WO2018149322A1 - 图像识别方法、装置、设备及存储介质 - Google Patents

图像识别方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2018149322A1
WO2018149322A1 PCT/CN2018/075379 CN2018075379W WO2018149322A1 WO 2018149322 A1 WO2018149322 A1 WO 2018149322A1 CN 2018075379 W CN2018075379 W CN 2018075379W WO 2018149322 A1 WO2018149322 A1 WO 2018149322A1
Authority
WO
WIPO (PCT)
Prior art keywords
identification
monitoring
identification record
area
coordinates
Prior art date
Application number
PCT/CN2018/075379
Other languages
English (en)
French (fr)
Inventor
王达峰
Original Assignee
腾讯科技（深圳）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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Priority to EP18754005.9A priority Critical patent/EP3585052B1/en
Publication of WO2018149322A1 publication Critical patent/WO2018149322A1/zh
Priority to US16/526,561 priority patent/US10936876B2/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/44—Event detection

Definitions

  • the present application relates to the field of video surveillance technologies, and in particular, to an image recognition method, a device device, and a storage medium.
  • monitoring systems can automatically recognize the monitoring image captured by the monitoring device. Specifically, the monitoring system performs frame-by-frame identification on all areas of the monitoring image captured by the monitoring device according to a predetermined identification algorithm, so as to timely identify a specified event in the monitoring image, for example, identifying a person or object that meets the specified requirement. The event, or, identifies the event in which the specified behavior occurred.
  • the monitoring system needs to identify all areas in the monitoring image captured by the monitoring device, the calculation amount of the monitoring system is large, the hardware requirement of the monitoring system is high, and the larger calculation amount also needs to be consumed. High power.
  • the present application provides an image recognition method, device, device and storage medium, and the technical solution is as follows:
  • an image recognition method comprising:
  • first identification record file includes at least two identification records, and each piece of identification record includes a coordinate corresponding to the specified event in the monitoring image when the monitoring image captured by the monitoring device is recognized ;
  • an image recognition apparatus comprising:
  • a first file obtaining module configured to acquire a first identification record file, where the first identification record file includes at least two identification records, each identification record includes a designated designation when the monitoring image captured by the monitoring device is recognized The event corresponds to the coordinates in the monitored image;
  • a region determining module configured to determine a monitoring area according to coordinates included in at least two of the first identification record files, where the monitoring area is a partial area in the monitoring image captured by the monitoring device;
  • the identification module is configured to identify, in the at least one frame monitoring image, an image corresponding to the monitoring area, when the at least one frame of the monitoring image that is subsequently captured by the monitoring device is identified.
  • an identification device comprising a processor and a memory, the memory storing instructions that are executed by the processor to implement the image recognition method described above.
  • a computer readable storage medium having stored therein instructions that are executed by a processor of an identification device to implement the image recognition method described above.
  • the identification device identifies the coordinates of the specified event according to the identification record of the monitoring image captured by the monitoring device, and determines a monitoring area that needs to be identified later, and when the monitoring image subsequently captured by the monitoring device is identified, It only needs to identify the identified monitoring area, which can reduce the identified image area and reduce the calculation amount of image recognition, thereby reducing the hardware requirements and power consumption of the monitoring system.
  • FIG. 1 is a schematic structural diagram of a monitoring system according to an embodiment of the present application.
  • FIG. 2 is a flowchart of an image recognition method according to an exemplary embodiment
  • Figure 3 is a schematic diagram of coordinates of the embodiment shown in Figure 2;
  • FIG. 4 is a schematic diagram of a monitoring area involved in the embodiment shown in FIG. 2;
  • FIG. 5 is a flowchart showing an implementation of the solution involved in the embodiment shown in Figure 2;
  • FIG. 6 is a flowchart of an image recognition method according to an exemplary embodiment
  • Figure 7 is a schematic view of a region involved in the embodiment shown in Figure 6;
  • FIG. 8 is a block diagram showing the structure of an image recognition apparatus according to an exemplary embodiment
  • FIG. 9 is a schematic structural diagram of a device according to an exemplary embodiment.
  • FIG. 1 is a schematic structural diagram of a monitoring system according to an embodiment of the present application.
  • the system includes a number of monitoring devices 120 and identification devices 140.
  • the monitoring device 120 may be a dedicated surveillance camera or other electronic device (such as a smart phone, a tablet, etc.) including a camera.
  • Identification device 140 can be a general purpose computer or server.
  • the monitoring device 120 and the identification device 140 are connected by a wired or wireless network.
  • the monitoring device 120 can send the captured monitoring image to the identification device 140 in real time through the wired or wireless network, and the identification device 140 identifies the monitoring image captured by the monitoring device 120 according to a predetermined recognition algorithm.
  • the system may also include a graphical display device 160 coupled to the identification device 140, such as the display device 160.
  • the identification device 140 recognizes a certain monitoring image
  • the monitoring image and the recognition result of the monitoring image may be collectively displayed in the graphic display device 160.
  • the monitoring device 120 and the identification device 140 may also be different components of the same electronic device.
  • the monitoring device 120 may be a camera in the electronic device, and the identification device 140 may be in the electronic device.
  • the processing function component including the processor and the memory, etc.
  • the monitoring device 120 and the identification device 140 are connected by a communication bus in the electronic device.
  • the identification device 140 may first acquire the first identification record file of the monitoring device during the process of identifying the monitoring image captured by the monitoring device 120, where the first identification record file includes At least two identification records, each of the identification records including the identification image captured by the monitoring device 120, the identified specified event corresponding to the coordinates in the monitoring image; the identification device 140 according to the first identification record file
  • the coordinate included in the at least two identification records determines a monitoring area, where the monitoring area is a partial area in the monitoring image captured by the monitoring device, and when the at least one frame of the monitoring image subsequently captured by the monitoring device is identified, the identifying device 140 In at least one frame of the monitoring image, the image corresponding to the monitoring area is identified.
  • the identification device 140 can identify the coordinates of the specified event according to the identification record of the monitoring image captured by the monitoring device 120, and determine the monitoring area that needs to be identified later.
  • the identification device 140 photographs the monitoring device 120 before the time node at a certain time node.
  • the identification record of the monitoring image is analyzed, and a part of the region in which the specified event identified by the identification record is relatively concentrated in the monitoring image is determined as a monitoring area, and after the time node, the new monitoring image captured by the monitoring device 120 on the monitoring device 120 is detected.
  • the specified event identification is performed, only the monitoring area in the new monitoring image is identified, and the image of the area other than the monitoring area in the new monitoring image is not recognized, compared to the whole of the monitoring image.
  • the scheme for identifying the image in the case where the recognition algorithm is the same, the application is implemented Illustrated embodiment can reduce the area of the image recognition, the image recognition calculation amount reduced, thereby reducing hardware requirements of the monitoring system, and power consumption.
  • FIG. 2 is a flowchart of an image recognition method according to an exemplary embodiment, which may be applied to the identification device 140 of the monitoring system shown in FIG. 1.
  • the image recognition method can include the following steps:
  • Step 201 Acquire a first identification record file, where the first identification record file includes at least two identification records.
  • the identified specified event corresponds to monitoring. The coordinates in the image.
  • the first identification record file is a file containing at least two identification records, each of the identification records includes one or more coordinates, and the coordinates may be identified by the identification device in the monitoring image corresponding to the identification record. Specifies the coordinates of the event in the corresponding surveillance image.
  • the specified event may be a specified person or object appearing in the monitoring image; for example, if the identification device needs to identify a person wearing black clothes or a black color in the monitoring device, the designated event may be a person wearing black clothes in the monitoring image. Or a vehicle of black color; or, assuming that the identification device needs to identify a traffic accident, the designated event may be a traffic accident event.
  • the specified event recognized by the identification device may correspond to only one coordinate.
  • the coordinate corresponding to the specified event may be the central coordinate of the region where the specified event is located in the monitoring image; specifically, when the event is specified For a traffic accident event, the coordinate corresponding to the specified event may be the central coordinate of the traffic accident occurrence area.
  • the coordinate corresponding to the specified event may also be the coordinate of a specific position of the object involved in the specified event. For example, when the specified event is a character, the coordinate corresponding to the specified event may be the coordinate of the center of the person's head. .
  • a specified event may also correspond to multiple coordinates.
  • the coordinate corresponding to the specified event may be the coordinate of the specific location of the object involved in the specified event, for example, when the specified event is present.
  • the coordinates corresponding to the specified event may be the coordinates of the center of the person's head and the coordinates of the center of the person's torso.
  • the coordinate may be a pixel coordinate corresponding to the resolution of the monitoring image, or may be a size coordinate corresponding to the size of the monitoring image.
  • FIG. 3 illustrates a schematic diagram of coordinates involved in the embodiment of the present application.
  • the resolution of the monitoring image captured by the monitoring device is 1080*720
  • the lower left corner of the monitoring image is taken as the origin
  • the lower boundary of the monitoring image is the horizontal axis
  • the left boundary is the vertical axis
  • the coordinate system is assumed.
  • the pixel of the specified event in the monitoring image of the certain frame captured by the monitoring device is the 850th pixel from the left in the monitoring image, and the 315th pixel is the next
  • the coordinate of the specified event may be set to (850) , 315).
  • the coordinate of the specified event may be set to (850, 315).
  • Step 202 Determine a monitoring area according to coordinates included in at least two of the first identification record files.
  • the monitoring area may be a partial area in the monitoring image captured by the monitoring device.
  • the identification device may determine, according to the historical monitoring record of the monitoring image captured by the monitoring device, an area in which the coordinates corresponding to the identified specified event are concentrated in the monitoring image as the monitoring area.
  • the identification device may determine target coordinates in coordinates included in at least two of the first identification record files, where the target coordinates are coordinates whose reliability is higher than a preset reliability threshold, the reliability A probability for indicating that the specified event is recognized again at the corresponding coordinates, and determining a minimum area including the target coordinates as the monitored area.
  • the target coordinate whose reliability is higher than the preset credibility threshold may be determined by a method of probability statistics. For example, when the identification device determines the target coordinate, at least two of the first identification record files may be calculated.
  • the strip identifies the standard deviation of the coordinates included in the record, and generates a two-dimensional normal distribution formula corresponding to the coordinates included in the at least two identification records in the first identification record file according to the standard deviation (the two-dimensional normal distribution formula corresponds to the space In the coordinate system is an axisymmetric bell-shaped surface), and according to the two-dimensional normal distribution formula and the credibility threshold, at least two of the first identification record files are identified in the coordinates included in the record, and are trusted The coordinates whose degree is higher than the reliability threshold are determined as the target coordinates.
  • the identification device determines the monitoring area as the minimum area including the coordinates whose reliability is higher than the reliability threshold
  • the area enclosed by the outermost coordinates of the determined target coordinates may be determined as Monitoring area.
  • FIG. 4 illustrates a schematic diagram of a monitoring area involved in an embodiment of the present application.
  • the area 40 is a complete coordinate area of the monitoring image
  • the coordinates of each "x" type mark in FIG. 4 are one coordinate included in the identification record in the first identification record file, and the identification device passes the probability statistical method. Determining the coordinates of the central region of the region 40 in FIG. 4 as the coordinates whose reliability is higher than the preset credibility threshold, and determining the coordinates including the credibility threshold higher than the preset credibility threshold, The area enclosed by each of the peripheral coordinates (ie, the area 41 in FIG. 4), and the coordinate area corresponding to the area 41 is the determined monitoring area.
  • Step 203 Identify at least one frame of the monitoring image that is subsequently captured by the monitoring device, and identify an image corresponding to the monitoring area in the at least one frame of the monitoring image.
  • the monitoring device After the monitoring device determines the monitoring area, after acquiring at least one frame of the monitoring image captured by the monitoring device, the monitoring device only needs to identify a part of the image corresponding to the monitoring area in the at least one frame of the monitoring image according to a predetermined identification algorithm to determine Whether a specified event occurs in a partial image of the area to be monitored.
  • the area of one frame of the monitoring image captured by the monitoring device is 100*100, and the identification device is at a certain time node, and the monitoring device is before the time node.
  • the captured identification record of the captured image is analyzed, and a part of the relatively concentrated part of the monitored image identified by the identification record is determined as a monitoring area (assuming that the monitoring area is an area of 50*50 in the surveillance image), After the time node, when the identification device performs the specified event recognition on the new monitoring image captured by the monitoring device, only the 50*50 monitoring area in the new monitoring image is identified, and the monitoring is performed in the new monitoring image.
  • the image of the other area outside the area is not recognized, and the solution shown in the embodiment of the present application only needs to be the entire image compared to the related art in which all areas in the entire image of the monitoring image are identified. Part of the area is identified, so that it can solve the problem of large amount of calculation when the monitoring image is recognized in the related art.
  • the requirements of the parts and the high power consumption result in the reduction of the amount of calculation of image recognition, thereby reducing the hardware requirements of the monitoring system and the effect of power consumption.
  • the identification device may further supplement the first identification record file when the subsequent monitoring image is partially identified according to the determined monitoring area.
  • the embodiment may further include the following steps 204 and 205. .
  • Step 204 For each frame monitoring image in the at least one frame monitoring image, calculate an event probability that the event identified in the monitoring image is the specified event.
  • the event probability of the identified event being the specified event may be calculated.
  • the monitoring device can display the character of each character appearing in the currently recognized monitoring image with the person in the specified event.
  • the clothing features are compared to calculate the similarity between the clothing characteristics of each character and the clothing characteristics of the characters in the specified event, and the similarity between the clothing characteristics of each character and the clothing characteristics of the characters in the specified event can be As the above event probability, or alternatively, the event probability may be obtained according to the similarity between the clothing features.
  • the identification device may multiply the above similarity by a predetermined coefficient (the predetermined coefficient may be a positive number that is close to and less than 1, such as 0.9), and use the obtained value as the event probability; or, the identification device may also store The correspondence between the interval in which the similarity is located and the event probability, after the recognition device calculates the similarity, the interval in which the similarity is located can be determined, and the event probability corresponding to the interval in which the similarity is located is further queried.
  • a predetermined coefficient may be a positive number that is close to and less than 1, such as 0.9
  • Step 205 When the event probability is higher than the preset probability threshold, generate an identification record of the event according to the coordinate corresponding to the event in the monitoring image, and add the identification record of the event to the first identification record file.
  • the probability threshold may be a percentage value set according to actual application requirements.
  • the probability threshold for example, 95%)
  • the identified event may be considered as the specified event.
  • the identifying device may obtain the above identified event corresponding to the current event. Identifying coordinates in the monitored image, generating an identification record including the coordinates, and adding the generated identification record to the first identification record file to timely replenish the first identification record file, and subsequently determining a new monitoring area, The monitoring area can be corrected according to the newly added identification record to improve the accuracy of the monitoring area determination.
  • the identification device may capture the identification device.
  • the entire area of the monitoring image is identified, or may be in accordance with a preset monitoring area (the preset monitoring area may be a default area in the monitoring image, or may be an area manually selected by the user or the monitoring personnel)
  • the monitoring image captured by the recognition device is identified, and an identification record is generated and added to the first identification record file when the specified event is recognized.
  • the identification device may acquire the first identification record file once every predetermined time (such as one day or one week), and determine the monitoring area according to the acquired first identification record file, each time after determining the monitoring area, based on The determined monitoring area performs event identification on the monitoring image subsequently captured by the monitoring device, generates an identification record when the specified event is recognized, and adds the identification record file to the first identification record file until the next acquisition of the first identification record file and determines a new monitoring area.
  • the identification device may also acquire the first identification record file and determine the monitoring area each time the monitoring image captured by the monitoring device is recognized, and perform event recognition based on the determined monitoring area to be recognized by the monitoring image, and identify An identification record is generated when the event is specified and added to the first identification record file.
  • FIG. 5 is a flowchart of an implementation of the solution in the embodiment of the present application.
  • the monitoring device is a surveillance camera
  • the identification device is a monitoring server in the monitoring room.
  • the captured monitoring image is sent to the monitoring server, and the monitoring image sent by the monitoring camera to the monitoring camera is initially detected.
  • the entire area is identified, and when the specified event is identified, the recognition result and the monitoring image are aggregated and displayed on the monitoring screen (ie, the graphic display device 160 shown in FIG. 1), and the corresponding event corresponding to the specified image is generated in the monitoring image.
  • the identification record of the coordinates is added and the identification record is added to the first identification record file.
  • the monitoring server obtains the first identification record file once every other day, and determines the monitoring area according to the identification record in the obtained first identification record file, and the surveillance camera is determined according to the determined monitoring area in the subsequent day.
  • the captured monitoring image is identified (ie, only part of the image of the corresponding monitoring area in the monitoring image is recognized), and when the specified event is recognized, an identification record containing the coordinates corresponding to the specified event in the monitoring image is generated and identified The record is added to the first identification record file.
  • the identification device identifies the coordinates of the specified event according to the identification record of the monitoring image captured by the monitoring device, and determines the monitoring area that needs to be identified later.
  • the monitoring image captured by the monitoring device is recognized, only the determined monitoring area needs to be identified, which can reduce the identified image area and reduce the calculation amount of image recognition, thereby reducing hardware requirements and power consumption of the monitoring system. .
  • the identification device when the identification device identifies the monitoring image that is subsequently captured by the monitoring device according to the determined monitoring area, if the probability that the identified event is a specified event is higher than a preset threshold, according to the The coordinate corresponding to the identified event is generated and added to the first identification record file to timely replenish the first identification record file.
  • the monitoring area may be performed according to the newly added identification record. Corrected to improve the accuracy of the monitoring area.
  • FIG. 6 is a flowchart of an image recognition method according to an exemplary embodiment, which may be applied to the identification device 140 of the monitoring system shown in FIG. 1.
  • the image recognition method can include the following steps:
  • Step 601 Acquire a first identification record file, where the first identification record file includes at least two identification records.
  • the identified specified event corresponds to monitoring. The coordinates in the image.
  • Step 602 Determine whether an expired record is included in the first identification record file.
  • the identification record in the first identification record file may include a corresponding recognition time in addition to the corresponding coordinates, and the expired record is an interval between the corresponding recognition time and the current time being greater than the preset time interval. record of.
  • the identified areas of the specified event set may change over time.
  • the shooting direction of the monitoring device may change over time, resulting in the coordinates of the specified event being concentrated.
  • the area has changed; at the same time, as time passes, the number of identification records in the first identification record file is also increasing.
  • the number of identification records is too large, it may cause a complicated increase in the calculation of the subsequent determination of the monitoring area. Therefore, in the embodiment of the present application, in the embodiment of the present application, in order to prevent the identified area of the coordinate corresponding to the specified event from being changed, the subsequent monitoring area is determined to be inaccurate, and the calculation complexity of the monitoring area is too high.
  • the interval between the recognition time and the current time corresponding to each identification record in the first identification record file may be calculated, and the interval between the recognition time distance and the current time is greater than the preset time interval.
  • the identification record (for example, 3 months or half a year) is determined to be an expired record.
  • Step 603 deleting the expired record from the first identification record file.
  • the identification device may delete the expired record in the first identification record file, so as to prevent the identified area of the specified event set from being changed, and the subsequent monitoring area is determined to be inaccurate. And to determine the computational complexity of the monitoring area is too high.
  • the identification device may further identify all areas in the partial image in the monitoring image captured by the monitoring device, and supplement the first identification record file according to the recognition result.
  • the embodiment of the present application may further include the following steps 604 to 607.
  • Step 604 Acquire a second identification record file, where each identification record in the second identification record file includes all the areas of the partial surveillance image captured by the monitoring device, and the identified designated event corresponds to the monitoring image. The coordinates in .
  • the partial monitoring image is an image other than the monitoring image corresponding to the identification record included in the first identification recording file.
  • the identification device can extract part of the monitoring from all the monitoring images captured by the monitoring device, in order to reduce the situation in which the identification device only recognizes the monitoring area in the monitoring image.
  • the image identifies the entire area of the partial monitoring image, and when the specified event is recognized, generates an identification record according to the coordinates of the identified specified event corresponding to the monitored image, and adds the identification record to the second identification record file.
  • a certain proportion (for example, 10%) of the monitoring image may be extracted from all the monitoring images captured by the monitoring device, and extracted. The entire area of the monitored image is identified, and the remaining portion of the monitored image is identified in accordance with the determined monitoring area.
  • the identification device when extracting a certain proportion of the monitoring image, may extract a certain proportion of the monitoring image from all the monitoring images according to a random algorithm; or the identification device may also sample from all the monitoring images according to a predetermined sampling rate, Extract a certain percentage of surveillance images. Specifically, if the above extraction ratio is 10%, the monitoring device captures and transmits 100 frames of monitoring images, and the identification device can randomly extract 10 frames of monitoring images from the 100-frame monitoring image for identification of all areas, and other 90 frames of monitoring.
  • the image is identified according to the determined monitoring area; or, the monitoring device can also sort the 100 frames of monitoring images according to the shooting time sequence of the 100-frame monitoring image, and will monitor the last frame in every 10 frames (or second)
  • the monitoring image is extracted to identify all areas, and the first 9 frames of monitoring images in every 10 frames of monitoring images are identified according to the determined monitoring area.
  • Step 605 Calculate a degree of fitting between the first identification record file and the second identification record file.
  • the degree of similarity is used to indicate a degree of similarity between the first area and the second area; wherein the first area is an area corresponding to the coordinates in the identification record included in the first identification record file, and the second area is the second area The area corresponding to the coordinates in the identification record included in the record is identified.
  • the identification device when calculating the degree of fit between the first identification record file and the second identification record file, may first determine the coordinate corresponding to the identification record included in the first identification record file. a first area, and a second area corresponding to the coordinates in the identification record included in the second identification record file, wherein the first area and the second area are determined in a manner similar to the step of determining the monitoring area in step 202 above, and the identifying device Calculating a ratio of an area of the common area between the second area and the first area to a total area of the second area, and calculating the above fitting degree according to a ratio of an area of the common area to a total area of the second area.
  • the identification device may directly compare the area of the common area to the total area of the second area as the coordinate in the identification record included in the first identification record file and the coordinate in the identification record included in the second identification record file. The degree of fit.
  • FIG. 7 shows a schematic diagram of a region involved in the embodiment of the present application.
  • the area 70 is a complete coordinate area of the monitoring image
  • the identification device calculates that the first area corresponding to the coordinates in the identification record included in the first identification record file is the area 71, and calculates the second identification record file.
  • the second area corresponding to the coordinates in the identification record is the area 72, wherein the common area of the area 71 and the area 72 is the area 73 (ie, the hatched portion in FIG. 7), and the identification device can calculate the area of the area 73 to occupy the area 72.
  • the ratio of the area is calculated as the degree of fit between the coordinates in the identification record included in the first identification record file and the coordinates in the identification record included in the second identification record file.
  • each identification record in the second identification record file may also include a corresponding identification time, in calculating a coordinate in the identification record included in the first identification record file and a coordinate in the identification record included in the second identification record file.
  • the identification device may also determine an expired record in the second identification record file and delete the expired record in the second identification record file.
  • Step 606 Detect whether the fitness is lower than a preset fitness threshold; if yes, go to step 607; otherwise, go to step 608.
  • the above fitness threshold may be a preset value or a proportional value set according to actual application requirements. For example, taking the ratio of the area of the common area to the total area of the second area as the above-mentioned fitting degree, when the area of the common area accounts for the total area of the second area, the ratio is not lower than a preset fitting threshold.
  • step 608 can be entered; otherwise, if the area of the common area is When the ratio of the total area of the second area is lower than the preset fitness threshold, it can be considered that the probability of occurrence of the specified event in other areas outside the monitoring area is higher in the monitoring image captured by the monitoring device.
  • Step 607 supplement the first identification record file according to the identification record included in the second identification record file.
  • the identification device can supplement the first identification record file and correct the monitoring area in time. Specifically, when the first identification record file is supplemented, the identification device may add the identification record included in the second identification record file to the first identification record file; or the identification device may only record the second identification record. In the identification record included in the file, the identification record whose corresponding coordinates are outside the above common area is added to the first identification record file.
  • the interval between the recognition time corresponding to the identification record included in the second identification record file and the current time is within a preset time interval (such as within one week or within one month),
  • a preset time interval such as within one week or within one month
  • the monitoring area is corrected.
  • the above threshold may be preset by a developer or a maintenance personnel.
  • Step 608 Determine a monitoring area according to coordinates included in at least two of the first identification record files.
  • Step 609 Identify at least one frame of the monitoring image that is subsequently captured by the monitoring device, and identify an image corresponding to the monitoring area in the at least one frame of the monitoring image.
  • the identification device may further supplement the first identification record file when the subsequent monitoring image is partially identified according to the determined monitoring area.
  • the embodiment of the present application may further include the following steps 610 and 611. .
  • Step 610 For each frame monitoring image in the at least one frame monitoring image, calculate an event probability that the event identified in the monitoring image is the specified event.
  • Step 611 When the event probability is higher than the preset probability threshold, generate an identification record of the event according to the coordinate corresponding to the event in the monitoring image, and add the identification record of the event to the first identification record file.
  • the embodiment of the present application may further include the following steps 612 and 613.
  • Step 612 Display the monitoring image when the event probability is higher than a preset probability threshold.
  • the identification device when the identification device detects that the event probability of an event is higher than the preset probability threshold, it may be considered that the suspected specified event is detected. At this time, the monitoring image corresponding to the identified event may be displayed. , by the user or monitoring personnel to determine whether the specified event does occur.
  • Step 613 Receive an event denial response sent for the monitoring image, where the event denial response is used to indicate that the event is not the specified event, and the identification record of the event is deleted from the first identification record file.
  • the user or the monitoring personnel may issue an event denial response based on the displayed monitoring image.
  • the identifying device records the identification record of the event from the first identification record file. Delete so that the incorrect recognition result affects the accuracy of the subsequent monitoring area determination.
  • the monitoring device is a surveillance camera
  • the identification device is a monitoring server in the monitoring room. After the surveillance camera is installed, the captured monitoring image is sent to the monitoring server.
  • the monitoring server On the first day after the installation of the surveillance camera, the monitoring server initially identifies all areas of the surveillance image sent by the surveillance camera, and when the specified event is identified, the recognition result and the surveillance image are aggregated and displayed on the monitoring screen, and the The specified event corresponds to the coordinates in the monitoring image and the identification record of the recognition time, and the identification record is added to the first identification record file.
  • the monitoring server acquires the first identification record file, determines the monitoring area according to the identification record in the acquired first identification record file, and monitors the image for every 10 frames of the surveillance image captured by the surveillance camera.
  • the identification device identifies the first nine frames of the monitoring image according to the determined monitoring area, and performs all area identification on the last frame of the monitoring image; when the first nine frames of the monitoring image are identified and the specified event is identified, the generating Include the coordinate of the specified event and the identification record of the recognition time, and add the identification record to the first identification record file; when the last frame of the monitoring image is identified and the specified event is recognized, the coordinate containing the specified event is generated And identifying the identification record of the time and adding the identification record to the second identification record file.
  • the monitoring server acquires the first identification record file and the second identification record file once, first deletes the expired record in the first identification record file and the second identification record file according to the recognition time, and then calculates the first identification record.
  • a degree of fit between the coordinates of the identification record in the file and the coordinates of the identification record in the second identification record file when the degree of fit between the two is not lower than a preset fitness threshold, the identification device is based on The identification record in the obtained first identification record file determines the monitoring area, and when the degree of fit between the two is lower than the preset fitness threshold, the identification device adds the identification record in the second identification record file.
  • the monitoring image for the surveillance camera is monitored every 10 frames, and the identification device is monitored according to the determination
  • the area identifies the first 9 frames of the monitoring image, and performs the identification of all the areas in the last frame of the monitoring image;
  • an identification record containing the coordinates of the specified event is generated, and the identification record is added to the first identification record file;
  • the last frame of the monitoring image is identified and identified
  • an identification record containing the coordinates of the specified event is generated, and the identification record is added to the second identification record file.
  • the identification device identifies the coordinates of the specified event according to the identification record of the monitoring image captured by the monitoring device, and determines the monitoring area that needs to be identified later.
  • the monitoring image captured by the monitoring device is recognized, only the determined monitoring area needs to be identified, which can reduce the identified image area and reduce the calculation amount of image recognition, thereby reducing hardware requirements and power consumption of the monitoring system. .
  • the identification device when the identification device identifies the monitoring image that is subsequently captured by the monitoring device according to the determined monitoring area, if the probability that the identified event is a specified event is higher than a preset threshold, according to the The coordinate corresponding to the identified event is generated and added to the first identification record file to timely replenish the first identification record file.
  • the monitoring area may be performed according to the newly added identification record. Corrected to improve the accuracy of the monitoring area.
  • the identification device deletes the expired record in the first identification record file before determining the monitoring area according to the first identification record file, so as to avoid that the identified area in the specified event set may change. As a result, the subsequent monitoring area is determined to be inaccurate, and the calculation complexity of the monitoring area is too high.
  • the identification device before the determining the monitoring area according to the first identification record file, the identification device further calculates a fitting degree between the first identification record file and the second identification record file, when the degree of fit is low.
  • the first identification record file is supplemented according to the identification record in the second identification record file to correct the monitoring area determined according to the first identification record file, thereby improving the accuracy of the detection area determination.
  • FIG. 8 is a block diagram showing the structure of an image recognition apparatus according to an exemplary embodiment.
  • the image recognition device can be used in the monitoring device 140 of the monitoring system shown in Figure 1 to perform all or part of the steps in the embodiment shown in Figure 2 or Figure 6.
  • the image recognition device may include:
  • the first file obtaining module 801 is configured to acquire a first identification record file, where the first identification record file includes at least two identification records, and each identification record includes a recognition image captured by the monitoring device, and is recognized The specified event corresponds to the coordinates in the monitored image;
  • the area determining module 802 is configured to determine a monitoring area according to coordinates included in at least two of the first identification record files, where the monitoring area is a partial area in the monitoring image captured by the monitoring device;
  • the identification module 803 is configured to identify, in the at least one frame of the monitoring image, an image corresponding to the monitoring area, when the at least one frame of the monitoring image that is subsequently captured by the monitoring device is identified.
  • the area determining module includes:
  • a coordinate determining unit configured to determine target coordinates in coordinates included in at least two of the first identification record files, where the target coordinates are coordinates whose reliability is higher than a preset reliability threshold, The credibility is used to indicate the probability of recognizing the specified event again at the corresponding coordinates;
  • a region determining unit configured to determine a minimum region including the target coordinates as the monitoring region.
  • the coordinate determining unit includes:
  • a calculating subunit configured to calculate a standard deviation of coordinates included in at least two of the first identification record files
  • Generating a subunit configured to generate, according to the standard deviation, a two-dimensional normal distribution formula corresponding to coordinates included in at least two identification records in the first identification record file;
  • a coordinate determining subunit configured to, according to the two-dimensional normal distribution formula and the credibility threshold, identify at least two of the first identification record files, and the reliability is higher than
  • the coordinates of the confidence threshold are determined as the target coordinates.
  • the device further includes:
  • a probability calculation module configured to monitor, for each frame of the at least one frame monitoring image, an event probability that the event identified in the monitoring image is the specified event
  • a record generating module configured to: when the event probability is higher than a preset probability threshold, generate an identification record of the event according to coordinates of the event corresponding to the monitoring image;
  • a adding module configured to add an identification record of the event to the first identification record file.
  • the device further includes:
  • a display module configured to display the monitoring image when the event probability is higher than a preset probability threshold
  • a response receiving module configured to receive an event denial response sent by the monitoring image, where the event denial response is used to indicate that the event is not the specified event;
  • a first deleting module configured to delete the identification record of the event from the first identification record file.
  • each of the identification records further includes a corresponding identification time
  • the device further includes:
  • a record determining module configured to determine whether an expired record is included in the first identification record file, and the expired record is corresponding to determining a monitoring area according to coordinates included in at least two of the first identification record files The interval between the recognition time and the current time is greater than the record of the preset time interval;
  • a second deleting module configured to delete the expired record from the first identification record file.
  • the device further includes:
  • a second file obtaining module configured to acquire a second identification record file, the second identification record, before the area determining module determines the monitoring area according to the coordinates included in the at least two identification records in the first identification record file
  • the identified specified event corresponds to coordinates in the monitoring image
  • the partial monitoring image is the first An image other than the monitoring image corresponding to the identification record included in the identification file
  • a fitness calculation module configured to calculate a fitting degree between a coordinate in the identification record included in the first identification record file and a coordinate in the identification record included in the second identification record file, the degree of fit And indicating a degree of similarity between the first area and the second area;
  • the first area is an area corresponding to coordinates in the identification record included in the first identification record file
  • the second area is the second area Identifying an area corresponding to the coordinates in the identification record included in the record;
  • a supplementing module configured to supplement the first identification record file according to the identification record included in the second identification record file when the degree of fit is lower than a preset fitness degree threshold.
  • the identification device identifies the coordinates of the specified event according to the identification record of the monitoring image captured by the monitoring device, and determines the monitoring area that needs to be identified later.
  • the monitoring image captured by the monitoring device is recognized, only the determined monitoring area needs to be identified, the area of the recognized image can be reduced, the calculation amount of image recognition can be reduced, thereby reducing the hardware requirements of the monitoring system and Power consumption.
  • the identification device when the identification device identifies the monitoring image that is subsequently captured by the monitoring device according to the determined monitoring area, if the probability that the identified event is a specified event is higher than a preset threshold, according to the The coordinate corresponding to the identified event is generated and added to the first identification record file to timely replenish the first identification record file.
  • the monitoring area may be performed according to the newly added identification record. Corrected to improve the accuracy of the monitoring area.
  • the identification device deletes the expired record in the first identification record file before determining the monitoring area according to the first identification record file, so as to avoid that the identified area in the specified event set may change. As a result, the subsequent monitoring area is determined to be inaccurate, and the calculation complexity of the monitoring area is too high.
  • the identification device calculates a fitting degree between the first identification record file and the second identification record file before determining the monitoring area according to the first identification record file, when the fitting degree is low.
  • the first identification record file is supplemented according to the identification record in the second identification record file to correct the monitoring area determined according to the first identification record file, thereby improving the accuracy of the detection area determination.
  • FIG. 9 is a schematic structural diagram of an apparatus 900 according to an exemplary embodiment.
  • device 900 can be identification device 140 in the monitoring system shown in FIG.
  • device 900 includes a processing component 922 that further includes one or more processors, and memory resources represented by memory 932 for storing instructions, such as applications, that are executable by processing component 922.
  • An application stored in memory 932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 922 is configured to execute instructions to perform the image recognition method performed by the identification device described above.
  • Device 900 can also include a power component 926 configured to perform power management of device 900, a wired or wireless network interface 950 configured to connect device 900 to the network, and an input/output (I/O) interface 958.
  • Device 900 can operate based on an operating system stored in memory 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • non-transitory computer readable storage medium comprising instructions, such as a memory comprising instructions executable by a processor of an identification device to perform 2 as shown in various embodiments of the present application Or the method of Figure 6.
  • the non-transitory computer readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Alarm Systems (AREA)
  • Closed-Circuit Television Systems (AREA)
  • Image Analysis (AREA)

Abstract

本申请是关于一种图像识别方法、装置、设备及存储介质。该方法包括：获取第一识别记录文件，根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识别；本方案只需要对确定的监控区域进行识别即可，能够减少识别的图像面积，降低图像识别的计算量，从而降低了对监控系统的硬件要求以及电量消耗。

Description

图像识别方法、装置、设备及存储介质
本申请要求于2017年02月15日提交中国专利局、申请号为201710081274.1、发明名称为“图像识别方法及装置”的中国专利申请的优先权，上述申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及视频监控技术领域，特别涉及一种图像识别方法、装置设备及存储介质。
背景技术
随着视频监控技术的不断发展，公共场所中安装的摄像头等监控设备也越来越多，人工识别已经无法满足对监控设备拍摄到的监控图像进行及时有效的识别的要求。
在相关技术中，为了节约对监控图像进行识别的人工成本，提高对监控图像进行识别的及时性和准确性，目前很多监控系统可以对监控设备拍摄到的监控图像进行自动识别。具体的，监控系统按照预定的识别算法对监控设备拍摄到的监控图像的全部区域进行逐帧识别，以便及时的识别出监控图像中的指定事件，比如，识别出满足指定要求的人或物的事件，或者，识别出发生指定行为的事件。
在相关技术中，监控系统需要对监控设备拍摄到的监控图像中的全部区域都进行识别，监控系统的计算量较大，对监控系统的硬件要求较高，较大的计算量也需要消耗较高的电量。
发明内容
为了解决相关技术中监控系统的计算量较大，对监控系统的硬件要求以及消耗的电量较高的问题，本申请提供了一种图像识别方法、装置、设备及存储介质，技术方案如下：
一方面，提供了一种图像识别方法，所述方法包括：
获取第一识别记录文件，所述第一识别记录文件包含至少两条识别记录，每条识别记录包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事 件对应在监控图像中的坐标；
根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；
在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识别。
另一方面，提供了一种图像识别装置，所述装置包括：
第一文件获取模块，用于获取第一识别记录文件，所述第一识别记录文件包含至少两条识别记录，每条识别记录包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标；
区域确定模块，用于根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；
识别模块，用于在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识别。
另一方面，提供了一种识别设备，所述识别设备包含处理器和存储器，所述存储器中存储有指令，所述指令由所述处理器执行以实现上述的图像识别方法。
另一方面，提供了一种计算机可读存储介质，所述存储介质中存储有指令，所述指令由识别设备的处理器执行，以实现上述的图像识别方法。
本申请提供的技术方案可以包括以下有益效果：
识别设备根据对该监控设备拍摄到的监控图像的识别记录中，识别出指定事件的坐标进行分析，确定后续需要进行识别的监控区域，在对该监控设备后续拍摄到的监控图像进行识别时，只需要对确定的监控区域进行识别即可，能够减少识别的图像面积，降低图像识别的计算量，从而降低了对监控系统的硬件要求以及电量消耗。
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，并不能限制本申请。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本申请的实施例，并与说明书一起用于解释本申请的原理。
图1是本申请一个实施例示出的一种监控系统的结构示意图；
图2是根据一示例性实施例示出的图像识别方法的流程图；
图3是图2所示实施例涉及的一种坐标示意图；
图4是图2所示实施例涉及的一种监控区域示意图；
图5是图2所示实施例涉及的方案的实现流程图；
图6是根据一示例性实施例示出的图像识别方法的流程图；
图7是图6所示实施例涉及的一种区域示意图；
图8是根据一示例性实施例示出的图像识别装置的结构方框图；
图9是根据一示例性实施例示出的设备的结构示意图。
具体实施方式
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本申请相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本申请的一些方面相一致的装置和方法的例子。
请参考图1，其示出了本申请一个实施例示出的一种监控系统的结构示意图。该系统包括：若干个监控设备120和识别设备140。
其中，监控设备120可以是专用的监控摄像头，或者，包含有摄像头的其它电子设备(比如智能手机、平板电脑等)。识别设备140可以是通用计算机或者服务器。
监控设备120和识别设备140之间通过有线或无线网络相连接。其中，监控设备120可以将拍摄到的监控图像通过该有线或无线网络实时发送给识别设备140，识别设备140按照预定的识别算法对监控设备120拍摄到的监控图像进行识别。
该系统中还可以包含与识别设备140相连的图形显示设备160，比如，该图形显示设备160可以是显示器。识别设备140在对某个监控图像进行识别时，可以将该监控图像以及对该监控图像的识别结果聚合显示在图形显示设备160中。
在另一种可能的实现方式中，监控设备120和识别设备140也可以是同一个电子设备的不同组成部分，比如，监控设备120可以是电子设备中的摄像头，识别设备140可以是电子设备中的处理功能组件(包括处理器以及存储器等)，监控设备120与识别设备140通过电子设备中的通信总线相连。
在本申请实施例所示的方案中，识别设备140对某个监控设备120拍摄的监控图像进行识别的过程中，可以首先获取该监控设备的第一识别记录文件，该第一识别记录文件包含至少两条识别记录，每条识别记录中包含对该监控设备120拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标；识别设备140根据第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，该监控区域是监控设备拍摄到的监控图像中的部分区域，在对监控设备后续拍摄到的至少一帧监控图像进行识别时，识别设备140对该至少一帧监控图像中，对应于监控区域中的图像进行识别。
通过本申请实施例所示的方案，识别设备140可以根据对监控设备120拍摄到的监控图像的识别记录中，识别出指定事件的坐标进行分析，以确定后续需要进行识别的监控区域，在对该监控设备120后续拍摄到的监控图像进行识别时，只需要对确定的监控区域进行识别即可；比如，识别设备140在某个时间节点上，对监控设备120在该时间节点之前拍摄到的监控图像的识别记录进行分析，将识别记录识别出的指定事件在监控图像中相对集中的部分区域确定为监控区域，在该时间节点之后，识别设备140对监控设备120拍摄到的新的监控图像进行指定事件识别时，只对新的监控图像中的监控区域进行识别，而对于新的监控图像中除了该监控区域之外的其它区域的图像则不予识别，相比于对监控图像的整幅图像进行识别的方案，在识别算法相同的情况下，本申请实施例所示的方案能够减少识别的图像面积，降低图像识别的计算量，从而降低对监控系统的硬件要求以及电量消耗。
图2是根据一示例性实施例示出的一种图像识别方法的流程图，该方法可以应用于如图1所示的监控系统的识别设备140中。该图像识别方法可以包括 如下几个步骤：
步骤201，获取第一识别记录文件，该第一识别记录文件中包含至少两条识别记录，每条识别记录中包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标。
其中，第一识别记录文件是一个包含有至少两条识别记录的文件，每条识别记录中包含一个或一个以上的坐标，该坐标可以是识别设备在该识别记录对应的监控图像中识别到的指定事件在对应的监控图像中的坐标。
上述指定事件可以是监控图像中出现指定的人或物；比如，假设识别设备需要识别监控设备中穿黑色衣服的人或黑颜色的车辆，则指定事件可以是监控图像中出现穿黑色衣服的人或黑颜色的车辆；或者，假设识别设备需要识别交通事故，则指定事件可以是交通事故事件。
在本申请实施例中，识别设备识别到的一个指定事件可以只对应一个坐标，比如，指定事件对应的坐标可以是指定事件在监控图像中所处的区域的中心坐标；具体的，当指定事件为交通事故事件时，该指定事件对应的坐标可以是交通事故发生区域的中心坐标。或者，指定事件对应的坐标也可以是指定事件所涉及的对象的某个特定位置的坐标，比如，当指定事件为出现人物时，该指定事件对应的坐标可以是人物的头部中心所在的坐标。
在另一种可能的实现方式中，一个指定事件也可以对应多个坐标，比如，指定事件对应的坐标可以是指定事件所涉及的对象的多个特定位置的坐标，例如，当指定事件为出现人物时，该指定事件对应的坐标可以是人物的头部中心所在的坐标以及人物的躯干中心所在的坐标。
其中，上述坐标可以是与监控图像的分辨率对应的像素坐标，或者，也可以是与监控图像的尺寸相对应的尺寸坐标。
比如，请参考图3，其示出了本申请明实施例涉及的一种坐标示意图。如图3所示，若监控设备拍摄的监控图像的分辨率为1080*720，则以监控图像的左下角为原点，监控图像的下边界为横轴，左边界为纵轴设置坐标系，假设该监控设备拍摄到的某一帧监控图像中的指定事件出现的像素为该监控图像中左起第850个像素，且为下起第315个像素，则该指定事件的坐标可以设置为(850，315)。
或者，在图3中，若监控设备拍摄的监控图像的尺寸为1080mm*720mm，则以监控图像的左下角为原点，监控图像的下边界为横轴，左边界为纵轴设置 坐标系，假设该监控设备拍摄到的某一帧监控图像中的指定事件出现的位置距离该监控图像的左边界850mm，且距离为该监控图像的左边界315mm，则该指定事件的坐标可以设置为(850，315)。
步骤202，根据该第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域。
该监控区域可以是监控设备拍摄到的监控图像中的部分区域。在本申请实施例中，识别设备可以根据对监控设备拍摄到的监控图像的历史监控记录，将识别出的指定事件对应的坐标在监控图像中较为集中的区域确定为监控区域。
其中，识别设备可以确定该第一识别记录文件中的至少两条识别记录包含的坐标中的目标坐标，该目标坐标是可信度高于预设的可信度阈值的坐标，该可信度用于指示在对应的坐标处再次识别出该指定事件的概率，将包含目标坐标的最小区域确定为该监控区域。
其中，上述可信度高于预设的可信度阈值的目标坐标可以通过概率统计的方法来确定，具体比如，识别设备在确定目标坐标时，可以计算该第一识别记录文件中的至少两条识别记录包含的坐标的标准差，根据该标准差生成该第一识别记录文件中的至少两条识别记录包含的坐标对应的二维正态分布公式(该二维正态分布公式对应在空间坐标系中是一个轴对称的钟型曲面)，并根据该二维正态分布公式以及该可信度阈值，将该第一识别记录文件中的至少两条识别记录包含的坐标中，可信度高于该可信度阈值的坐标确定为目标坐标。
可选的，识别设备在包含可信度高于可信度阈值的坐标的最小区域确定为监控区域时，可以将确定出的目标坐标中，处于最外围的各个坐标所围成的区域确定为监控区域。
比如，请参考图4，其示出了本申请实施例涉及的一种监控区域示意图。如图4所示，区域40为监控图像的完整坐标区域，图4中每个“x”型标记所在坐标都是第一识别记录文件中的识别记录包含的一个坐标，识别设备通过概率统计方法确定出图4中集中在区域40中部区域的坐标为可信度高于预设的可信度阈值的坐标，并确定出包含可信度高于预设的可信度阈值的坐标中，最外围的各个坐标所围成的区域(即图4中的区域41)，该区域41对应的坐标区域就是确定出的监控区域。
步骤203，在对该监控设备后续拍摄到的至少一帧监控图像进行识别时，对该至少一帧监控图像中，对应于该监控区域中的图像进行识别。
监控设备在确定出监控区域后，后续获取到监控设备拍摄的至少一帧监控图像后，只需要按照预定的识别算法对该至少一帧监控图像中对应该监控区域的部分图像进行识别，以确定对应该监控区域的部分图像中是否发生指定事件。
通过本申请实施例上述步骤201至步骤203所示的方案，假设监控设备拍摄到的一帧监控图像的面积是100*100，识别设备在某个时间节点上，对监控设备在该时间节点之前拍摄到的监控图像的识别记录进行分析，将识别记录识别出的指定事件在监控图像中相对集中的部分区域确定为监控区域(假设监控区域是监控图像中面积为50*50的区域)，在该时间节点之后，识别设备对监控设备拍摄到的新的监控图像进行指定事件识别时，只对新的监控图像中面积为50*50监控区域进行识别，而对于新的监控图像中除了该监控区域之外的其它区域的图像则不予识别，相比于相关技术中对监控图像的整幅图像中的所有区域都进行识别的方案，本申请实施例所示的方案只需要对整幅图像中的部分区域进行识别，因而能够解决相关技术中对监控图像进行识别时的计算量较大，对硬件的要求以及消耗的电量较高的问题，达到降低图像识别的计算量，从而降低对监控系统的硬件要求以及电量消耗的效果。
可选的，识别设备在按照确定的监控区域对后续的监控图像进行部分识别时，还可以对第一识别记录文件进行补充，相应的，本申请实施例还可以包含下述步骤204和步骤205。
步骤204，对于该至少一帧监控图像中的每一帧监控图像，计算在该监控图像中识别到的事件为该指定事件的事件概率。
监控设备在按照预定的识别算法，对监控图像中对应监控区域的部分图像进行识别时，可以计算出识别出的事件为指定事件的事件概率。
具体比如，假设指定事件为出现身穿皮夹克黑裤子黑皮鞋，头戴白色鸭舌帽的人物的事件，监控设备可以将当前识别的监控图像中出现的每个人物的服饰特征与指定事件中的人物的服饰特征进行比对，以计算每个人物的服饰特征与指定事件中的人物的服饰特征的相似度，每个人物的服饰特征与指定事件中的人物的服饰特征之间的相似度即可以作为上述事件概率，或者，也可以根据该服饰特征之间的相似度得到上述事件概率。比如，识别设备可以将上述相似度乘以预定系数(该预定系数为可以是接近且小于1的正数，比如0.9)，并将得到的数值作为上述事件概率；或者，识别设备中也可以存储相似度所在的区 间与事件概率之间的对应关系，识别设备计算获得上述相似度之后，即可以确定该相似度所在的区间，并进一步查询该相似度所在的区间对应的事件概率。
步骤205，当该事件概率高于预设的概率阈值时，根据该事件对应在该监控图像中的坐标生成该事件的识别记录，将该事件的识别记录添加至该第一识别记录文件。
其中，上述概率阈值可以是根据实际应用需求设置的一个百分比数值。当识别设备在上述步骤204中计算出的事件概率高于概率阈值(比如95％)时，可以认为上述识别到的事件是指定事件，此时，识别设备可以获取上述识别到的事件对应在当前识别的监控图像中的坐标，生成包含该坐标的识别记录，并将生成的识别记录添加至第一识别记录文件中，以对第一识别记录文件进行及时补充，后续确定新的监控区域时，可以根据新补充的识别记录对监控区域进行修正，以提高监控区域确定的准确性。
在实际应用中，在识别设备初始安装后，识别设备首次接收到该识别设备拍摄的监控图像时，第一识别记录文件中尚不存在识别记录，此时，识别设备可以对该识别设备拍摄的监控图像的全部区域进行识别，或者也可以按照预设的监控区域(该预设的监控区域可以是监控图像中的默认区域，或者，也可以是用户或监控人员人工框选的区域)对该识别设备拍摄的监控图像进行识别，在识别出指定事件时生成识别记录并添加至第一识别记录文件。在此之后，识别设备可以每隔预定的时间(比如一天或一星期)获取一次上述第一识别记录文件，并根据获取到的第一识别记录文件确定监控区域，每次确定监控区域之后，基于确定的监控区域对监控设备后续拍摄到的监控图像进行事件识别，在识别出指定事件时生成识别记录并添加至第一识别记录文件，直至下一次获取第一识别记录文件并确定新的监控区域；或者，识别设备也可以在每次对监控设备拍摄到的监控图像进行识别之前获取上述第一识别记录文件并确定监控区域，基于确定的监控区域对待识别的监控图像进行事件识别，在识别出指定事件时生成识别记录并添加至第一识别记录文件。
具体的，请参考图5，其示出了本申请实施例涉及方案的实现流程图。假设监控设备为监控摄像头，识别设备为监控室中的监控服务器，如图5所示，监控摄像头安装后，将拍摄到的监控图像发送给监控服务器，监控服务器初始时对监控摄像头发送的监控图像的全部区域进行识别，在识别出指定事件时，将识别结果和监控图像聚合显示在监控屏幕(即图1所示的图形显示设备160) 中，同时生成包含该指定事件对应在监控图像中的坐标的识别记录，并将识别记录添加至第一识别记录文件中。后续每隔一天，监控服务器获取一次该第一识别记录文件，并根据获取到的第一识别记录文件中的识别记录确定监控区域，在后续的一天时间内，按照确定的监控区域对该监控摄像头拍摄的监控图像进行识别(即只对监控图像中对应监控区域的部分图像进行识别)，并在识别出指定事件时，生成包含该指定事件对应在监控图像中的坐标的识别记录，并将识别记录添加至第一识别记录文件中。
综上所述，本申请实施例所示的方法，识别设备根据对监控设备拍摄到的监控图像的识别记录中，识别出指定事件的坐标进行分析，确定后续需要进行识别的监控区域，在对该监控设备后续拍摄到的监控图像进行识别时，只需要对确定的监控区域进行识别即可，能够减少识别的图像面积，降低图像识别的计算量，从而降低对监控系统的硬件要求以及电量消耗。
此外，本申请实施例所示的方案，识别设备按照确定的监控区域对监控设备后续拍摄到的监控图像进行识别时，如果识别出的事件是指定事件的概率高于预设阈值，则根据该识别出的事件对应的坐标生成识别记录并添加至第一识别记录文件中，以对第一识别记录文件进行及时补充，后续确定新的监控区域时，可以根据新补充的识别记录对监控区域进行修正，以提高监控区域确定的准确性。
图6是根据一示例性实施例示出的一种图像识别方法的流程图，该方法可以应用于如图1所示的监控系统的识别设备140中。该图像识别方法可以包括如下几个步骤：
步骤601，获取第一识别记录文件，该第一识别记录文件中包含至少两条识别记录，每条识别记录中包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标。
步骤602，确定该第一识别记录文件中是否包含过期记录。
其中，第一识别记录文件中的每个识别记录除了包含对应的坐标之外，还可以包含对应的识别时间，而过期记录则是对应的识别时间与当前时间之间的间隔大于预设时间间隔的记录。
在实际应用中，随着时间的推移，识别出的指定事件集中的区域可能会发生变化，比如，监控设备的拍摄方向可能会随着时间的推移产生变化，导致指 定事件对应的坐标所集中的区域发生了变化；同时，随着时间的推移，第一识别记录文件中的识别记录也越来越多，当识别记录过多时，也可能会导致后续确定监控区域的计算复杂的增加。因此，为了避免识别出的指定事件对应的坐标所集中的区域可能发生变化而导致后续监控区域确定不准确，以及确定监控区域的计算复杂度太高的情形，在本申请实施例中，监控设备在获取到第一识别记录文件后，可以计算第一识别记录文件中每条识别记录对应的识别时间距离当前时间之间的间隔，并将识别时间距离当前时间之间的间隔大于预设时间间隔(比如3个月或半年)的识别记录确定为过期记录。
步骤603，将该过期记录从该第一识别记录文件中删除。
在确定出第一识别记录文件中的过期记录后，识别设备可以将第一识别记录文件中的过期记录删除，以避免识别出的指定事件集中的区域可能发生变化而导致后续监控区域确定不准确，以及确定监控区域的计算复杂度太高的问题。
可选的，本申请实施例所示的方案中，识别设备还可以对监控设备拍摄到的监控图像中的部分图像中的全部区域进行识别，并根据识别结果对第一识别记录文件进行补充。相应的，本申请实施例还可以包含下述步骤604至步骤607。
步骤604，获取第二识别记录文件，该第二识别记录文件中的每条识别记录包含对该监控设备拍摄到的部分监控图像的全部区域进行识别时，识别到的该指定事件对应在监控图像中的坐标。
其中，上述的部分监控图像是第一识别记录文件中包含的识别记录对应的监控图像之外的图像。
在实际应用中，并不是所有的指定事件都发生在识别设备确定的监控区域中，即在监控设备拍摄到的监控图像中的监控区域之外，也有可能会发生指定事件。为了减少因识别设备只对监控图像中的监控区域进行识别而导致的部分指定事件漏识别的情形，在本申请实施例中，识别设备可以从监控设备拍摄到的全部监控图像中提取出部分监控图像，对该部分监控图像的全部区域进行识别，并在识别出指定事件时，根据识别出的指定事件对应在监控图像中的坐标生成识别记录，并添加至第二识别记录文件中。
在一种可能的实现方式中，识别设备对监控设备拍摄到的监控图像进行识别时，可以从监控设备拍摄到的全部监控图像中提取出一定比例(比如10％) 的监控图像，并对提取出的监控图像的全部区域进行识别，而其余部分的监控图像则按照确定的监控区域进行识别。
其中，在提取一定比例的监控图像时，识别设备可以按照随机算法从全部监控图像中提取出一定比例的监控图像；或者，识别设备也可以按照预定的采样率从全部监控图像中进行采样，以提取一定比例的监控图像。具体比如，假设上述提取比例为10％，监控设备拍摄并发送了100帧监控图像，识别设备可以从该100帧监控图像中随机提取出10帧监控图像进行全部区域的识别，而其它90帧监控图像则按照确定的监控区域进行识别；或者，监控设备也可以按照100帧监控图像的拍摄时间顺序对该100帧监控图像进行排序，并将每10帧监控图像中的最后一帧(或者第二帧、第三帧等)监控图像提取出来进行全部区域的识别，而每10帧监控图像中的前9帧监控图像则按照确定的监控区域进行识别。
步骤605，计算该第一识别记录文件与该第二识别记录文件之间的拟合度。
其中，该拟合度用于指示第一区域和第二区域之间的相似程度；其中，第一区域是第一识别记录文件包含的识别记录中的坐标对应的区域，第二区域是第二识别记录包含的识别记录中的坐标对应的区域。
具体的，在本申请实施例中，识别设备在计算第一识别记录文件与该第二识别记录文件之间的拟合度时，可以首先确定第一识别记录文件包含的识别记录中的坐标对应的第一区域，以及第二识别记录文件包含的识别记录中的坐标对应的第二区域，其中，第一区域和第二区域的确定方式与上述步骤202中确定监控区域的步骤类似，识别设备计算第二区域与第一区域之间的共同区域的面积占第二区域的总面积的比例，并根据该共同区域的面积占第二区域的总面积的比例计算上述拟合度。比如，识别设备可以直接将上述共同区域的面积占第二区域的总面积的比例作为第一识别记录文件包含的识别记录中的坐标与该第二识别记录文件包含的识别记录中的坐标之间的拟合度。
比如，请参考7，其示出了本申请实施例涉及的一种区域示意图。如图7所示，区域70为监控图像的完整坐标区域，识别设备计算出第一识别记录文件包含的识别记录中的坐标对应的第一区域为区域71，计算出第二识别记录文件包含的识别记录中的坐标对应的第二区域为区域72，其中，区域71与区域72的共同区域为区域73(即图7中斜线部分)，则识别设备可以计算区域73的面积占区域72的面积的比例，并将计算出的结果作为第一识别记录文件包 含的识别记录中的坐标与该第二识别记录文件包含的识别记录中的坐标之间的拟合度。
可选的，第二识别记录文件中的每条识别记录也可以包含对应的识别时间，在计算第一识别记录文件包含的识别记录中的坐标与第二识别记录文件包含的识别记录中的坐标之前，识别设备也可以确定第二识别记录文件中的过期记录，并将第二识别记录文件中的过期记录删除。
步骤606，检测该拟合度是否低于预设的拟合度阈值；若是，进入步骤607，否则，进入步骤608。
其中，上述拟合度阈值可以是预先设置，或者按照实际的应用需求设置的一个比例数值。比如，以上述共同区域的面积占第二区域的总面积的比例为上述拟合度为例，当上述共同区域的面积占第二区域的总面积的比例不低于预设的拟合度阈值(比如98％)时，可以认为监控设备拍摄到的监控图像中，处于监控区域之外的其他区域发生指定事件的概率较低，此时可以进入步骤608；反之，若当上述共同区域的面积占第二区域的总面积的比例低于预设的拟合度阈值时，可以认为监控设备拍摄到的监控图像中，处于监控区域之外的其他区域发生指定事件的概率较高，此时可以进入步骤607处理后，再进入步骤608。
步骤607，根据该第二识别记录文件包含的识别记录对该第一识别记录文件进行补充。
在本申请实施例中，当第一识别记录文件包含的识别记录中的坐标与该第二识别记录文件包含的识别记录中的坐标之间的拟合度低于预设的拟合度阈值时，识别设备可以对第一识别记录文件进行补充，以及时修正监控区域。具体的，在对第一识别记录文件进行补充时，识别设备可以将第二识别记录文件中包含的识别记录都添加至第一识别记录文件中；或者，识别设备也可以只将第二识别记录文件包含的识别记录中，对应的坐标处于上述共同区域之外的识别记录添加至第一识别记录文件中。
在另一种可能的实现方式中，第二识别记录文件中包含的识别记录对应的识别时间距离当前时间之间的间隔处于预设的时间间隔之内(比如一星期内或者一个月内)，识别设备确定是否需要将第二识别记录文件包含的识别记录添加至第一识别记录文件中时，也可以不计算两者之间的拟合度，而是在确定第一识别记录文件包含的识别记录中的坐标对应的第一区域，以及第二识别记录文件包含的识别记录中的坐标对应的第二区域后，计算上述第二识别记录文件 中的识别记录包含的坐标中，处于第一区域之外的坐标的个数是否大于预设的个数阈值，若是，则确定将第二识别记录文件包含的识别记录添加至第一识别记录文件中，以对根据第一识别记录文件确定出的监控区域进行修正。其中，上述个数阈值可以由开发人员或者维护人员预先设置。
步骤608，根据该第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域。
步骤609，在对该监控设备后续拍摄到的至少一帧监控图像进行识别时，对该至少一帧监控图像中，对应于该监控区域中的图像进行识别。
可选的，识别设备在按照确定的监控区域对后续的监控图像进行部分识别时，还可以对第一识别记录文件进行补充，相应的，本申请实施例还可以包含下述步骤610和步骤611。
步骤610，对于该至少一帧监控图像中的每一帧监控图像，计算在该监控图像中识别到的事件为该指定事件的事件概率。
步骤611，当该事件概率高于预设的概率阈值时，根据该事件对应在该监控图像中的坐标生成该事件的识别记录，将该事件的识别记录添加至该第一识别记录文件。
上述步骤608至步骤611的实现过程可以参考图2所示实施例中的步骤202至步骤205下的描述，此处不再赘述。
可选的，在本申请实施例所示的方案中，在将识别出的事件的识别记录添加至第一识别记录文件后，当用户确定该识别出的事件不是指定事件时，还可以将对应的识别记录从第一识别记录文件中删除，相应的，本申请实施例还可以包含下述步骤612和步骤613。
步骤612，当该事件概率高于预设的概率阈值时，展示该监控图像。
在本申请实施例中，识别设备检测出某个事件的事件概率高于预设的概率阈值时，可以认为检测到疑似的指定事件，此时，可以对识别出的事件对应的监控图像进行展示，由用户或监控人员确定是否真的发生指定事件。
步骤613，接收针对该监控图像发出的事件否认响应，该事件否认响应用于指示该事件不是该指定事件，将该事件的识别记录从该第一识别记录文件中删除。
当用户或监控人员根据展示的监控图像确定未发生指定事件时，用户或监控人员可以基于展示的监控图像发出事件否认响应，此时，识别设备将该事件 的识别记录从第一识别记录文件中删除，以便错误的识别结果影响后续监控区域确定的准确性。
具体的，假设监控设备为监控摄像头，识别设备为监控室中的监控服务器，监控摄像头安装后，将拍摄到的监控图像发送给监控服务器。
在监控摄像头安装后的第一天，监控服务器初对监控摄像头发送的监控图像的全部区域进行识别，在识别出指定事件时，将识别结果和监控图像聚合显示在监控屏幕中，同时生成包含该指定事件对应在监控图像中的坐标以及识别时间的识别记录，并将识别记录添加至第一识别记录文件中。
在监控摄像头安装后的第二天，监控服务器获取该第一识别记录文件，根据获取到的第一识别记录文件中的识别记录确定监控区域，对于该监控摄像头拍摄的监控图像每10帧监控图像，识别设备按照确定的监控区域对其中的前9帧监控图像进行识别，并对其中最后一帧监控图像进行全部区域识别；在对上述前9帧监控图像进行识别且识别出指定事件时，生成包含该指定事件的坐标以及识别时间的识别记录，并将识别记录添加至第一识别记录文件中；在对上述最后一帧监控图像进行识别且识别出指定事件时，生成包含该指定事件的坐标以及识别时间的识别记录，并将识别记录添加至第二识别记录文件中。
后续每隔一天，监控服务器获取一次该第一识别记录文件和第二识别记录文件，先按照识别时间将第一识别记录文件和第二识别记录文件中的过期记录删除，再计算第一识别记录文件中的识别记录的坐标与第二识别记录文件中的识别记录的坐标之间的拟合度，当两者之间的拟合度不低于预设的拟合度阈值时，识别设备根据获取到的第一识别记录文件中的识别记录确定监控区域，而当两者之间的拟合度低于预设的拟合度阈值时，识别设备将第二识别记录文件中的识别记录添加至第一识别记录文件，并根据添加了识别记录之后的第一识别记录文件确定监控区域；在确定监控区域之后，对于该监控摄像头拍摄的监控图像每10帧监控图像，识别设备按照确定的监控区域对其中的前9帧监控图像进行识别，并对其中最后一帧监控图像进行全部区域识别；在对上述前9帧监控图像进行识别且识别出指定事件时，生成包含该指定事件的坐标的识别记录，并将识别记录添加至第一识别记录文件中；在对上述最后一帧监控图像进行识别且识别出指定事件时，生成包含该指定事件的坐标的识别记录，并将识别记录添加至第二识别记录文件中。
综上所述，本申请实施例所示的方法，识别设备根据对监控设备拍摄到的 监控图像的识别记录中，识别出指定事件的坐标进行分析，确定后续需要进行识别的监控区域，在对该监控设备后续拍摄到的监控图像进行识别时，只需要对确定的监控区域进行识别即可，能够减少识别的图像面积，降低图像识别的计算量，从而降低对监控系统的硬件要求以及电量消耗。
此外，本申请实施例所示的方案，识别设备按照确定的监控区域对监控设备后续拍摄到的监控图像进行识别时，如果识别出的事件是指定事件的概率高于预设阈值，则根据该识别出的事件对应的坐标生成识别记录并添加至第一识别记录文件中，以对第一识别记录文件进行及时补充，后续确定新的监控区域时，可以根据新补充的识别记录对监控区域进行修正，以提高监控区域确定的准确性。
另外，本申请实施例所示的方法，识别设备在根据第一识别记录文件确定监控区域之前，先将第一识别记录文件中的过期记录删除，避免识别出的指定事件集中的区域可能发生变化而导致后续监控区域确定不准确，以及确定监控区域的计算复杂度太高的情形。
另外，本申请实施例所示的方法，识别设备在根据第一识别记录文件确定监控区域之前，还计算第一识别记录文件和第二识别记录文件之间的拟合度，当拟合度低于拟合度阈值时，根据第二识别记录文件中的识别记录对第一识别记录文件进行补充，以修正按照第一识别记录文件确定出的监控区域，从而提高监控区域确定的准确性。
图8是根据一示例性实施例示出的一种图像识别装置的结构方框图。该图像识别装置可以用于设置在图1所示的监控系统的监控设备140中，以执行图2或图6所示实施例中的全部或者部分步骤。该图像识别装置可以包括：
第一文件获取模块801，用于获取第一识别记录文件，所述第一识别记录文件包含至少两条识别记录，每条识别记录中包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标；
区域确定模块802，用于根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；
识别模块803，用于在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识 别。
可选的，所述区域确定模块，包括：
坐标确定单元，用于确定所述第一识别记录文件中的至少两条识别记录包含的坐标中的目标坐标，所述目标坐标是可信度高于预设的可信度阈值的坐标，所述可信度用于指示在对应的坐标处再次识别出所述指定事件的概率；
区域确定单元，用于将包含所述目标坐标的最小区域确定为所述监控区域。
可选的，所述坐标确定单元，包括：
计算子单元，用于计算所述第一识别记录文件中的至少两条识别记录包含的坐标的标准差；
生成子单元，用于根据所述标准差生成所述第一识别记录文件中的至少两条识别记录包含的坐标对应的二维正态分布公式；
坐标确定子单元，用于根据所述二维正态分布公式以及所述可信度阈值，将所述第一识别记录文件中的至少两条识别记录包含的坐标中，可信度高于所述可信度阈值的坐标确定为所述目标坐标。
可选的，所述装置还包括：
概率计算模块，用于对于所述至少一帧监控图像中的每一帧监控图像，计算在所述监控图像中识别到的事件为所述指定事件的事件概率；
记录生成模块，用于当所述事件概率高于预设的概率阈值时，根据所述事件对应在所述监控图像中的坐标生成所述事件的识别记录；
添加模块，用于将所述事件的识别记录添加至所述第一识别记录文件。
可选的，所述装置还包括：
展示模块，用于当所述事件概率高于预设的概率阈值时，展示所述监控图像；
响应接收模块，用于接收针对所述监控图像发出的事件否认响应，所述事件否认响应用于指示所述事件不是所述指定事件；
第一删除模块，用于将所述事件的识别记录从所述第一识别记录文件中删除。
可选的，所述每条识别记录中还包含对应的识别时间，所述装置还包括：
记录确定模块，用于在根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，确定所述第一识别记录文件中是否包含过期记 录，所述过期记录是对应的识别时间与当前时间之间的间隔大于预设时间间隔的记录；
第二删除模块，用于将所述过期记录从所述第一识别记录文件中删除。
可选的，所述装置还包括：
第二文件获取模块，用于在所述区域确定模块根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，获取第二识别记录文件，所述第二识别记录文件中的每条识别记录包含对所述监控设备拍摄到的部分监控图像的全部区域进行识别时，识别到的所述指定事件对应在监控图像中的坐标，所述部分监控图像是所述第一识别记录文件中包含的识别记录对应的监控图像之外的图像；
拟合度计算模块，用于计算所述第一识别记录文件包含的识别记录中的坐标与所述第二识别记录文件包含的识别记录中的坐标之间的拟合度，所述拟合度用于指示第一区域和第二区域之间的相似程度；所述第一区域是所述第一识别记录文件包含的识别记录中的坐标对应的区域，所述第二区域是所述第二识别记录包含的识别记录中的坐标对应的区域；
补充模块，用于当所述拟合度低于预设的拟合度阈值时，根据所述第二识别记录文件中包含的识别记录对所述第一识别记录文件进行补充。
综上所述，本申请实施例所示的装置，识别设备根据对该监控设备拍摄到的监控图像的识别记录中，识别出指定事件的坐标进行分析，确定后续需要进行识别的监控区域，在对该监控设备后续拍摄到的监控图像进行识别时，只需要对确定的监控区域进行识别即可，能够减少识别的图像面积，降低图像识别的计算量，从而降低了对监控系统的硬件要求以及电量消耗。
此外，本申请实施例所示的装置，识别设备按照确定的监控区域对监控设备后续拍摄到的监控图像进行识别时，如果识别出的事件是指定事件的概率高于预设阈值，则根据该识别出的事件对应的坐标生成识别记录并添加至第一识别记录文件中，以对第一识别记录文件进行及时补充，后续确定新的监控区域时，可以根据新补充的识别记录对监控区域进行修正，以提高监控区域确定的准确性。
另外，本申请实施例所示的装置，识别设备在根据第一识别记录文件确定监控区域之前，先将第一识别记录文件中的过期记录删除，避免识别出的指定事件集中的区域可能发生变化而导致后续监控区域确定不准确，以及确定监控 区域的计算复杂度太高的情形。
另外，本申请实施例所示的装置，识别设备在根据第一识别记录文件确定监控区域之前，还计算第一识别记录文件和第二识别记录文件之间的拟合度，当拟合度低于拟合度阈值时，根据第二识别记录文件中的识别记录对第一识别记录文件进行补充，以修正按照第一识别记录文件确定出的监控区域，从而提高监控区域确定的准确性。
图9是根据一示例性实施例示出的一种设备900的结构示意图。例如，设备900可以是图1所示的监控系统中的识别设备140。参照图9，设备900包括处理组件922，其进一步包括一个或多个处理器，以及由存储器932所代表的存储器资源，用于存储可由处理组件922执行的指令，例如应用程序。存储器932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件922被配置为执行指令，以执行上述由识别设备执行的图像识别方法。
设备900还可以包括一个电源组件926，被配置为执行设备900的电源管理，一个有线或无线网络接口950，被配置为将设备900连接到网络，和一个输入输出(I/O)接口958。设备900可以操作基于存储在存储器932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
在示例性实施例中，还提供了一种包括指令的非临时性计算机可读存储介质，例如包括指令的存储器，上述指令可由识别设备的处理器执行以完成本申请各个实施例所示的2或图6方法。例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
本领域技术人员在考虑说明书及实践这里公开的申请后，将容易想到本申请的其它实施方案。本申请旨在涵盖本申请的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本申请的一般性原理并包括本申请未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本申请的真正范围和精神由下面的权利要求指出。
应当理解的是，本申请并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本申请的范围仅由所附的权利要求来限制。

Claims (16)

  1. 一种图像识别方法，其特征在于，所述方法包括：
    识别设备获取第一识别记录文件，所述第一识别记录文件包含至少两条识别记录，每条识别记录包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标；
    所述识别设备根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；
    所述识别设备在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识别。
  2. 根据权利要求1所述的方法，其特征在于，所述识别设备根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，包括：
    所述识别设备确定所述第一识别记录文件中的至少两条识别记录包含的坐标中的目标坐标，所述目标坐标是可信度高于预设的可信度阈值的坐标，所述可信度用于指示在对应的坐标处再次识别出所述指定事件的概率；
    所述识别设备将包含所述目标坐标的最小区域确定为所述监控区域。
  3. 根据权利要求2所述的方法，其特征在于，所述识别设备确定所述第一识别记录文件中的至少两条识别记录包含的坐标中的目标坐标，包括：
    所述识别设备计算所述第一识别记录文件中的至少两条识别记录包含的坐标的标准差；
    所述识别设备根据所述标准差生成所述第一识别记录文件中的至少两条识别记录包含的坐标对应的二维正态分布公式；
    所述识别设备根据所述二维正态分布公式以及所述可信度阈值，将所述第一识别记录文件中的至少两条识别记录包含的坐标中，可信度高于所述可信度阈值的坐标确定为所述目标坐标。
  4. 根据权利要求1至3任一所述的方法，其特征在于，所述方法还包括：
    对于所述至少一帧监控图像中的每一帧监控图像，所述识别设备计算在所 述监控图像中识别到的事件为所述指定事件的事件概率；
    当所述事件概率高于预设的概率阈值时，所述识别设备根据所述事件对应在所述监控图像中的坐标生成所述事件的识别记录；
    所述识别设备将所述事件的识别记录添加至所述第一识别记录文件。
  5. 根据权利要求4所述的方法，其特征在于，所述方法还包括：
    当所述事件概率高于预设的概率阈值时，所述识别设备展示所述监控图像；
    所述识别设备接收针对所述监控图像发出的事件否认响应，所述事件否认响应用于指示所述事件不是所述指定事件；
    所述识别设备将所述事件的识别记录从所述第一识别记录文件中删除。
  6. 根据权利要求1至3任一所述的方法，其特征在于，所述每条识别记录中还包含对应的识别时间，所述方法还包括：
    所述识别设备在根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，确定所述第一识别记录文件中是否包含过期记录，所述过期记录是对应的识别时间与当前时间之间的间隔大于预设时间间隔的记录；
    所述识别设备将所述过期记录从所述第一识别记录文件中删除。
  7. 根据权利要求1至3任一所述的方法，其特征在于，所述识别设备在根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，所述方法还包括：
    所述识别设备获取第二识别记录文件，所述第二识别记录文件中的每条识别记录包含对所述监控设备拍摄到的部分监控图像的全部区域进行识别时，识别到的所述指定事件对应在监控图像中的坐标，所述部分监控图像是所述第一识别记录文件中包含的识别记录对应的监控图像之外的图像；
    所述识别设备计算所述第一识别记录文件与所述第二识别记录文件之间的拟合度，所述拟合度用于指示第一区域和第二区域之间的相似程度；所述第一区域是所述第一识别记录文件包含的识别记录中的坐标对应的区域，所述第二区域是所述第二识别记录包含的识别记录中的坐标对应的区域；
    当所述拟合度低于预设的拟合度阈值时，所述识别设备根据所述第二识别 记录文件中包含的识别记录对所述第一识别记录文件进行补充。
  8. 一种图像识别装置，其特征在于，所述装置包括：
    第一文件获取模块，用于获取第一识别记录文件，所述第一识别记录文件包含至少两条识别记录，每条识别记录包含对监控设备拍摄到的监控图像进行识别时，识别到的指定事件对应在监控图像中的坐标；
    区域确定模块，用于根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域，所述监控区域是所述监控设备拍摄到的监控图像中的部分区域；
    识别模块，用于在对所述监控设备后续拍摄到的至少一帧监控图像进行识别时，对所述至少一帧监控图像中，对应于所述监控区域中的图像进行识别。
  9. 根据权利要求8所述的装置，其特征在于，所述区域确定模块，包括：
    坐标确定单元，用于确定所述第一识别记录文件中的至少两条识别记录包含的坐标中的目标坐标，所述目标坐标是可信度高于预设的可信度阈值的坐标，所述可信度用于指示在对应的坐标处再次识别出所述指定事件的概率；
    区域确定单元，用于将包含所述目标坐标的最小区域确定为所述监控区域。
  10. 根据权利要求9所述的装置，其特征在于，所述坐标确定单元，包括：
    计算子单元，用于计算所述第一识别记录文件中的至少两条识别记录包含的坐标的标准差；
    生成子单元，用于根据所述标准差生成所述第一识别记录文件中的至少两条识别记录包含的坐标对应的二维正态分布公式；
    坐标确定子单元，用于根据所述二维正态分布公式以及所述可信度阈值，将所述第一识别记录文件中的至少两条识别记录包含的坐标中，可信度高于所述可信度阈值的坐标确定为所述目标坐标。
  11. 根据权利要求8至10任一所述的装置，其特征在于，所述装置还包括：
    概率计算模块，用于对于所述至少一帧监控图像中的每一帧监控图像，计算在所述监控图像中识别到的事件为所述指定事件的事件概率；
    记录生成模块，用于当所述事件概率高于预设的概率阈值时，根据所述事 件对应在所述监控图像中的坐标生成所述事件的识别记录；
    添加模块，用于将所述事件的识别记录添加至所述第一识别记录文件。
  12. 根据权利要求11所述的装置，其特征在于，所述装置还包括：
    展示模块，用于当所述事件概率高于预设的概率阈值时，展示所述监控图像；
    响应接收模块，用于接收针对所述监控图像发出的事件否认响应，所述事件否认响应用于指示所述事件不是所述指定事件；
    第一删除模块，用于将所述事件的识别记录从所述第一识别记录文件中删除。
  13. 根据权利要求8至10任一所述的装置，其特征在于，所述每条识别记录中还包含对应的识别时间，所述装置还包括：
    记录确定模块，用于在根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，确定所述第一识别记录文件中是否包含过期记录，所述过期记录是对应的识别时间与当前时间之间的间隔大于预设时间间隔的记录；
    第二删除模块，用于将所述过期记录从所述第一识别记录文件中删除。
  14. 根据权利要求8至10任一所述的装置，其特征在于，所述装置还包括：
    第二文件获取模块，用于在所述区域确定模块根据所述第一识别记录文件中的至少两条识别记录包含的坐标确定监控区域之前，获取第二识别记录文件，所述第二识别记录文件中的每条识别记录包含对所述监控设备拍摄到的部分监控图像的全部区域进行识别时，识别到的所述指定事件对应在监控图像中的坐标，所述部分监控图像是所述第一识别记录文件中包含的识别记录对应的监控图像之外的图像；
    拟合度计算模块，用于计算所述第一识别记录文件包含的识别记录中的坐标与所述第二识别记录文件包含的识别记录中的坐标之间的拟合度，所述拟合度用于指示第一区域和第二区域之间的相似程度；所述第一区域是所述第一识别记录文件包含的识别记录中的坐标对应的区域，所述第二区域是所述第二识别记录包含的识别记录中的坐标对应的区域；
    补充模块，用于当所述拟合度低于预设的拟合度阈值时，根据所述第二识别记录文件中包含的识别记录对所述第一识别记录文件进行补充。
  15. 一种识别设备，其特征在于，所述识别设备包含处理器和存储器，所述存储器中存储有指令，所述指令由所述处理器执行以实现如权利要求1至7任一所述的图像识别方法。
  16. 一种计算机可读存储介质，其特征在于，所述存储介质中存储有指令，所述指令由识别设备的处理器执行，以实现如权利要求1至7任一所述的图像识别方法。
PCT/CN2018/075379 2017-02-15 2018-02-06 图像识别方法、装置、设备及存储介质 WO2018149322A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP18754005.9A EP3585052B1 (en) 2017-02-15 2018-02-06 Image identification method, device, apparatus, and data storage medium
US16/526,561 US10936876B2 (en) 2017-02-15 2019-07-30 Content recognition method, system and storage medium thereof

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710081274.1 2017-02-15
CN201710081274.1A CN108289191B (zh) 2017-02-15 2017-02-15 图像识别方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/526,561 Continuation US10936876B2 (en) 2017-02-15 2019-07-30 Content recognition method, system and storage medium thereof

Publications (1)

Publication Number Publication Date
WO2018149322A1 true WO2018149322A1 (zh) 2018-08-23

Family

ID=62831521

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/075379 WO2018149322A1 (zh) 2017-02-15 2018-02-06 图像识别方法、装置、设备及存储介质

Country Status (5)

Country Link
US (1) US10936876B2 (zh)
EP (1) EP3585052B1 (zh)
CN (1) CN108289191B (zh)
MA (1) MA47512A (zh)
WO (1) WO2018149322A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112468843A (zh) * 2020-10-26 2021-03-09 国家广播电视总局广播电视规划院 视频去重方法及装置
CN116030388A (zh) * 2022-12-30 2023-04-28 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10832043B2 (en) * 2018-02-06 2020-11-10 Disney Enterprises, Inc. Variable resolution recognition
CN112118478B (zh) * 2020-09-23 2022-08-19 广州虎牙科技有限公司 文本处理方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1722825A (zh) * 2004-07-16 2006-01-18 赖金鼎 设定影像监控区的装置及其方法
CN103327306A (zh) * 2013-06-14 2013-09-25 广东威创视讯科技股份有限公司 视频监控存储方法及装置
JP2014036414A (ja) * 2012-08-10 2014-02-24 Hitachi Kokusai Electric Inc 監視システム
CN103646257A (zh) * 2013-12-30 2014-03-19 中国科学院自动化研究所 一种基于视频监控图像的行人检测和计数方法
CN103686074A (zh) * 2013-11-20 2014-03-26 南京熊猫电子股份有限公司 视频监控中移动目标的定位方法

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101392294B1 (ko) * 2006-04-17 2014-05-27 오브젝트비디오 인코퍼레이티드 통계적인 픽셀 모델링을 이용한 비디오 분할
CN101729867A (zh) * 2009-11-26 2010-06-09 天津市电视技术研究所 铁路道口机车检测图像预警系统
JP5763965B2 (ja) * 2011-05-11 2015-08-12 キヤノン株式会社 情報処理装置、情報処理方法、及びプログラム
CN103200357A (zh) * 2012-01-04 2013-07-10 苏州科泽数字技术有限公司 全景凝视网络摄像机构建方法与装置
US9020189B2 (en) * 2012-12-07 2015-04-28 The Nielsen Company (Us), Llc Methods and apparatus to monitor environments
KR102130316B1 (ko) * 2013-09-30 2020-07-08 한국전자통신연구원 영상 인식 장치 및 방법
US9514366B2 (en) * 2014-02-03 2016-12-06 Xerox Corporation Vehicle detection method and system including irrelevant window elimination and/or window score degradation
CN104021570B (zh) * 2014-06-04 2016-11-09 电子科技大学 一种视频中有方向的越界和拌线检测方法
JP6482196B2 (ja) 2014-07-09 2019-03-13 キヤノン株式会社 画像処理装置、その制御方法、プログラム、及び記憶媒体
WO2016025485A2 (en) * 2014-08-14 2016-02-18 Qualcomm Incorporated Detection of action frames of a video stream
CN104754302B (zh) * 2015-03-20 2017-08-08 安徽大学 一种基于枪球联动系统的目标检测跟踪方法
JP6364372B2 (ja) * 2015-03-24 2018-07-25 トヨタホーム株式会社 地域監視システム
US10095933B2 (en) * 2016-12-05 2018-10-09 Google Llc Systems and methods for locating image data for selected regions of interest

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1722825A (zh) * 2004-07-16 2006-01-18 赖金鼎 设定影像监控区的装置及其方法
JP2014036414A (ja) * 2012-08-10 2014-02-24 Hitachi Kokusai Electric Inc 監視システム
CN103327306A (zh) * 2013-06-14 2013-09-25 广东威创视讯科技股份有限公司 视频监控存储方法及装置
CN103686074A (zh) * 2013-11-20 2014-03-26 南京熊猫电子股份有限公司 视频监控中移动目标的定位方法
CN103646257A (zh) * 2013-12-30 2014-03-19 中国科学院自动化研究所 一种基于视频监控图像的行人检测和计数方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3585052A4 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112468843A (zh) * 2020-10-26 2021-03-09 国家广播电视总局广播电视规划院 视频去重方法及装置
CN116030388A (zh) * 2022-12-30 2023-04-28 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质
CN116030388B (zh) * 2022-12-30 2023-08-11 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质

Also Published As

Publication number Publication date
EP3585052A4 (en) 2020-12-30
CN108289191A (zh) 2018-07-17
US20190354768A1 (en) 2019-11-21
EP3585052B1 (en) 2024-04-03
CN108289191B (zh) 2020-01-10
MA47512A (fr) 2019-12-25
US10936876B2 (en) 2021-03-02
EP3585052A1 (en) 2019-12-25

Similar Documents

Publication Publication Date Title
WO2018149322A1 (zh) 图像识别方法、装置、设备及存储介质
WO2018219180A1 (zh) 确定人脸图像质量的方法和装置、电子设备和计算机存储介质
US9277165B2 (en) Video surveillance system and method using IP-based networks
WO2020094091A1 (zh) 一种图像抓拍方法、监控相机及监控系统
US11024052B2 (en) Stereo camera and height acquisition method thereof and height acquisition system
JP2014157452A (ja) 画像処理装置、画像処理方法、および画像処理プログラム
US20200013180A1 (en) Height calculation system, information processing apparatus, and non-transitory computer readable medium storing program
CN109960969B (zh) 移动路线生成的方法、装置及系统
CN111597879A (zh) 一种基于监控视频的姿态检测方法、装置及系统
CN113228626A (zh) 视频监控系统和方法
JP7103229B2 (ja) 不審度推定モデル生成装置
US20200084416A1 (en) Information processing apparatus, control method, and program
US10965858B2 (en) Image processing apparatus, control method thereof, and non-transitory computer-readable storage medium for detecting moving object in captured image
CN110505438B (zh) 一种排队数据的获取方法和摄像机
CN105718033A (zh) 疲劳检测系统及方法
US11132778B2 (en) Image analysis apparatus, image analysis method, and recording medium
US10783365B2 (en) Image processing device and image processing system
JP2012124658A (ja) 特定人物検知システムおよび検知方法
WO2023231479A1 (zh) 瞳孔检测方法、装置、存储介质及电子设备
CN109190495B (zh) 性别识别方法、装置及电子设备
CN113947795B (zh) 口罩佩戴检测方法、装置、设备及存储介质
WO2022057329A1 (zh) 安全监控方法、装置、系统和存储介质
JP6632632B2 (ja) 監視システム
US20220319232A1 (en) Apparatus and method for providing missing child search service based on face recognition using deep-learning
CN108551493A (zh) 基于移动云计算的智慧城市环境监测装置及方法

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: 18754005

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018754005

Country of ref document: EP

Effective date: 20190916