplarily, the label corresponding to the forward line is set as 0, the label corresponding to the left turn is 1, the label corresponding to the right turn is 2, the label corresponding to the brake is 3, . . . .
  • S103 Determine target state information corresponding to the target object based on the initial state information and state sequence information.
  • the flow chart of determining the target state information can be referred to as shown in FIG. 2 , where 32 represents the target state information, which can be obtained by combining the initial state information 29 and the state sequence information 30 .
  • the state sequence information also includes target validity information corresponding to each historical state information, wherein the target validity information is information for evaluating whether the historical state information is valid.
  • the target validity information is valid, that is, the historical state information corresponding to the target validity information is valid.
  • “Valid” indicates that the vehicle light state can be used to optimize the initial state corresponding to the initial state information
  • "invalid” indicates that the detected vehicle light state cannot be used to optimize the initial state corresponding to the initial state information.
  • the forward state is a preset invalid state.
  • the initial state information can be further optimized by using the historical state information representing the target object at multiple historical moments, because the historical state information at multiple historical moments is different from the current initial state information Correlation, therefore, optimizing the initial state information by combining historical state information at multiple historical moments can reduce the impact on vehicle light detection in occlusion, missed detection, or dense scenes, and can improve the accuracy of the optimized target state information.
  • the computer device may determine valid state information corresponding to the target object in the state sequence information based on a plurality of pieces of historical state information and target validity information corresponding to each piece of historical state information. Afterwards, based on the initial state information and the effective state information, the target state information corresponding to the target object is determined.
  • the effective state information is used to represent the effective historical state of the target object at historical moments, and the effective historical state can be selected from the state sequence indicated by the state sequence information, that is, the effective state information selected from multiple historical state information Historical state information.
  • Determining the valid state information corresponding to the target object includes the following steps:
  • the initial state information of the vehicle output by the vehicle light state detection model is optimized by using multiple historical state information included in the state sequence information.
  • the output label of the vehicle light state detection model is 1
  • determine the initial state information of the vehicle m(t) 1, that is, the initial state is turning left, where t represents the number of frames, and m(t) represents the corresponding initial state of the vehicle determined by detecting the t-th frame of the image to be detected information.
  • the target state information corresponding to the vehicle can be referred to formula 1:
  • ⁇ represents the weight of the initial state information output by the lamp state detection model in the optimization process
  • ⁇ represents the weight of the effective state information in the optimization process
  • the computer device may determine the target based on multiple historical state information and target validity information corresponding to each historical state information. Effective state information corresponding to the object; based on the initial state information and the effective state information, determine the target state information corresponding to the target object in the state sequence information. In this way, after accumulating a certain amount of historical state information in the state sequence information, the state information of the target object in a certain historical period can be reflected through multiple historical state information, that is, effective state information, thereby improving the accuracy of the state information of the target object .
  • the preset number may be set according to the image sampling frequency when the photographing device collects the image to be detected, which is not limited in this embodiment of the present disclosure.
  • the preset number is generally greater than the amount of historical state information in the standard state sequence of a preset state, so that the state sequence information composed of the preset amount of historical state information can represent the preset state corresponding to the target object within a period of time.
  • the preset state represents an execution behavior state, such as the left turn state corresponding to the left turn behavior, including multiple historical state information, such as turn left, move forward, turn left, move forward, turn left, move forward, determine the left Turn on.
  • a standard state sequence of preset states may include a state-on subsequence and a state-end subsequence of corresponding preset states.
  • the preset state may include left steering, right steering, forward driving, braking, double flashing and so on.
  • the turn-on subsequence of the state of the vehicle light turning left includes 6 historical state information, such as turning left, moving forward, turning left, moving forward, turning left, moving forward, and it is determined that the turn left is turned on.
  • the end subsequence of the left-turn state includes 4 historical state information, that is, other vehicle light states except for left-turn, such as forward, forward, brake, and brake, to determine the end of the left turn.
  • the number of historical state information stored in the state sequence information needs to be greater than the preset number in order to more accurately determine the preset state corresponding to the target object, and then more accurately determine Reflect the effective state information corresponding to the target object.
  • the description of determining the target state information corresponding to the target object can refer to the detailed content in the above formula 1, and the repeated part will not be repeated here.
  • the initial validity information corresponding to the initial state information is used as the target validity information corresponding to the initial state information, and The initial state information and the target validity information are added to the sequence corresponding to the state sequence information to obtain the updated state sequence information.
  • the initial validity information corresponding to the initial state information is continuously determined and added to the sequence corresponding to the state sequence information until the accumulated historical state information in the updated state sequence information reaches the expected amount. This helps to improve the accuracy of the determined target state information.
  • the initial validity information is information for evaluating whether the initial state information output by the vehicle light state detection model is valid, and if the initial validity information is valid, the initial state information corresponding to the initial validity information is valid.
  • the computer device determining the initial validity information corresponding to the initial state information may include: obtaining the credibility corresponding to the initial state information; when the credibility is greater than the first preset value, determining The initial validity information corresponding to the initial state information is valid.
  • the first preset value may be set according to empirical values, which is not limited in this embodiment of the present disclosure.
  • the first preset value can be determined based on empirical values, it can more accurately evaluate whether the initial validity information corresponding to the initial state information is valid. Therefore, by comparing the credibility corresponding to the initial state information with the first preset The size between the values can accurately determine whether the initial validity information is valid, that is, if the reliability is greater than the first preset value, it can be determined that the initial validity information is valid.
  • the historical state information included in the state sequence information may be arranged according to the chronological order in which the historical state information is determined. Therefore, the state sequence information also includes timing information corresponding to the historical state information, and the timing information is the determined historical state information. time sequence. As shown in FIG. 2 , historical state information is added sequentially from bottom to top in chronological order. After determining the target state information and the target validity information corresponding to the target state information, according to the determined time of the target state information, add the target state information and target validity information to the state sequence in sequence, that is, add the target state information in the In FIG. 2 , the state sequence information 30 is at the top of the area, and the updated state sequence information also includes timing information corresponding to the target state information.
  • updating the state sequence information by using the optimized target state information and the target validity information corresponding to the target state information can improve the accuracy of optimizing the initial state information corresponding to the target object in the target image in the next frame.
  • determining the target validity information corresponding to the target state information by the computer device may include: first, acquiring the credibility corresponding to the initial state information.
  • the initial state information is the initial state information output by the lamp state detection model before the target state information is optimized.
  • the credibility corresponding to the initial state information can be used as a basis for judging whether the target validity information is valid.
  • the credibility corresponding to the initial state information can be used as the primary basis for judging the validity of the target validity information.
  • Ways of judging effectiveness may include:
  • Mode 1 In response to the reliability being greater than a first preset value, determine that the target validity information corresponding to the target state information is valid.
  • the credibility is the primary basis for judging the validity of the target validity information
  • it can be directly determined that the target validity information corresponding to the target state information is valid, Judgment on other conditions will not be performed.
  • the secondary basis includes the comparison between the valid state information and the initial state information, and the comparison between the reliability and the second preset value.
  • Mode 2 In response to the reliability being less than or equal to the first preset value, the valid state information being the same as the initial state information, and the reliability being greater than the second preset value, determine that the target validity information corresponding to the target state information is valid. Wherein, the first preset value is greater than the second preset value.
  • the importance of the credibility can be reduced, that is, the credibility is greater than the second preset value, and the target validity information corresponding to the target state information can be determined to be valid.
  • Way 3 In response to the reliability being less than or equal to the first preset value and the effective state information being different from the initial state information, determine that the target validity information corresponding to the target state information is invalid.
  • Mode 4 In response to the reliability being less than or equal to the first preset value and the reliability being less than or equal to the second preset value, determine that the target validity information corresponding to the target state information is invalid.
  • the above modes 3 and 4 illustrate that if the primary basis does not satisfy the condition, and if any one of the comparison results in the secondary basis does not meet the condition, it is determined that the target validity information is invalid.
  • the first preset value and the second preset value may be determined based on empirical values, which are not limited in the embodiments of the present disclosure.
  • the first preset value and the second preset value can more accurately evaluate the correspondence between target state information whether the target effectiveness information is valid; combining the first preset value, the second preset value, the effective state information and the initial state information, the accuracy of determining the target effectiveness information is improved.
  • FIG. 3 it is a flowchart of a method for determining target status information, including S301-S302.
  • the timing information includes the order of the historical state information in the sequence corresponding to the state sequence information, and is arranged according to the order in which each historical state information is stored.
  • Each preset state, standard state sequence, and standard state sequence corresponding to the target object include the state start subsequence and state end subsequence of the corresponding preset state.
  • the predicted state information predicts the latest acquired state information corresponding to the target object on the image to be detected.
  • the known state sequence information includes 20 historical state information, which are brake, forward, forward, forward, left turn, forward, left turn, forward, left turn, forward, left turn, forward Go, turn left, move forward, turn left, move forward, turn left, move forward, turn left, move forward, turn left, move forward, turn left, move forward, turn left, move forward, move forward, turn left, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward, move forward
  • the timing information and the standard state sequence predict the state of the vehicle's lights in the 21st frame of the image to be detected.
  • the state of each car light keeps turning left and moving forward, and the corresponding sequence of the continuous cycle includes the state opening subsequence of the light turning left. It can be determined that the left turning is turned on.
  • the 20th historical state information it is the forward state, and the latest acquisition can be predicted.
  • the predicted state information corresponding to the target object on the image to be detected is left turn.
  • S302. Determine target state information corresponding to the target object based on the initial state information, effective state information, and predicted state information.
  • the predicted state information s(t) is determined, and s(t) represents the predicted state information corresponding to the predicted vehicle when the t-th frame of the image to be detected is detected. According to formula 2, determine the target state information y(t) corresponding to the target object:
  • ⁇ represents the weight of the initial state information output by the vehicle light state detection model in the optimization process
  • ⁇ represents the weight of the effective state information in the optimization process
  • p(t) is a periodic term, which represents the periodicity of the effective state information
  • ⁇ represents the weight of the predicted state information in the optimization process
  • an embodiment of the present disclosure shows a schematic flowchart of a method for determining target status information, as shown in FIG. 4 , the method may include the following steps:
  • l1 is greater than l2, and l1 and l2 are used to control the switching time between states.
  • the target state information can be calculated by formula 2.
  • combining the predicted state information and the effective state information, and further optimizing the initial state information at the same time can improve the accuracy of the target state information corresponding to the finally determined target object.
  • the computer device can also obtain a preset image sampling frequency; then, based on the timing information corresponding to each historical state information in the state sequence information, the standard state sequence of each preset state corresponding to the target object, and The image sampling frequency determines the predicted state information corresponding to the target object.
  • the preset image sampling frequency may be set according to the frequency at which the photographing device captures the image to be detected, and different photographing devices have different photographing frequencies. Therefore, the image sampling frequency is not limited in this embodiment of the present disclosure.
  • the computer device can determine the estimated frequency of the flashing lights of the car lights according to the empirical values of the flickering frequencies of various types of car lights (such as the flickering frequency of the left car lights in left turning); after that, according to the image acquisition Frequency and Estimated Frequency to determine the period at which the lights flash. For example, if it is predicted that the vehicle is in the state of turning left, and the period of left turning is 1, then the flickering status of the lights of the left turning vehicle is turning left, moving forward, turning left, moving forward, ...; when the period is 2, the vehicle turning left The flashing conditions of the light are turn left, turn left, move forward, move forward, turn left, turn left, move forward, move forward, ....
  • the number corresponding to the period is the frame number of the image. For example, if the period is 1, it means that every frame is cycled to the left turn state; that is, the state of the light corresponding to the vehicle in the first frame of the target image is left turn, and The state of the light corresponding to the vehicle in the frame target image is moving forward, the state of the light corresponding to the vehicle in the third frame of target image is turning left, the state of the light corresponding to the vehicle in the fourth frame of target image is moving forward, ... .
  • the standard state sequence of each preset state corresponding to the target object can be accurately determined according to the period of the vehicle light flashing.
  • Historical status information namely left turn, left turn, forward, forward, left turn, left turn, forward, forward, left turn, left turn; the end of the state of the light turning left Historical status information corresponding to other vehicle light statuses other than turning.
  • the predicted state information corresponding to the target object is determined.
  • the known state sequence information includes 20 historical state information, which are brake, forward, forward, left turn, left turn, forward, forward, left turn, left turn, front Go, move forward, turn left, turn left, move forward, move forward, turn left, turn left, move forward, move forward, turn left; the subsequence of turning on the state of the known car light turning left includes 10 historical state information, That is, turn left, turn left, move forward, move forward, turn left, turn left, move forward, move forward, turn left, turn left; the state of the car light turning left ends.
  • the historical state information corresponding to the state of the car light; the known flickering period of the car light is 2.
  • the standard state sequence and the cycle of the lights flashing predict the state of the lights of the vehicle in the 21st frame of the image to be detected. From the 1st car light state to the 20th car light state, turn left, turn left, forward, forward continuously, and the sequence corresponding to the continuous cycle includes the subsequence of turning on the state of the car light turning left, and it can be determined that the left turning is turned on. According to the 20th
  • the historical state information is the state of turning left, and the cycle of the lights flickering is 2. It can be predicted that the predicted state information corresponding to the target object on the newly acquired image to be detected is turning left.
  • the flickering cycle of the car lights can be obtained, and the standard state sequence of each preset state corresponding to the target object can be accurately determined according to the flickering cycle of the car lights.
  • the accurate standard state sequence and the timing information The historical status information that has been arranged can predict more accurate forecast status information.
  • the implementation of determining the sub-object image may include: identifying the image to be detected, and obtaining the position information of each preset part in the image to be detected among the plurality of preset parts contained in the target object ; Afterwards, based on the position information, the computer device can respectively intercept sub-images containing each preset part from the image to be detected to obtain a sub-object image corresponding to each preset part. After that, multiple sub-object images are spliced.
  • the sub-object images corresponding to each preset part can be spliced to obtain a target image.
  • the computer device can use the vehicle light detection model to perform vehicle light recognition on the image to be detected, and obtain the detection frame corresponding to each vehicle light component, that is, each vehicle light component in the image to be detected After that, the sub-image framed by the detection frame corresponding to each lamp assembly is intercepted from the image to be detected to obtain the sub-object image corresponding to each lamp assembly. After that, the computer device can determine the splicing position of each sub-object image according to the relative positional relationship of each detection frame in the image to be detected, and then splice the sub-object images according to the splicing position to obtain the target image.
  • the sub-object images corresponding to the preset parts are spliced according to the position information, and the obtained target image not only focuses on the preset parts, but also retains the relative position information of the preset parts on the target object, that is, retains the preset
  • the structure information corresponding to the position therefore, the detection based on the target image can improve the efficiency and accuracy of the state detection.
  • the writing order of each step does not mean a strict execution order and constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possible
  • the inner logic is OK.
  • the embodiment of the present disclosure also provides a state detection device corresponding to the state detection method. Since the problem-solving principle of the device in the embodiment of the present disclosure is similar to the above-mentioned state detection method of the embodiment of the present disclosure, the implementation of the device Reference can be made to the implementation of the method, and repeated descriptions will not be repeated.
  • the target state information is determined by combining single-needle image detection with timing optimization; in this way, the computer device can still accurately determine the target when the hardware computing resource capability is limited, or the timing convolution algorithm cannot be supported.
  • the target state information of the object improves the robustness of state detection.
  • FIG. 5 it is a schematic diagram of a state detection device provided by an embodiment of the present disclosure, the device includes: an information acquisition part 401, an initial state detection part 402 and a state optimization part 403; wherein,
  • the information acquisition part 401 is configured to acquire sub-object images corresponding to multiple preset parts contained in the target object in the image to be detected, and state sequence information corresponding to the target object; the state sequence information includes Describe the historical state information of the target object at multiple historical moments;
  • the initial state detection part 402 is configured to stitch multiple sub-object images into a target image, and perform object state detection on the target image to obtain initial state information corresponding to the target object;
  • the state optimization part 403 is configured to determine target state information corresponding to the target object based on the initial state information and the state sequence information.
  • the state sequence information further includes target validity information corresponding to each historical state information
  • the state optimization part 403 is further configured to, when the number of historical state information stored in the state sequence information is greater than a preset number, based on a plurality of the historical state information and each of the historical state information corresponds to The target validity information of the target object is determined to determine the valid state information corresponding to the target object in the state sequence information;
  • the device includes a sequence updating part 404;
  • the sequence update part 404 is configured to use the initial validity information corresponding to the initial state information as the initial validity information when the amount of historical state information stored in the state sequence information is less than or equal to the preset amount.
  • the target validity information corresponding to the initial state information is added, and the initial state information and the target validity information are added to the sequence corresponding to the state sequence information to obtain the updated state sequence information.
  • sequence updating part 404 is further configured to obtain the initial validity information corresponding to the initial state information as the target validity information corresponding to the initial state information The credibility corresponding to the initial state information;
  • the reliability is greater than a first preset value, it is determined that the initial validity information corresponding to the initial state information is valid.
  • sequence updating part 404 is further configured to, after determining the target state information corresponding to the target object based on the initial state information and the effective state information, determine the target Target validity information corresponding to status information;
  • the target state information and the target validity information are added to a state sequence to obtain updated state sequence information.
  • sequence updating part 404 is further configured to obtain the credibility corresponding to the initial state information
  • the target state information in response to the reliability being less than or equal to the first preset value, the valid state information being the same as the initial state information, and the reliability being greater than a second preset value
  • the corresponding target validity information is valid; wherein, the first preset value is greater than the second preset value.
  • sequence updating part 404 is further configured to respond to the reliability being less than or equal to the first preset value, and the effective status information is different from the initial status information different, determine that the target validity information corresponding to the target state information is invalid;
  • the target validity information corresponding to the target state information is invalid.
  • the state sequence information further includes timing information corresponding to each of the historical state information
  • the state optimization part 403 is further configured to determine the target based on the timing information corresponding to each of the historical state information in the state sequence information and the standard state sequence of each preset state corresponding to the target object.
  • the state optimization part 403 is further configured to obtain a preset image sampling frequency
  • the standard state sequence of each preset state corresponding to the target object Based on the timing information corresponding to each of the historical state information in the state sequence information, the standard state sequence of each preset state corresponding to the target object, and the image sampling frequency, determine the prediction corresponding to the target object State information; wherein, the standard state sequence includes a state start subsequence and a state end subsequence of corresponding preset states.
  • the information acquiring part 401 is further configured to identify the image to be detected, and obtain the position of each preset part in the multiple preset parts included in the target object. in the image to be detected;
  • the initial state detection part 402 is further configured to splice the sub-object images corresponding to each preset part according to the position information of each preset part in the image to be detected to obtain the target image.
  • a "part" may be a part of a circuit, a part of a processor, a part of a program or software, etc., of course it may also be a unit, a module or a non-modular one.
  • FIG. 6 it is a schematic structural diagram of a computer device provided by an embodiment of the present disclosure, including:
  • Processor 51 memory 52 and bus 53 .
  • the memory 52 stores machine-readable instructions executable by the processor 51
  • the processor 51 is used to execute the machine-readable instructions stored in the memory 52.
  • the processor 51 executes The following steps: S101: Obtain the sub-object images corresponding to multiple preset parts contained in the target object in the image to be detected, and the state sequence information corresponding to the target object; the state sequence information includes information used to characterize the target object at multiple historical moments historical state information; S102: Concatenate multiple sub-object images into a target image, and perform object state detection on the target image to obtain the initial state information corresponding to the target object; S103: Determine the corresponding target object based on the initial state information and state sequence information target state information.
  • Above-mentioned storer 52 comprises internal memory 521 and external memory 522;
  • Internal memory 521 here is also called internal memory, is used for temporarily storing computing data in processor 51, and the data exchanged with external memory 522 such as hard disk, and processor 51 communicates with external memory 521 through internal memory 521.
  • the external memory 522 performs data exchange.
  • the processor 51 communicates with the memory 52 through the bus 53, so that the processor 51 executes the execution instructions mentioned in the above method embodiments.
  • Embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the state detection method described in the foregoing method embodiments are executed.
  • the storage medium may be a volatile or non-volatile computer-readable storage medium.
  • An embodiment of the present disclosure further provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, the steps of the above state detection method are realized.
  • the computer program product can be any product that can realize the above state detection method, and some or all of the solutions in the computer program product that contribute to the prior art can be implemented as a software product (such as a software development kit (Software Development Kit, SDK) ), the software product may be stored in a storage medium, and the included computer instructions cause the relevant equipment or processor to perform some or all of the steps of the above state detection method.
  • the disclosed devices and methods may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the modules is only a logical function division.
  • multiple modules or components can be combined.
  • some features can be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces, and the indirect coupling or communication connection of devices or modules may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional module in each embodiment of the present disclosure may be integrated into one processing module, each module may exist separately physically, or two or more modules may be integrated into one module.
  • the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor.
  • the technical solution of the present disclosure is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disc and other media that can store program codes. .
  • the image of the non-detection part is removed, the image of the preset part to be detected (that is, the sub-object image) is focused. Therefore, the object state detection is performed on the target image, and the accurate target object corresponding Initial state information. Moreover, since the historical state information at multiple historical moments is associated with the current initial state information, optimizing the initial state information in combination with the historical state information at multiple historical moments can improve the accuracy of the target state information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Traffic Control Systems (AREA)
  • Image Analysis (AREA)

Abstract

本公开提供了一种状态检测方法、装置、计算机设备、存储介质和程序产品，其中，该方法包括：获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及目标对象对应的状态序列信息；状态序列信息包括用于表征目标对象在多个历史时刻的历史状态信息；将多个子对象图像拼接为目标图像，并对目标图像进行对象状态检测，得到目标对象对应的初始状态信息；基于初始状态信息和状态序列信息，确定目标对象对应的目标状态信息。

Description

状态检测方法、装置、计算机设备、存储介质和程序产品
相关申请的交叉引用
本公开实施例基于申请号为202111437066.3、申请日为2021年11月29日、申请名称为“一种状态检测方法、装置、计算机设备和存储介质”的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本公开作为参考。
技术领域
本公开涉及图像处理技术领域，尤其涉及一种状态检测方法、装置、计算机设备、存储介质和程序产品。
背景技术
在自动驾驶场景中的车辆意图判定、交通管理中，车灯状态检测对于判断驾驶员是否合法驾驶等领域有着巨大的应用价值。实际中可以通过对包括车灯的图像进行检测，确定出车灯状态。
相关技术中，通常是通过车灯状态分类器对整张车辆图像进行检测，有监督的训练出车灯的状态。由于整张车辆图像包含了较多的车辆信息，因此，利用整张车辆图像进行训练，导致车灯状态的检测精度较低。
发明内容
本公开实施例至少提供一种状态检测方法、装置、计算机设备、存储介质和程序产品，针对检测出的目标对象对应的初始状态信息做进一步的优化，能够得到较高精度的目标状态信息。
第一方面，本公开实施例提供了一种状态检测方法，包括：
获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及所述目标对象对应的状态序列信息；所述状态序列信息包括用于表征所述目标对象在多个历史时刻的历史状态信息；
将多个所述子对象图像拼接为目标图像，并对所述目标图像进行对象状态检测，得到所述目标对象对应的初始状态信息；
基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息。
该方面，将多个子对象图像拼接为目标图像，使得目标图像仅包括目标对象待检测的预设部位的子对象图像，由于去除了非检测部位的图像，聚焦了待检测的预设部位的图像(即子对象图像)，因此，对目标图像进行对象状态检测，能够得到精准的目标对象对应的初始状态信息。之后，利用包括表征目标对象在多个历史时刻的历史状态信息对初始状态信息做进一步的优化处理，由于多个历史时刻的历史状态信息与当前的初始状态信息有关联，因 此，结合多个历史时刻的历史状态信息对初始状态信息进行优化，能够提高优化得到的目标状态信息的准确性。
一种可选的实施方式中，所述状态序列信息还包括每个历史状态信息对应的目标有效性信息；
所述基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息，包括：
在所述状态序列信息中存储的历史状态信息的数量大于预设数量的情况下，基于多个所述历史状态信息以及每个所述历史状态信息对应的目标有效性信息，确定在所述状态序列信息中所述目标对象对应的有效状态信息；
基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息。
该实施方式中，由于在状态序列信息中累计一定数量的历史状态信息，通过多个历史状态信息能够较为准确地反映目标对象在某一历史时段的状态信息，即上述有效状态信息，利用历史状态信息对应的目标有效性信息，比如利用有效的历史状态信息能够较为准确地确定上述有效状态信息。考虑到初始状态信息受各种条件因素影响产生的误差，在优化初始状态信息的过程中，利用有效状态信息优化初始状态信息，能够提高最终确定的目标对象对应的目标状态信息的准确性。
一种可选的实施方式中，还包括：
在所述状态序列信息中存储的历史状态信息的数量小于或等于所述预设数量的情况下，将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息，并将所述初始状态信息以及目标有效性信息添加到状态序列信息对应的序列中，得到更新后的状态序列信息。
该实施方式中，由于状态序列信息中累计的历史状态信息的数量不够，因此无法准确地反映目标对象在某一历史时段的状态信息，此时，需要在状态序列信息中添加历史状态信息。通过不断确定初始状态信息对应的初始有效性信息，并将其添加到状态序列信息对应的序列中，直到更新后的状态序列信息中累计的历史状态信息达到预计数量为止，这样有利于提高确定的目标状态信息的准确性。
一种可选的实施方式中，在所述将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息之前，还包括确定所述初始有效性信息的步骤：
获取所述初始状态信息对应的可信度；
在所述可信度大于第一预设值的情况下，确定所述初始状态信息对应的初始有效性信息为有效。
该实施方式中，由于第一预设值可以是根据经验值确定出来的，其能够较为准确的评估初始状态信息对应的初始有效性信息是否有效，因此，通过比较初始状态信息对应的可信度与第一预设值之间大小，能够准确地判定出初始有效性信息是否有效，如果可信度大于第一预设值，即判定初始有效性 信息为有效。
一种可选的实施方式中，在基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息之后，还包括：
确定所述目标状态信息对应的目标有效性信息；
按照所述目标状态信息确定的时间顺序，将所述目标状态信息以及所述目标有效性信息添加到状态序列中，得到更新后的状态序列信息。
该实施方式，目标状态信息是优化后的目标对象的状态信息，将优化后的目标状态信息以及目标状态信息对应的目标有效性信息添加到状态序列信息中，能够提高状态序列信息中的历史状态信息(包括已添加的目标状态信息)的有效性。
一种可选的实施方式中，所述确定所述目标状态信息对应的目标有效性信息，包括：
获取所述初始状态信息对应的可信度；
响应于所述可信度大于第一预设值，确定所述目标状态信息对应的目标有效性信息为有效；
响应于所述可信度小于或等于所述第一预设值、所述有效状态信息与所述初始状态信息相同，并且所述可信度大于第二预设值，确定所述目标状态信息对应的目标有效性信息为有效；其中，所述第一预设值大于所述第二预设值。
该实施方式，第一预设值和第二预设值可以是根据经验值确定出来的，其能够较为准确的评估目标状态信息对应的目标有效性信息是否有效；结合第一预设值、第二预设值、有效状态信息以及所述初始状态信息，能够较为准确地确定目标有效性信息。
一种可选的实施方式中，所述确定所述目标状态信息对应的目标有效性信息，还包括：
响应于所述可信度小于或等于所述第一预设值，并且所述有效状态信息与所述初始状态信息不相同，确定所述目标状态信息对应的目标有效性信息为无效；
响应于所述可信度小于或等于所述第一预设值，并且所述可信度小于或等于所述第二预设值，确定所述目标状态信息对应的目标有效性信息为无效。
该实施方式，可信度较小(小于或等于第二预设值)，或者可信度较大(大于第二预设值，但是小于或等于第一预设值)，并且有效状态信息与初始状态信息不相同时，目标状态信息大概率为无效，因此结合第一预设值、第二预设值、有效状态信息以及所述初始状态信息，能够较为准确地确定目标有效性信息。
一种可选的实施方式中，所述状态序列信息还包括每个所述历史状态信息对应的时序信息；
所述基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息，包括：
基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息；
基于所述初始状态信息、所述有效状态信息和所述预测状态信息，确定所述目标对象对应的目标状态信息。
该实施方式，考虑到初始状态信息受各种条件因素影响产生的误差，在优化初始状态信息的过程中，还可以参考理论上估计出的状态信息，即预测状态信息，因此，这里结合预测状态信息和有效状态信息同时对初始状态信息做进一步的优化处理，能够提高最终确定的目标对象对应的目标状态信息的准确性。
一种可选的实施方式中，所述基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息，包括：
获取预先设置的图像采样频率；
基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，和所述图像采样频率，确定所述目标对象对应的预测状态信息；其中，所述标准状态序列包括对应的预设状态的状态开启子序列和状态结束子序列。
该实施方式，根据图像采样频率能够准确地确定目标对象对应的每种预设状态的标准状态序列，结合准确的标准状态序列和按照时序信息已经排列好的历史状态信息，能够预测出较为精准的预测状态信息。
一种可选的实施方式中，所述获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，包括：
对所述待检测图像进行识别，得到所述目标对象所包含的多个预设部位中每个预设部位在所述待检测图像中的位置信息；
基于所述位置信息，从所述待检测图像中分别截取包含每个预设部位的子图像，得到每个预设部位对应的所述子对象图像；
所述将多个所述子对象图像拼接为目标图像，包括：
按照每个预设部位在所述待检测图像中的位置信息，将每个预设部位对应的子对象图像进行拼接，得到所述目标图像。
该实施方式，根据位置信息拼接预设部位对应的子对象图像，得到的目标图像既聚焦了预设部位，同时也保留了预设部位在目标对象上的相对位置信息，即保留了预设位置对应的结构信息，因此，基于目标图像进行检测能够提高状态检测的效率以及精确度。
第二方面，本公开实施例还提供一种状态检测装置，包括：
信息获取部分，被配置为获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及所述目标对象对应的状态序列信息；所述状态序列信息包括用于表征所述目标对象在多个历史时刻的历史状态信息；
初始状态检测部分，被配置为将多个所述子对象图像拼接为目标图像， 并对所述目标图像进行对象状态检测，得到所述目标对象对应的初始状态信息；
状态优化部分，被配置为基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述状态序列信息还包括每个历史状态信息对应的目标有效性信息；
所述状态优化部分，还被配置为在所述状态序列信息中存储的历史状态信息的数量大于预设数量的情况下，基于多个所述历史状态信息以及每个所述历史状态信息对应的目标有效性信息，确定在所述状态序列信息中所述目标对象对应的有效状态信息；
基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述装置包括序列更新部分；
所述序列更新部分，被配置为在所述状态序列信息中存储的历史状态信息的数量小于或等于所述预设数量的情况下，将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息，并将所述初始状态信息以及目标有效性信息添加到状态序列信息对应的序列中，得到更新后的状态序列信息。
一种可选的实施方式中，所述序列更新部分，还被配置为在所述将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息之前，获取所述初始状态信息对应的可信度；
在所述可信度大于第一预设值的情况下，确定所述初始状态信息对应的初始有效性信息为有效。
一种可选的实施方式中，所述序列更新部分，还被配置为在基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息之后，确定所述目标状态信息对应的目标有效性信息；
按照所述目标状态信息确定的时间顺序，将所述目标状态信息以及所述目标有效性信息添加到状态序列中，得到更新后的状态序列信息。
一种可选的实施方式中，所述序列更新部分，还被配置为获取所述初始状态信息对应的可信度；
响应于所述可信度大于第一预设值，确定所述目标状态信息对应的目标有效性信息为有效；
响应于所述可信度小于或等于所述第一预设值、所述有效状态信息与所述初始状态信息相同，并且所述可信度大于第二预设值，确定所述目标状态信息对应的目标有效性信息为有效；其中，所述第一预设值大于所述第二预设值。
一种可选的实施方式中，所述序列更新部分，还被配置为响应于所述可信度小于或等于所述第一预设值，并且所述有效状态信息与所述初始状态信息不相同，确定所述目标状态信息对应的目标有效性信息为无效；
响应于所述可信度小于或等于所述第一预设值，并且所述可信度小于或等于所述第二预设值，确定所述目标状态信息对应的目标有效性信息为无效。
一种可选的实施方式中，所述状态序列信息还包括每个所述历史状态信息对应的时序信息；
所述状态优化部分，还被配置为基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息；
基于所述初始状态信息、所述有效状态信息和所述预测状态信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述状态优化部分，还被配置为获取预先设置的图像采样频率；
基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，和所述图像采样频率，确定所述目标对象对应的预测状态信息；其中，所述标准状态序列包括对应的预设状态的状态开启子序列和状态结束子序列。
一种可选的实施方式中，所述信息获取部分，还被配置为对所述待检测图像进行识别，得到所述目标对象所包含的多个预设部位中每个预设部位在所述待检测图像中的位置信息；
基于所述位置信息，从所述待检测图像中分别截取包含每个预设部位的子图像，得到每个预设部位对应的所述子对象图像；
所述初始状态检测部分，还被配置为按照每个预设部位在所述待检测图像中的位置信息，将每个预设部位对应的子对象图像进行拼接，得到所述目标图像。
第三方面，本公开实施例还提供一种计算机设备，包括：处理器、存储器和总线，所述存储器存储有所述处理器可执行的机器可读指令，当计算机设备运行时，所述处理器与所述存储器之间通过总线通信，所述机器可读指令被所述处理器执行时执行上述第一方面，或第一方面中任一种可能的状态检测方法的步骤。
第四方面，本公开实施例还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行上述第一方面，或第一方面中任一种可能的状态检测方法的步骤。
第五方面，本公开实施例还提供一种计算机程序产品，包括计算机可读代码，所述计算机程序产品包括计算机程序或指令，在所述计算机程序或指令在计算机设备上运行的情况下，使得所述计算机设备执行上述第一方面，或第一方面中任一种可能的状态检测方法的步骤。
关于上述状态检测装置、计算机设备和存储介质的效果描述参见上述状态检测方法的说明，这里不再赘述。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开。
为使本公开的上述目的、特征和优点能更明显易懂，下文特举较佳实施例，并配合所附附图，作详细说明如下。
附图说明
为了更清楚地说明本公开实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，此处的附图被并入说明书中并构成本说明书中的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。应当理解，以下附图仅示出了本公开的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他相关的附图。
图1示出了本公开实施例所提供的一种状态检测方法的流程图；
图2示出了本公开实施例所提供的车灯状态检测过程的流程示意图；
图3示出了本公开实施例所提供的一种确定目标状态信息的流程图；
图4示出了本公开实施例所提供的一种确定目标状态信息的流程图；
图5示出了本公开实施例所提供的一种状态检测装置的示意图；
图6示出了本公开实施例所提供的一种计算机设备的结构示意图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例中附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。通常在此处附图中描述和示出的本公开实施例的组件可以以各种不同的配置来布置和设计。因此，以下对在附图中提供的本公开的实施例的详细描述并非旨在限制要求保护的本公开的范围，而是仅仅表示本公开的选定实施例。基于本公开的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
另外，本公开实施例中的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。
在本文中提及的“多个或者若干个”是指两个或两个以上。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。
通常，车灯状态检测在实际中可以通过对包括车灯的图像进行检测，确定出车灯状态。相关技术是通过车灯状态分类器对整张车辆图像进行检测，有监督的训练出车灯的状态。由于整张车辆图像包含了较多的车辆信息，因此，利用整张车辆图像进行训练，导致车灯状态的检测精度较低。
本公开提供了一种状态检测方法，将多个子对象图像拼接为目标图像，使得目标图像仅包括目标对象待检测的预设部位的子对象图像，由于去除了 非检测部位的图像，聚焦了待检测的预设部位的图像(即子对象图像)，因此，对目标图像进行对象状态检测，能够得到精准的目标对象对应的初始状态信息。之后，利用包括表征目标对象在多个历史时刻的历史状态信息对初始状态信息做进一步的优化处理，由于多个历史时刻的历史状态信息与当前的初始状态信息有关联，因此，结合多个历史时刻的历史状态信息对初始状态信息进行优化，能够提高优化得到的目标状态信息的准确性。
针对以上方案所存在的缺陷，均是发明人在经过实践并仔细研究后得出的结果，因此，上述问题的发现过程以及下文中本公开针对上述问题所提出的解决方案，都应该是发明人在本公开过程中对本公开做出的贡献。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步定义和解释。
为便于对本实施例进行理解，首先对本公开实施例所公开的一种状态检测方法进行详细介绍，本公开实施例所提供的一种状态检测方法的执行主体一般为具有一定计算能力的计算机设备。其中，计算机设备指的可以是服务器、笔记本电脑、平板电脑、台式计算机、智能电视、机顶盒、移动设备(例如移动电话、便携式视频播放器、个人数字助理、专用消息设备、便携式游戏设备)等具备状态检测能力的设备。在一些可能的实现方式中，该状态检测方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
下面以执行主体为计算机设备为例对本公开实施例提供的状态检测方法加以说明。
首先对本公开实施例所公开的一种状态检测方法的应用场景进行介绍，本公开实施例提供的状态检测方法可以应用于检测机动车灯的车灯状态，车灯状态的基本状态包括左转、右转、双闪、刹车、不转(左、右车灯均灭，可能是白天的前行，这种状态需要做进一步的优化处理才能确定最终的车灯状态)。车灯检测的状态包括左转、右转、双闪、不转这四种基本状态，与刹车、刹车/前行(左、右车灯均亮，可能是白天的刹车，也可能是晚上的前行，这种状态需要做进一步的优化处理)、不刹车这三种状态的排列，确定12种不同的车灯检测状态，即左转+刹车、右转+刹车、双闪+刹车、不转+刹车(即为刹车)、左转+刹车/前行、右转+刹车/前行、双闪+刹车/前行、不转+刹车/前行(即为刹车前行)、左转+不刹车(即为左转)、右转+不刹车(即为右转)、双闪+不刹车(即为双闪)、不转+不刹车(即为不转)。
参见图1所示，为本公开实施例提供的状态检测方法的流程图，所述方法包括步骤S101～S103，其中：
S101：获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及目标对象对应的状态序列信息；状态序列信息包括用于表征目标对象在多个历史时刻的历史状态信息。
本步骤中，待检测图像包括目标对象，目标对象为等待状态检测的对象，比如车辆、行人重识别(Person re-identification，reid)中的特定行人、3D单 目目标检测(monocular 3d object detection，mono3D)中的组件等。
以目标对象为车辆为例，预设部位包括车灯组件，比如左车灯组件、右车灯组件、刹车灯组件(即顶灯组件)等。其中，左车灯组件包括左转向灯、雾灯和示廓灯等。左车灯组件包括右转向灯、雾灯和示廓灯等。刹车灯组件包括刹车灯等。需要说明的是，不同类型的车辆，车灯布局方式不同，比如，某一类型车辆没有刹车灯组件。目标对象所包含的多个预设部位对应的子对象图像可以包括车灯对应的多个子对象图像。
车灯组件安装在车辆中的不同位置，利用车灯检测模型能够检测出车灯组件以及每个车灯组件所在的位置，得到位置信息；之后，根据位置信息，从待检测图像中将车灯组件对应的图像裁剪出来，得到多个子对象图像，比如，左车灯组件图像、右车灯组件图像、刹车灯组件图像。
状态序列信息30包括用于表征目标对象在多个历史时刻的历史状态信息31。状态序列信息对应的状态序列包括按照时序信息顺序排列的多个历史状态信息对应的历史状态。如图2所示，多个历史状态信息31按照时序信息t进行排列，从上到下时序依次为t-1，t-2，t-3，……，t-n。
这里，历史状态可以包括历史时刻获取到的初始状态信息对应的初始状态经过下述优化过程(S102～S103)优化后的状态，确定历史状态信息的过程可参见下述S102～S103中对初始状态信息优化为目标状态信息并添加到状态序列信息的描述，重复部分在此不再赘述。这里，将目标状态信息添加到状态序列信息后，目标状态信息就变成了历史状态信息。
S102：将多个子对象图像拼接为目标图像，并对目标图像进行对象状态检测，得到目标对象对应的初始状态信息。
以车灯状态检测为例，参见图2所示，其为车灯状态检测过程的流程示意图，首先，利用车灯检测模型21(比如Faster RCNN等神经网络模型，利用眼部车灯样本图像进行训练)检测待检测图像20中的车灯组件，输出包含车灯组件的子对象图像，其中，子对象图像包括左车灯组件图像22、右车灯组件图像23、刹车灯组件图像24、左转向灯组件图像25、右转向灯图像26；之后，将上述多个子对象图像按照每个车灯组件在车辆上的位置顺序进行拼接，确定目标图像27；之后，将目标图像27输入到车灯状态检测模型(比如Faster RCNN等神经网络模型，利用车灯不同状态样本图像进行训练)28进行车灯状态检测，确定车辆对应的初始状态信息29。
将上述多个子对象图像按照每个车灯组件在车辆上的位置顺序进行拼接，可以包括：将左车灯组件图像22拼接在目标图像27中部左侧，右车灯组件图像23拼接在目标图像27中部右侧，刹车灯组件图像24拼接在目标图像27顶部，左转向灯组件图像25拼接在目标图像27下部左侧，右转向灯图像26拼接在目标图像27下部右侧；由于图2中车辆的左车灯组件中包括左转向灯，右车灯组件包括右转向灯，无刹车灯，因此，刹车灯组件图像24，左转向灯组件图像25和右转向灯图像26，无实质内容可展示，即为黑框。
这里，由于减少了针对车辆中一些其他组件(即非车灯组件)的检测， 因此，能够提高车灯状态检测的效率。同时，只针对车灯组件进行车灯状态检测，由于降低了非车灯组件干扰因素的影响，因此，能够降低车灯状态检测的误差，提高车灯状态检测的精度。
车灯状态检测模型28输出车辆对应的初始状态信息，以及初始状态信息的可信度，其中，可信度越低车灯状态检测模型输出的初始状态信息越不可信，即车灯状态检测模型检测误差较大。
初始状态信息包括上述列举的12种不同车灯检测状态对应的状态信息，比如标签信息，即每个车辆检测状态对应一个标签，车灯状态检测模型输出标签，标签用于指示当前检测到的车灯的状态。示例性的，预先设置前行对应的标签为0，左转对应的标签为1，右转对应的标签为2，刹车对应的标签为3，……。
S103：基于初始状态信息和状态序列信息，确定目标对象对应的目标状态信息。
这里，确定目标状态信息的流程图可以参见图2所示，其中，32表示目标状态信息，可以结合初始状态信息29和状态序列信息30得到。
本步骤中，状态序列信息还包括每个历史状态信息对应的目标有效性信息，其中，目标有效性信息为评估历史状态信息是否有效的信息。当目标有效性信息为有效的情况下，即该目标有效性信息对应的历史状态信息有效。“有效”表示车灯状态可用于对初始状态信息对应的初始状态进行优化处理；“无效”表示检测的车灯状态不可用于对初始状态信息对应的初始状态进行优化处理。需要说明的是，前行状态为预先设置好的无效状态。
在一种可能的实施方式中，可以利用包括表征目标对象在多个历史时刻的历史状态信息对初始状态信息做进一步的优化处理，由于多个历史时刻的历史状态信息与当前的初始状态信息有关联，因此，结合多个历史时刻的历史状态信息对初始状态信息进行优化，可以减少遮挡、漏检或稠密场景中对车灯检测的影响，能够提高优化得到的目标状态信息的准确性。在一些实施例中，计算机设备可以基于多个历史状态信息以及每个历史状态信息对应的目标有效性信息，确定在状态序列信息中目标对象对应的有效状态信息。之后，基于初始状态信息和有效状态信息，确定目标对象对应的目标状态信息。
这里，有效状态信息用于表征目标对象在历史时刻的有效历史状态，有效历史状态可以是从状态序列信息指示的状态序列中筛选出的，也即从多个历史状态信息中筛选出的有效的历史状态信息。
确定目标对象对应的有效状态信息包括如下步骤：
S1031、确定序列中有效的历史状态信息，并从有效的历史状态信息中筛选出有效数量最多的历史状态信息A，可以将该历史状态信息A作为有效状态信息。
S1032、确定序列中有效的历史状态信息，在从有效的历史状态信息中筛选出有效数量最多的多个历史状态信息B和历史状态信息C的情况下，计算历史状态信息B对应的可信度(车灯状态检测模型输出的可信度)之和、以 及历史状态信息C对应的可信度之和，并比较大小，将可信度之和较大的历史状态信息作为有效状态信息。
示例性的，以车灯状态检测为例，利用状态序列信息中包括的多个历史状态信息优化车灯状态检测模型输出的车辆的初始状态信息，例如，在车灯状态检测模型输出标签为1的情况下，确定车辆的初始状态信息m(t)＝1，即初始状态为左转，其中，t表示帧数，m(t)表示检测第t帧待检测图像确定的车辆对应的初始状态信息。已知状态序列信息对应的序列中包括多个历史状态信息以及每个历史状态信息对应的目标有效性信息，比如，左转(有效)、前行(无效)、左转(有效)、前行(无效)、左转(有效)、前行(无效)、前行(无效)、前行(无效)，刹车(有效)、刹车(有效)，确定有效状态信息为左转，标签为1，即有效状态信息p(t)＝1，p(t)表示检测第t帧待检测图像确定的车辆对应的有效状态信息。之后，根据初始状态信息m(t)＝1和有效状态信息p(t)＝1，确定目标对象对应的目标状态信息y(t)，y(t)表示检测第t帧待检测图像确定的车辆对应的目标状态信息，可以参见公式一：
y(t)＝αm(t)+βp(t)        公式一
其中，α表示车灯状态检测模型输出的初始状态信息在优化过程中所占权重；β表示有效状态信息在优化过程中所占权重；其中，α+β＝1。因此，计算得到y(t)＝α+β＝1，即初始状态信息优化后的目标状态信息为左转。
在一些实施例中，计算机设备可以在状态序列信息中存储的历史状态信息的数量大于预设数量的情况下，基于多个历史状态信息以及每个历史状态信息对应的目标有效性信息，确定目标对象对应的有效状态信息；基于初始状态信息和有效状态信息，确定在状态序列信息中目标对象对应的目标状态信息。如此，在状态序列信息中累计一定数量的历史状态信息后，可以通过多个历史状态信息反映目标对象在某一历史时段的状态信息，即有效状态信息，从而提高目标对象的状态信息的准确性。
这里，预设数量可以根据拍摄设备采集待检测图像时的图像采样频率设定，本公开实施例不进行限定。预设数量一般大于某一预设状态的标准状态序列中历史状态信息的数量，以使预设数量的历史状态信息所组成的状态序列信息能够表示一段时间内目标对象对应的预设状态。这里，预设状态表示一个执行行为状态，比如左转向行为对应的左转向状态，包括多个历史状态信息，比如，左转、前行、左转、前行、左转、前行，确定左转向开启。预设状态的标准状态序列可以包括对应的预设状态的状态开启子序列和状态结束子序列。
示例性的，预设状态可以包括左转向、右转向、前行、刹车和双闪等。
示例性的，车灯左转向的状态开启子序列包括6个历史状态信息，比如左转、前行、左转、前行、左转、前行，确定左转向开启。左转状态结束子序列包括4个历史状态信息，即除了左转以外的其他车灯状态，比如前行、前行、刹车、刹车，确定左转向结束。
由于确定预设状态需要多个历史状态信息，因此，需要状态序列信息中存储的历史状态信息的数量大于预设数量，才能较为准确地确定反映目标对象对应的预设状态，进而较为准确地确定反映目标对象对应的有效状态信息。
之后，确定目标对象对应的目标状态信息的说明可以参见上述公式一中的详细内容，重复部分在此不再赘述。
在一些实施例中，在状态序列信息中存储的历史状态信息的数量小于或等于预设数量的情况下，将初始状态信息对应的初始有效性信息作为初始状态信息对应的目标有效性信息，并将初始状态信息以及目标有效性信息添加到状态序列信息对应的序列中，得到更新后的状态序列信息。
这里，由于状态序列信息中累计的历史状态信息的数量不够，因此无法准确地反映目标对象在某一历史时段的状态信息，此时，需要在状态序列信息中添加历史状态信息。在本公开实施例中，不断确定初始状态信息对应的初始有效性信息，并将其添加到状态序列信息对应的序列中，直到更新后的状态序列信息中累计的历史状态信息达到预计数量为止，这样有利于提高确定的目标状态信息的准确性。
其中，初始有效性信息为评估车灯状态检测模型输出的初始状态信息是否有效的信息，在初始有效性信息为有效的情况下，该初始有效性信息对应的初始状态信息有效。
在本公开的一些实施例中，计算机设备确定初始状态信息对应的初始有效性信息，可以包括：获取初始状态信息对应的可信度；在可信度大于第一预设值的情况下，确定初始状态信息对应的初始有效性信息为有效。
这里，第一预设值可以根据经验值设定，本公开实施例不进行限定。
由于第一预设值可以是根据经验值确定出来的，其能够较为准确的评估初始状态信息对应的初始有效性信息是否有效，因此，通过比较初始状态信息对应的可信度与第一预设值之间大小，能够准确地判定出初始有效性信息是否有效，即在可信度大于第一预设值的情况下，可以判定初始有效性信息为有效。
针对S103，在确定了目标状态信息之后，还需要进一步确定目标状态信息对应的目标有效性信息；之后，按照目标状态信息确定的时间顺序，将目标状态信息以及目标有效性信息添加到状态序列中，得到更新后的状态序列信息。能够提高状态序列信息中的历史状态信息(包括已添加的目标状态信息)的有效性。
这里，状态序列信息中包括的历史状态信息可以是按照历史状态信息被确定的时间顺序排列的，因此，状态序列信息还包括历史状态信息对应的时序信息，所述时序信息即为确定历史状态信息的时间顺序。如图2所示，按照时间顺序从下到上依次添加历史状态信息。在确定了目标状态信息以及目标状态信息对应的目标有效性信之后，根据确定的目标状态信息的时间，将目标状态信息以及目标有效性信息顺序添加到状态序列中，也即将目标状态信息添加在图2中状态序列信息30所在区域的最上方，同时更新后的状态序 列信息还包括目标状态信息对应的时序信息。
另外，利用优化后的目标状态信息以及目标状态信息对应的目标有效性信息，更新状态序列信息，能够提高优化下一帧目标图像中目标对象对应的初始状态信息的准确性。
在本公开的一些实施例中，计算机设备确定目标状态信息对应的目标有效性信息，可以包括：首先，可以获取初始状态信息对应的可信度。这里，初始状态信息为目标状态信息优化前的、车灯状态检测模型输出的初始状态信息。该初始状态信息对应的可信度可以作为判断目标有效性信息是否有效的依据。
这里，初始状态信息对应的可信度可以作为目标有效性信息有效性判断的首要依据。有效性判断的方式可以包括：
方式1、响应于可信度大于第一预设值，确定目标状态信息对应的目标有效性信息为有效。
这里，由于可信度作为目标有效性信息有效性判断的首要依据，因此，在确定可信度大于第一预设值的情况下，可以直接确定目标状态信息对应的目标有效性信息为有效，不再进行其他条件的判断。
如果可信度小于或等于第一预设值，即首要依据不满足预设条件，还可以根据次要依据继续进行目标有效性信息有效性的判断。次要依据包括有效状态信息与初始状态信息之间的比较，以及可信度与第二预设值之间的比较。
方式2、响应于可信度小于或等于第一预设值、有效状态信息与初始状态信息相同，并且可信度大于第二预设值，确定目标状态信息对应的目标有效性信息为有效。其中，第一预设值大于第二预设值。
这里，一旦有效状态信息与初始状态信息相同，则可以降低可信度的重要程度，即可信度大于第二预设值，即可确定目标状态信息对应的目标有效性信息为有效。
方式3、响应于可信度小于或等于第一预设值，并且有效状态信息与初始状态信息不相同，确定目标状态信息对应的目标有效性信息为无效。
方式4、响应于可信度小于或等于所述第一预设值，并且可信度小于或等于第二预设值，确定目标状态信息对应的目标有效性信息为无效。
上述方式3和方式4说明，在首要依据不满足条件的情况下，次要依据中的任意一个比较结果不满足条件的情况下，均判定目标有效性信息为无效。
上述，第一预设值和第二预设值可以是根据经验值确定出来的，本公开实施例不进行限定，第一预设值和第二预设值能够较为准确的评估目标状态信息对应的目标有效性信息是否有效；结合第一预设值、第二预设值、有效状态信息以及初始状态信息，提高了确定目标有效性信息的准确性。
在另一种可能的实施方式中，考虑到初始状态信息受各种条件因素影响产生的误差，在优化初始状态信息的过程中，还可以参考理论上估计出的状态信息，即预测状态信息。参见图3所示，其为一种确定目标状态信息的方法流程图，包括S301～S302。
S301、基于状态序列信息中每个历史状态信息对应的时序信息、目标对象对应的每种预设状态的标准状态序列，确定目标对象对应的预测状态信息。
本步骤中，时序信息包括历史状态信息在状态序列信息对应的序列中的排列顺序，按照每个历史状态信息存储时的先后顺序进行排列。目标对象对应的每种预设状态、标准状态序列、以及标准状态序列包括对应的预设状态的状态开启子序列和状态结束子序列，参见上述的详细说明，重复部分在此不再赘述。
预测状态信息预测的是最新获取的待检测图像上的目标对象对应的状态信息。
示例性的，已知状态序列信息中包括20个历史状态信息，按照时序信息排列分别为刹车、前行、前行、前行、左转、前行、左转、前行、左转、前行、左转、前行、左转、前行、左转、前行、左转、前行、左转、前行，默认车灯闪烁的周期为1(下述有针对车灯闪烁的周期的详细说明)。已知车灯左转向的状态开启子序列包括6个历史状态信息，即左转、前行、左转、前行、左转、前行；车灯左转向的状态结束启子序列包括4个除了左转以外的其他车灯状态对应的历史状态信息。根据时序信息和标准状态序列，预测第21帧待检测图像中车辆的车灯状态，例如，在确定第20个历史状态信息为前进状态的情况下，由于从第5个车灯状态到第20个车灯状态不断循环左转和前进，且不断循环对应的序列中包括车灯左转向的状态开启子序列，可以确定左转向开启，根据第20个历史状态信息为前进状态，能够预测最新获取的待检测图像上的目标对象对应的预测状态信息为左转。
S302、基于初始状态信息、有效状态信息和预测状态信息，确定目标对象对应的目标状态信息。
确定预测状态信息s(t)，s(t)表示检测第t帧待检测图像时预测的车辆对应的预测状态信息。根据公式二，确定目标对象对应的目标状态信息y(t)：
y(t)＝γm(t)+δp(t)+εs(t)        公式二
其中，γ表示车灯状态检测模型输出的初始状态信息在优化过程中所占权重；δ表示有效状态信息在优化过程中所占权重，p(t)为周期项，表征有效状态信息的周期性；ε表示预测状态信息在优化过程中所占权重；s(t)为趋势项，表征时序队列中出现的有效属性；其中，γ+δ+ε＝1。
延续上例，已知状态序列信息中包括20个历史状态信息，车灯状态检测模型输出标签为1，即m(t)＝1，根据序列中多个历史状态信息的目标有效性，确定左转有效个数为8，刹车有效个数为1，前进无效个数为11，因此，确定有效状态信息为左转，即标签为1，p(t)＝1，已知上例预测状态信息为左转，即标签为1，s(t)＝1；之后，根据公式二计算得到y(t)＝γ+δ+ε＝1，即初始状态信息优化后的目标状态信息为左转。
示例性的，本公开实施例示出了一种确定目标状态信息的方法流程示意图，如图4所示，该方法可以包括如下步骤：
S1、从采集的视频帧序列中获取当前待检测图像；
S2、通过状态检测模型检测当前待检测图像的初始状态信息；
S3、判断历史状态队列中的历史状态信息数量是否大于或者等于l1；若是，则执行S4-S7；否则执行S8；
S4、确定目标状态信息的目标有效性信息，得到目标对象的有效状态信息；
S5、判断历史状态队列中的历史状态信息数量是否大于或者等于l2；若是，则执行S6-S7；否则执行S8；
在S5中，l1大于l2，l1和l2用于控制状态之间的切换时长。
S6、基于历史状态队列中每个历史状态信息，确定预测状态信息；
S7、通过公式二计算目标状态信息并执行S4，直到视频帧序列结束。
在S7中，计算机设备在得到状态检测模型输出的初始状态信息，以及S4中的有效状态信息，还有S6中的预测状态信息后，可以通过公式二计算目标状态信息。
S8、将初始状态信息对应的初始有效性信息作为初始状态信息对应的目标有效性信息，并将初始状态信息以及目标有效性信息添加到历史状态队列中，更新历史状态队列。
这里结合预测状态信息和有效状态信息，同时对初始状态信息做进一步的优化处理，能够提高最终确定的目标对象对应的目标状态信息的准确性。
在一些实施例中，计算机设备还可以获取预先设置的图像采样频率；之后，基于状态序列信息中每个历史状态信息对应的时序信息、目标对象对应的每种预设状态的标准状态序列，和图像采样频率，确定目标对象对应的预测状态信息。
其中，预先设置的图像采样频率可以根据拍摄设备拍摄待检测图像的频率进行设定，不同拍摄设备的拍摄频率不同，因此，对于图像采样频率，本公开实施例不进行限定。
在本公开实施例中，计算机设备可以根据多种不同类型车灯闪烁的频率(比如左转向中左车灯闪烁的频率)的经验值，确定车灯闪烁的预估频率；之后，根据图像采集频率和预估频率，确定车灯闪烁的周期。比如，预测车辆处于左转向状态下，左转向的周期为1，则左转向的车灯闪烁情况为左转、前行、左转、前行、……；周期为2，则左转向的车灯闪烁情况为左转、左转、前行、前行、左转、左转、前行、前行、……。周期对应的数字为图像的帧数，比如，周期为1，即为每间隔一帧循环到左转状态；也就是说，第一帧目标图像中车辆对应的车灯状态为左转，第二帧目标图像中车辆对应的车灯状态为前行，第三帧目标图像中车辆对应的车灯状态为左转，第四帧目标图像中车辆对应的车灯状态为前行，……。
这里，根据车灯闪烁的周期能够准确地确定目标对象对应的每种预设状态的标准状态序列，比如，车灯闪烁的周期为2，可以确定车灯左转向的状态开启子序列包括10个历史状态信息，即左转、左转、前行、前行、左转、左 转、前行、前行、左转、左转；车灯左转向的状态结束启子序列包括6个除了左转以外的其他车灯状态对应的历史状态信息。
之后，根据时序信息、标准状态序列和车灯闪烁的周期，确定目标对象对应的预测状态信息。
示例性的，已知状态序列信息中包括20个历史状态信息，按照时序信息排列分别为刹车、前行、前行、左转、左转、前行、前行、左转、左转、前行、前行、左转、左转、前行、前行、左转、左转、前行、前行、左转；已知车灯左转向的状态开启子序列包括10个历史状态信息，即左转、左转、前行、前行、左转、左转、前行、前行、左转、左转；车灯左转向的状态结束启子序列包括6个除了左转以外的其他车灯状态对应的历史状态信息；已知车灯闪烁的周期为2。根据时序信息、标准状态序列和车灯闪烁的周期，预测第21帧待检测图像中车辆的车灯状态，例如，在确定第20个历史状态信息为左转状态的情况下，由于从第4个车灯状态到第20个车灯状态不断循环左转、左转、前进、前进，且不断循环对应的序列中包括车灯左转向的状态开启子序列，可以确定左转向开启，根据第20个历史状态信息为左转状态，车灯闪烁的周期为2，能够预测最新获取的待检测图像上的目标对象对应的预测状态信息为左转。
这里，根据图像采样频率能够得到车灯闪烁的周期，根据车灯闪烁的周期能够准确地确定目标对象对应的每种预设状态的标准状态序列，之后，结合准确的标准状态序列和按照时序信息已经排列好的历史状态信息，能够预测出较为精准的预测状态信息。
在一些实施例中，针对S101，确定子对象图像的实现，可以包括：对待检测图像进行识别，得到目标对象所包含的多个预设部位中每个预设部位在待检测图像中的位置信息；之后，计算机设备可以基于位置信息，从待检测图像中分别截取包含每个预设部位的子图像，得到每个预设部位对应的子对象图像。之后，对多个子对象图像进行拼接，这里，可以按照每个预设部位在待检测图像中的位置信息，将每个预设部位对应的子对象图像进行拼接，得到目标图像。
在本公开实施例中，参照图2所示，计算机设备可以利用车灯检测模型对待检测图像进行车灯识别，得到每个车灯组件对应的检测框，即每个车灯组件在待检测图像中的位置信息，之后，从待检测图像中截取每个车灯组件对应的检测框所框出的子图像，得到每个车灯组件对应的子对象图像。之后，计算机设备可以按照每个检测框在待检测图像中的相对位置关系，确定每个子对象图像的拼接位置，之后，按照拼接位置将子对象图像进行拼接得到目标图像。
这里，根据位置信息拼接预设部位对应的子对象图像，得到的目标图像的方式，既聚焦了预设部位，同时也保留了预设部位在目标对象上的相对位置信息，即保留了预设位置对应的结构信息，因此，基于目标图像进行检测能够提高状态检测的效率以及精确度。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
基于同一发明构思，本公开实施例中还提供了与状态检测方法对应的状态检测装置，由于本公开实施例中的装置解决问题的原理与本公开实施例上述状态检测方法相似，因此装置的实施可以参见方法的实施，重复之处不再赘述。
本公开实施例，通过单针图片检测结合时序优化的方式，确定目标状态信息；如此，在硬件计算资源能力受限，或者无法支持时序卷积算法的情况下，计算机设备仍然可以准确地确定目标对象的目标状态信息，提高了状态检测的鲁棒性。
参照图5所示，为本公开实施例提供的一种状态检测装置的示意图，所述装置包括：信息获取部分401、初始状态检测部分402和状态优化部分403；其中，
信息获取部分401，被配置为获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及所述目标对象对应的状态序列信息；所述状态序列信息包括用于表征所述目标对象在多个历史时刻的历史状态信息；
初始状态检测部分402，被配置为将多个所述子对象图像拼接为目标图像，并对所述目标图像进行对象状态检测，得到所述目标对象对应的初始状态信息；
状态优化部分403，被配置为基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述状态序列信息还包括每个历史状态信息对应的目标有效性信息；
所述状态优化部分403，还被配置为在所述状态序列信息中存储的历史状态信息的数量大于预设数量的情况下，基于多个所述历史状态信息以及每个所述历史状态信息对应的目标有效性信息，确定在所述状态序列信息中所述目标对象对应的有效状态信息；
基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述装置包括序列更新部分404；
所述序列更新部分404，被配置为在所述状态序列信息中存储的历史状态信息的数量小于或等于所述预设数量的情况下，将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息，并将所述初始状态信息以及目标有效性信息添加到状态序列信息对应的序列中，得到更新后的状态序列信息。
一种可选的实施方式中，所述序列更新部分404，还被配置为在所述将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息之前，获取所述初始状态信息对应的可信度；
在所述可信度大于第一预设值的情况下，确定所述初始状态信息对应的初始有效性信息为有效。
一种可选的实施方式中，所述序列更新部分404，还被配置为在基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息之后，确定所述目标状态信息对应的目标有效性信息；
按照所述目标状态信息确定的时间顺序，将所述目标状态信息以及所述目标有效性信息添加到状态序列中，得到更新后的状态序列信息。
一种可选的实施方式中，所述序列更新部分404，还被配置为获取所述初始状态信息对应的可信度；
响应于所述可信度大于第一预设值，确定所述目标状态信息对应的目标有效性信息为有效；
响应于所述可信度小于或等于所述第一预设值、所述有效状态信息与所述初始状态信息相同，并且所述可信度大于第二预设值，确定所述目标状态信息对应的目标有效性信息为有效；其中，所述第一预设值大于所述第二预设值。
一种可选的实施方式中，所述序列更新部分404，还被配置为响应于所述可信度小于或等于所述第一预设值，并且所述有效状态信息与所述初始状态信息不相同，确定所述目标状态信息对应的目标有效性信息为无效；
响应于所述可信度小于或等于所述第一预设值，并且所述可信度小于或等于所述第二预设值，确定所述目标状态信息对应的目标有效性信息为无效。
一种可选的实施方式中，所述状态序列信息还包括每个所述历史状态信息对应的时序信息；
所述状态优化部分403，还被配置为基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息；
基于所述初始状态信息、所述有效状态信息和所述预测状态信息，确定所述目标对象对应的目标状态信息。
一种可选的实施方式中，所述状态优化部分403，还被配置为获取预先设置的图像采样频率；
基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，和所述图像采样频率，确定所述目标对象对应的预测状态信息；其中，所述标准状态序列包括对应的预设状态的状态开启子序列和状态结束子序列。
一种可选的实施方式中，所述信息获取部分401，还被配置为对所述待检测图像进行识别，得到所述目标对象所包含的多个预设部位中每个预设部位在所述待检测图像中的；
基于所述位置信息，从所述待检测图像中分别截取包含每个预设部位的子图像，得到每个预设部位对应的所述子对象图像；
所述初始状态检测部分402，还被配置为按照每个预设部位在所述待检测 图像中的位置信息，将每个预设部位对应的子对象图像进行拼接，得到所述目标图像。
在本公开实施例以及其他的实施例中，“部分”可以是部分电路、部分处理器、部分程序或软件等等，当然也可以是单元，还可以是模块也可以是非模块化的。
关于状态检测装置中的各模块的处理流程、以及各模块之间的交互流程的描述可以参照上述状态检测方法实施例中的相关说明，这里不再详述。
基于同一技术构思，本公开实施例还提供了一种计算机设备。参照图6所示，为本公开实施例提供的计算机设备的结构示意图，包括：
处理器51、存储器52和总线53。其中，存储器52存储有处理器51可执行的机器可读指令，处理器51用于执行存储器52中存储的机器可读指令，所述机器可读指令被处理器51执行时，处理器51执行下述步骤：S101：获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及目标对象对应的状态序列信息；状态序列信息包括用于表征目标对象在多个历史时刻的历史状态信息；S102：将多个子对象图像拼接为目标图像，并对目标图像进行对象状态检测，得到目标对象对应的初始状态信息；S103：基于初始状态信息和状态序列信息，确定目标对象对应的目标状态信息。
上述存储器52包括内存521和外部存储器522；这里的内存521也称内存储器，用于暂时存放处理器51中的运算数据，以及与硬盘等外部存储器522交换的数据，处理器51通过内存521与外部存储器522进行数据交换，当计算机设备运行时，处理器51与存储器52之间通过总线53通信，使得处理器51在执行上述方法实施例中所提及的执行指令。
本公开实施例还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行上述方法实施例中所述的状态检测方法的步骤。其中，该存储介质可以是易失性或非易失的计算机可读取存储介质。
本公开实施例还提供一种计算机程序产品，包括计算机指令，所述计算机指令被处理器执行时实现上述的状态检测方法的步骤。其中，计算机程序产品可以是任何能实现上述状态检测方法的产品，该计算机程序产品中对现有技术做出贡献的部分或全部方案可以以软件产品(例如软件开发包(Software Development Kit，SDK))的形式体现，该软件产品可以被存储在一个存储介质中，通过包含的计算机指令使得相关设备或处理器执行上述状态检测方法的部分或全部步骤。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的装置的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。在本公开所提供的几个实施例中，应该理解到，所揭露的装置和方法，可以通过其它的方式实现。以上所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，又例如，多个模块或组件可以结合，或一些特征可以忽略，或 不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些通信接口，装置或模块的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。
所述功能如果以软件功能模块的形式实现并作为独立的产品销售或使用时，可以存储在一个处理器可执行的非易失的计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是：以上所述实施例，仅为本公开的具体实施方式，用以说明本公开的技术方案，而非对其限制，本公开的保护范围并不局限于此，尽管参照前述实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本公开揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本公开实施例技术方案的精神和范围，都应涵盖在本公开的保护范围之内。因此，本公开的保护范围应所述以权利要求的保护范围为准。
工业实用性
本公开实施例中，由于去除了非检测部位的图像，聚焦了待检测的预设部位的图像(即子对象图像)，因此，对目标图像进行对象状态检测，能够得到精准的目标对象对应的初始状态信息。并且，由于多个历史时刻的历史状态信息与当前的初始状态信息有关联，因此，结合多个历史时刻的历史状态信息对初始状态信息进行优化，能够提高目标状态信息的准确性。

Claims (14)

  1. 一种状态检测方法，包括：
    获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及所述目标对象对应的状态序列信息；所述状态序列信息包括用于表征所述目标对象在多个历史时刻的历史状态信息；
    将多个所述子对象图像拼接为目标图像，并对所述目标图像进行对象状态检测，得到所述目标对象对应的初始状态信息；
    基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息。
  2. 根据权利要求1所述的方法，其中，所述状态序列信息还包括每个历史状态信息对应的目标有效性信息；
    所述基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息，包括：
    在所述状态序列信息中存储的历史状态信息的数量大于预设数量的情况下，基于多个所述历史状态信息以及每个所述历史状态信息对应的目标有效性信息，确定在所述状态序列信息中所述目标对象对应的有效状态信息；
    基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息。
  3. 根据权利要求2所述的方法，其中，还包括：
    在所述状态序列信息中存储的历史状态信息的数量小于或等于所述预设数量的情况下，将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息，并将所述初始状态信息以及目标有效性信息添加到状态序列信息对应的序列中，得到更新后的状态序列信息。
  4. 根据权利要求3所述的方法，其中，在所述将所述初始状态信息对应的初始有效性信息作为所述初始状态信息对应的目标有效性信息之前，还包括确定所述初始有效性信息的步骤：
    获取所述初始状态信息对应的可信度；
    在所述可信度大于第一预设值的情况下，确定所述初始状态信息对应的初始有效性信息为有效。
  5. 根据权利要求2所述的方法，其中，在基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息之后，还包括：
    确定所述目标状态信息对应的目标有效性信息；
    按照所述目标状态信息确定的时间顺序，将所述目标状态信息以及所述目标有效性信息添加到状态序列中，得到更新后的状态序列信息。
  6. 根据权利要求5所述的方法，其中，所述确定所述目标状态信息对应的目标有效性信息，包括：
    获取所述初始状态信息对应的可信度；
    响应于所述可信度大于第一预设值，确定所述目标状态信息对应的目标有效性信息为有效；
    响应于所述可信度小于或等于所述第一预设值、所述有效状态信息与所述初始状态信息相同，并且所述可信度大于第二预设值，确定所述目标状态信息对应的目标有效性信息为有效；其中，所述第一预设值大于所述第二预设值。
  7. 根据权利要求6所述的方法，其中，所述确定所述目标状态信息对应的目标有效性信息，还包括：
    响应于所述可信度小于或等于所述第一预设值，并且所述有效状态信息与所述初始状态信息不相同，确定所述目标状态信息对应的目标有效性信息为无效；
    响应于所述可信度小于或等于所述第一预设值，并且所述可信度小于或等于所述第二预设值，确定所述目标状态信息对应的目标有效性信息为无效。
  8. 根据权利要求2所述的方法，其中，所述状态序列信息还包括每个所述历史状态信息对应的时序信息；
    所述基于所述初始状态信息和所述有效状态信息，确定所述目标对象对应的目标状态信息，包括：
    基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息；
    基于所述初始状态信息、所述有效状态信息和所述预测状态信息，确定所述目标对象对应的目标状态信息。
  9. 根据权利要求8所述的方法，其中，所述基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，确定所述目标对象对应的预测状态信息，包括：
    获取预先设置的图像采样频率；
    基于所述状态序列信息中每个所述历史状态信息对应的时序信息、所述目标对象对应的每种预设状态的标准状态序列，和所述图像采样频率，确定所述目标对象对应的预测状态信息；其中，所述标准状态序列包括对应的预设状态的状态开启子序列和状态结束子序列。
  10. 根据权利要求1所述的方法，其中，所述获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，包括：
    对所述待检测图像进行识别，得到所述目标对象所包含的多个预设部位中每个预设部位在所述待检测图像中的位置信息；
    基于所述位置信息，从所述待检测图像中分别截取包含每个预设部位的子图像，得到每个预设部位对应的所述子对象图像；
    所述将多个所述子对象图像拼接为目标图像，包括：
    按照每个预设部位在所述待检测图像中的位置信息，将每个预设部位对应的子对象图像进行拼接，得到所述目标图像。
  11. 一种状态检测装置，包括：
    信息获取部分，被配置为获取待检测图像中目标对象所包含的多个预设部位对应的子对象图像，以及所述目标对象对应的状态序列信息；所述状态序列信息包括用于表征所述目标对象在多个历史时刻的历史状态信息；
    初始状态检测部分，被配置为将多个所述子对象图像拼接为目标图像，并对所述目标图像进行对象状态检测，得到所述目标对象对应的初始状态信息；
    状态优化部分，被配置为基于所述初始状态信息和所述状态序列信息，确定所述目标对象对应的目标状态信息。
  12. 一种计算机设备，包括：处理器、存储器和总线，所述存储器存储有所述处理器可执行的机器可读指令，当计算机设备运行时，所述处理器与所述存储器之间通过总线通信，所述机器可读指令被所述处理器执行时执行如权利要求1至10任一项所述的状态检测方法的步骤。
  13. 一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行如权利要求1至10任一项所述的状态检测方法的步骤。
  14. 一种计算机程序产品，包括计算机可读代码，所述计算机程序产品包括计算机程序或指令，在所述计算机程序或指令在电子设备上运行的情况下，使得所述电子设备执行权利要求1至10中任意一项所述的方法。
PCT/CN2022/096575 2021-11-29 2022-06-01 状态检测方法、装置、计算机设备、存储介质和程序产品 WO2023092982A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111437066.3 2021-11-29
CN202111437066.3A CN114092896A (zh) 2021-11-29 2021-11-29 一种状态检测方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2023092982A1 true WO2023092982A1 (zh) 2023-06-01

Family

ID=80305545

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/096575 WO2023092982A1 (zh) 2021-11-29 2022-06-01 状态检测方法、装置、计算机设备、存储介质和程序产品

Country Status (2)

Country Link
CN (1) CN114092896A (zh)
WO (1) WO2023092982A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117351485A (zh) * 2023-12-04 2024-01-05 深圳赛陆医疗科技有限公司 基因测序训练数据集的增强方法及装置、设备及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114092896A (zh) * 2021-11-29 2022-02-25 上海商汤临港智能科技有限公司 一种状态检测方法、装置、计算机设备和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109255356A (zh) * 2018-07-24 2019-01-22 阿里巴巴集团控股有限公司 一种文字识别方法、装置及计算机可读存储介质
CN110969647A (zh) * 2019-12-04 2020-04-07 苏州智加科技有限公司 一种整合车辆的识别跟踪与车灯检测的方法
CN112528923A (zh) * 2020-12-21 2021-03-19 上海眼控科技股份有限公司 视频分析方法、装置、电子设备和存储介质
CN112712057A (zh) * 2021-01-13 2021-04-27 腾讯科技（深圳）有限公司 交通信号识别方法、装置、电子设备及存储介质
US20210295058A1 (en) * 2020-03-23 2021-09-23 Toyota Jidosha Kabushiki Kaisha Apparatus, method, and computer program for identifying state of object, and controller
CN114092896A (zh) * 2021-11-29 2022-02-25 上海商汤临港智能科技有限公司 一种状态检测方法、装置、计算机设备和存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109255356A (zh) * 2018-07-24 2019-01-22 阿里巴巴集团控股有限公司 一种文字识别方法、装置及计算机可读存储介质
CN110969647A (zh) * 2019-12-04 2020-04-07 苏州智加科技有限公司 一种整合车辆的识别跟踪与车灯检测的方法
US20210295058A1 (en) * 2020-03-23 2021-09-23 Toyota Jidosha Kabushiki Kaisha Apparatus, method, and computer program for identifying state of object, and controller
CN112528923A (zh) * 2020-12-21 2021-03-19 上海眼控科技股份有限公司 视频分析方法、装置、电子设备和存储介质
CN112712057A (zh) * 2021-01-13 2021-04-27 腾讯科技（深圳）有限公司 交通信号识别方法、装置、电子设备及存储介质
CN114092896A (zh) * 2021-11-29 2022-02-25 上海商汤临港智能科技有限公司 一种状态检测方法、装置、计算机设备和存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117351485A (zh) * 2023-12-04 2024-01-05 深圳赛陆医疗科技有限公司 基因测序训练数据集的增强方法及装置、设备及存储介质
CN117351485B (zh) * 2023-12-04 2024-03-08 深圳赛陆医疗科技有限公司 基因测序训练数据集的增强方法及装置、设备及存储介质

Also Published As

Publication number Publication date
CN114092896A (zh) 2022-02-25

Similar Documents

Publication Publication Date Title
WO2023092982A1 (zh) 状态检测方法、装置、计算机设备、存储介质和程序产品
US10691952B2 (en) Adapting to appearance variations when tracking a target object in video sequence
US10740654B2 (en) Failure detection for a neural network object tracker
CN108725440B (zh) 前向碰撞控制方法和装置、电子设备、程序和介质
US20150371398A1 (en) Method and system for updating background model based on depth
CN111292531B (zh) 交通信号灯的跟踪方法、装置、设备及存储介质
Ciberlin et al. Object detection and object tracking in front of the vehicle using front view camera
JP2017033559A (ja) 放置物の検出装置、方法及びシステム
CN111814746A (zh) 一种识别车道线的方法、装置、设备及存储介质
CN113112525B (zh) 目标跟踪方法、网络模型及其训练方法、设备和介质
CN112733818B (zh) 基于注意力机制的车灯状态识别方法、装置、终端和介质
CN117425916A (zh) 遮挡感知的多对象跟踪
CN110647818A (zh) 一种遮挡目标物体的识别方法及装置
WO2021017341A1 (zh) 识别智能行驶设备的行驶状态的方法及装置、设备
Zhang et al. Pedestrian detection based on hierarchical co-occurrence model for occlusion handling
RU2764223C2 (ru) Способ и система для определения состояния светофора
US20220067479A1 (en) Vehicle entry detection
CN113674317B (zh) 一种高位视频的车辆跟踪方法及装置
CN110799984A (zh) 跟踪控制方法、设备、计算机可读存储介质
JP2024516642A (ja) 行動検出方法、電子機器およびコンピュータ読み取り可能な記憶媒体
CN113344975A (zh) 一种多目标跟踪方法、装置及电子设备
CN112699781A (zh) 车灯状态检测方法、装置、计算机设备和可读存储介质
US20240257535A1 (en) Systems and methods for determining road object importance based on forward facing and driver facing video data
Moon et al. Robust visual detection of brake-lights in front for commercialized dashboard camera
CN111597959B (zh) 行为检测方法、装置及电子设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22897081

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE