WO2019080881A1 - 行人流量漏斗生成方法及装置、存储介质、电子设备 - Google Patents

行人流量漏斗生成方法及装置、存储介质、电子设备

Info

Publication number
WO2019080881A1
WO2019080881A1 PCT/CN2018/111713 CN2018111713W WO2019080881A1 WO 2019080881 A1 WO2019080881 A1 WO 2019080881A1 CN 2018111713 W CN2018111713 W CN 2018111713W WO 2019080881 A1 WO2019080881 A1 WO 2019080881A1
Authority
WO
WIPO (PCT)
Prior art keywords
head
shoulder
tracking
frame image
current frame
Prior art date
Application number
PCT/CN2018/111713
Other languages
English (en)
French (fr)
Inventor
车广富
陈宇
安山
刘强
翁志
Original Assignee
北京京东尚科信息技术有限公司
北京京东世纪贸易有限公司
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 北京京东尚科信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京京东尚科信息技术有限公司
Priority to JP2020520032A priority Critical patent/JP7270617B2/ja
Priority to US16/758,958 priority patent/US11210795B2/en
Publication of WO2019080881A1 publication Critical patent/WO2019080881A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06M—COUNTING MECHANISMS; COUNTING OF OBJECTS NOT OTHERWISE PROVIDED FOR
    • G06M11/00—Counting of objects distributed at random, e.g. on a surface
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
    • 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
    • G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • 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
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30242—Counting objects in image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking

Definitions

  • the present disclosure relates to the field of data processing technologies, and in particular, to a pedestrian flow funnel generating method and apparatus, a storage medium, and an electronic device.
  • the application range of video surveillance systems is becoming wider and wider.
  • Surveillance cameras are often installed at the entrances and exits of supermarkets, shopping malls, stadiums, and airport stations, so that security personnel and managers can monitor the entrances and exits of these places.
  • the flow of people entering and leaving the supermarkets, shopping malls, stadiums, and airport stations is of great significance to the operators or managers of the above-mentioned places.
  • the flow of people refers to the number of people moving in a certain direction. Refers to the number of people moving in both directions of entry/exit.
  • the traffic statistics are mainly realized by manual enumeration by monitoring personnel, but there is a problem that the accuracy is difficult to guarantee, and the labor cost is high.
  • a pedestrian flow funnel generating method including:
  • a motion trajectory of each head and shoulder region in the tracking sequence set is analyzed to count the pedestrians, and when the current frame image is the last frame image, a pedestrian flow funnel is generated based on the counting result of the pedestrian.
  • the method before the acquiring the current frame image, the method further includes:
  • the method further includes:
  • the age range and gender of each head and shoulder region in the set of tracking sequences are identified according to a gender age recognition model.
  • the generating a pedestrian flow funnel based on the counting result of the pedestrian includes:
  • the pedestrian flow funnel is generated based on the counting result of the pedestrian and in combination with the age and gender of each head and shoulder region in the tracking sequence set.
  • the tracking and updating the head and shoulder regions in the tracking sequence set in the current frame image according to the multi-target tracking algorithm includes:
  • the tracking and updating, respectively, the head and shoulder regions in the set of tracking sequences in the current frame image according to a kernel correlation filter tracking algorithm includes:
  • the method further includes: calculating a tracker corresponding to each of the head and shoulder regions, including:
  • a regression model is trained according to the first training sample set corresponding to each of the head and shoulder regions to obtain a tracker corresponding to each of the head and shoulder regions.
  • the method further includes generating the head and shoulder recognition model according to a convolutional neural network, including:
  • the updating the tracking sequence set according to a head and shoulder region in the current frame image includes:
  • the set of tracking sequences is updated according to the similarity.
  • the calculating the similarity between each of the head and shoulder regions in the current frame image and each of the head and shoulder regions in the tracking sequence set includes:
  • sim(Q i , Q j ) is the similarity between the i-th head and shoulder region Q i in the current frame image and the j-th head and shoulder region Q j in the tracking target set
  • Q i area is The area of the i-th head and shoulder region Q i in the current frame image
  • Q j area is the area of the j-th head and shoulder region Q j in the set of tracking sequences.
  • the updating the set of tracking sequences according to the similarity includes:
  • the replacing the head-shoulder region in the tracking sequence set corresponding to the successfully matched head-shoulder region in the current frame image includes:
  • the header and shoulder regions corresponding to the successful matching in the current frame image are used. Replace the head and shoulder regions in the set of tracking sequences.
  • the calculation formula of the confidence is:
  • conf(obj) is the confidence of obj
  • obj is the head and shoulder area in the current frame image or the head and shoulder area in the tracking sequence set
  • area(obj) is the area of obj
  • Score(obj) is The head-to-shoulder recognition model calculates the class attribution probability of obj, where ⁇ is 0.5 and B is 1000.
  • the method before the analyzing the motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrians, the method further includes:
  • the first preset frame number when there is a head-shoulder region in the tracking state that is not updated from the head-shoulder region acquired in each frame image in the image of the first preset frame number, the first preset frame number will be The head and shoulder regions of the head and shoulder regions that are not acquired from each frame of the image are changed from the tracking state to the abnormal state;
  • the tracking sequence set when there is an abnormal state of the head and shoulder region that is not updated from the head and shoulder regions acquired in each frame image in the image of the second preset frame number, the tracking sequence set is deleted in the tracking sequence set.
  • the image of the second predetermined number of frames is not in the head and shoulder region of the abnormal state that is not updated from the head and shoulder regions acquired from each frame image.
  • the analyzing the motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrian includes:
  • Pedestrians are counted based on the trajectory of each head and shoulder region combined with a virtual count line.
  • the method further includes: constructing the gender age recognition model, including:
  • the gender age network is trained using the third training sample set to obtain the gender age recognition model, wherein the gender age network includes 3 convolutional layers and 3 fully connected layers.
  • a pedestrian flow funnel generating apparatus including:
  • a tracking update module configured to acquire a current frame image, and track and update a head and shoulder region in the tracking sequence set in the current frame image according to a multi-target tracking algorithm
  • a detection update module configured to acquire a head and shoulder region in the current frame image according to a head and shoulder recognition model, and update the tracking sequence set according to a head and shoulder region in the current frame image;
  • a technique generating module configured to analyze a motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrian, and generate, based on the counting result of the pedestrian when the current frame image is the last frame image A pedestrian flow funnel.
  • a computer readable storage medium having stored thereon a computer program, the computer program being executed by a processor to implement the pedestrian traffic funnel generating method of any of the above.
  • an electronic device including:
  • a memory for storing executable instructions of the processor
  • the processor is configured to perform the pedestrian traffic funnel generation method according to any one of the above, by executing the executable instruction.
  • the pedestrian flow funnel generating method and apparatus, storage medium, and electronic device provided by an exemplary embodiment of the present disclosure. First acquiring the current frame image, tracking and updating the head and shoulder regions in the tracking sequence set in the current frame image according to the multi-target tracking algorithm, and then acquiring the head and shoulder regions in the current frame image according to the head and shoulder recognition model, And updating the tracking sequence set according to the head and shoulder regions in the current frame image, and finally analyzing motion trajectories of each head and shoulder region in the tracking sequence set to count the pedestrians, and in the current frame When the image is the last frame image, a pedestrian flow funnel is generated based on the counting result of the pedestrian.
  • the multi-target tracking algorithm combined with the head and shoulder recognition model can avoid the missed detection of the head and shoulder area, improve the accuracy of detecting the head and shoulder area, and thus improve the accuracy of pedestrian counting, thereby improving the accuracy of the pedestrian flow funnel.
  • the head and shoulder recognition model can quickly acquire the head and shoulder area and the cost is low, thereby increasing the counting speed of the head and shoulder area and reducing the counting cost, thereby increasing the speed of generating the pedestrian flow funnel and reducing the speed. The cost of generating a pedestrian traffic funnel.
  • FIG. 1 is a flow chart of a method for generating a pedestrian traffic funnel according to the present disclosure
  • FIG. 2 is a schematic diagram of a mask layout provided by an exemplary embodiment of the present disclosure
  • FIG. 3 is a block diagram of a method for generating a pedestrian traffic funnel according to the present disclosure
  • FIG. 4 is a schematic block diagram of an electronic device according to an exemplary embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram of a program product in an exemplary embodiment of the present disclosure.
  • the traffic statistics are mainly realized by manual enumeration by monitoring personnel.
  • This method of manually counting human traffic is relatively reliable in the case of short monitoring time and sparse traffic.
  • the statistical accuracy will be greatly reduced ( That is, missed detection), and the method of manual statistics requires a lot of labor costs.
  • you want to count the gender and age of people traffic it will further reduce the efficiency and accuracy of statistics, and increase the cost.
  • the accuracy of the flow funnel based on inaccurate statistics on human traffic and gender age is not high, making it impossible for statisticians to get through the traffic funnel. Analyze data more accurately.
  • a pedestrian traffic funnel generating method is first disclosed.
  • the pedestrian traffic funnel generating method may include the following steps:
  • Step S1 acquiring a current frame image, and tracking and updating a head and shoulder region in the tracking sequence set in the current frame image according to a multi-target tracking algorithm
  • Step S2 acquiring a head and shoulder area in the current frame image according to a head and shoulder recognition model, and updating the tracking sequence set according to a head and shoulder area in the current frame image;
  • Step S3 analyzing a motion track of each head and shoulder region in the tracking sequence set to count the pedestrian, and generating a pedestrian flow based on the counting result of the pedestrian when the current frame image is the last frame image.
  • the multi-target tracking algorithm combined with the head and shoulder recognition model can avoid the missed detection of the head and shoulder regions, improve the accuracy of detecting the head and shoulder regions, and thereby improve the pedestrians.
  • the accuracy of the counting increases the accuracy of the pedestrian flow funnel; on the other hand, the head and shoulder recognition model can quickly acquire the head and shoulder area and the cost is low, thereby increasing the counting speed of the head and shoulder area and reducing the counting cost. This in turn increases the speed at which the pedestrian traffic funnel is generated and reduces the cost of generating a pedestrian traffic funnel.
  • Step S1 Acquire a current frame image, and track and update the head and shoulder regions in the tracking sequence set in the current frame image according to the multi-target tracking algorithm.
  • the method before the acquiring the current frame image, the method further includes: acquiring a first frame image, and acquiring a head and shoulder region in the first frame image according to the head and shoulder recognition model, and passing the The head and shoulder area initializes the set of tracking sequences.
  • the video material transmitted by the monitoring device can be received in real time, and the first frame image is intercepted in the video material.
  • an image of the mall business start time may be acquired as the first frame image, for example, the image is acquired from eight o'clock.
  • the head and shoulder regions in the first frame image are identified according to the head and shoulder recognition model; after the head and shoulder regions are recognized, the tracking sequence set is initialized according to the identified head and shoulder regions, that is, the identification is to be recognized The head and shoulder areas are added to the tracking sequence set. It should be noted that, when the head and shoulder regions in the first frame image are identified by the head and shoulder recognition model, the score value of each head and shoulder region, that is, the category attribution probability of each head and shoulder region, can also be calculated by the head and shoulder recognition model.
  • the head and shoulder recognition model may be generated according to a convolutional neural network.
  • the process of generating the head and shoulder recognition model according to the convolutional neural network includes the following steps:
  • Step S111 training the MobileNet network according to the ImageNet classification data set to obtain the weight value of the MobileNet network.
  • the ImageNet is currently the largest database for image recognition in the world.
  • the MobileNet network is a streamlined architecture that uses a deeply separable convolutional lightweight deep neural network that can resolve standard volume integration into a deep convolution and a point convolution, ie each volume
  • the core is applied to each channel, and the 1x1 convolution is used to combine the output of the channel convolution. It has been proved that this decomposition can effectively reduce the amount of calculation, reduce the size of the model, and the network performance is good and the speed is also very fast.
  • step S11 an image of each marked over-shoulder region in the ImageNet classification data set can be input into the MobileNet network to obtain a weight value of the MobileNet network.
  • Step S112 adding a predetermined number of layers of the volume base layer on the MobileNet network to obtain a head and shoulder detection network, wherein the size of the volume base layer is decremented layer by layer.
  • the preset number of layers may be set according to specific requirements, for example, may be 3 layers or 6 layers, which is not specifically limited in this exemplary embodiment.
  • the size of the roll base layer is decremented layer by layer to complete the prediction of the multi-dimensional head and shoulder area by using the roll base layer which is reduced in size by layer.
  • Step S113 acquiring a second training sample set that marks the head and shoulder area, and training the head and shoulder detection network that retains the weight value of the MobileNet network according to the second training sample set to obtain a Head and shoulder recognition model.
  • the image acquisition module can acquire different angles and different backgrounds (twig occlusion, construction). Head and shoulder images in the background of the object occlusion, and the head and shoulder regions in each head and shoulder image are marked to obtain a second set of training samples that mark the head and shoulder regions.
  • the head and shoulder detection network retaining the weight value of the MobileNet network is trained based on the second training sample set and using a mobility learning method to obtain a head and shoulder recognition model.
  • the image of the head and shoulder area marked in the second training sample set is input to the head and shoulder detection network to obtain the weight value in the head and shoulder detection network, and the loss of the head and shoulder detection network is basically not stable, and the average value is At 0.8, the head and shoulder detection network is the head and shoulder recognition model.
  • the tracking sequence set in step S1 is an updated tracking sequence set in the previous frame graphic of the current frame image. It should be noted that when the previous frame image of the current frame image is the first frame image, the tracking sequence set at this time is the initial or subsequent tracking sequence set in the first frame image.
  • the tracking and updating the head and shoulder regions in the tracking sequence set in the current frame image according to the multi-target tracking algorithm may include: respectively, in the current frame image according to a kernel correlation filtering tracking algorithm
  • the head and shoulder regions in the set of tracking sequences are tracked and updated.
  • the tracking and updating the head and shoulder regions in the tracking sequence set in the current frame image according to the kernel correlation filter tracking algorithm may include:
  • Step S121 Acquire, in the current frame image, a candidate tracking target corresponding to each head and shoulder region in the tracking sequence set, respectively, based on a position of each of the head and shoulder regions in the tracking sequence set in the previous frame image.
  • the candidate tracking targets corresponding to the head and shoulder regions in the tracking sequence set may be sequentially acquired, wherein the candidate tracking targets corresponding to each of the head and shoulder regions may include multiple.
  • the candidate tracking target corresponding to the first head and shoulder region in the tracking sequence set is taken as an example, and the position of the first head and shoulder region in the previous frame image is centered in the current frame image.
  • a plurality of candidate tracking targets are obtained, and the plurality of candidate tracking targets are candidate tracking targets corresponding to the first head and shoulder regions, and the preset ranges may be set by a developer.
  • the obtaining principle is the same as the candidate tracking target corresponding to the first head and shoulder region, and only the acquiring position changes, so it will not be described here.
  • Step 122 Calculate a response value of a candidate tracking target corresponding to each of the head and shoulder regions according to a tracker corresponding to each of the head and shoulder regions.
  • the calculating the tracker corresponding to each of the head and shoulder regions may include: acquiring, in the previous frame image, based on positions of the head and shoulder regions in the previous frame image in the tracking sequence set respectively a first training sample set corresponding to each head and shoulder region in the tracking sequence set; training a regression model according to the first training sample set corresponding to each of the head and shoulder regions, respectively, to obtain a corresponding to each of the head and shoulder regions Tracker.
  • a description is made by taking an example of calculating a tracker corresponding to the first head and shoulder area in the tracking sequence set.
  • a plurality of first training samples are acquired in a previous frame image centering on a position of the first head and shoulder region in the previous frame image to form a first training sample set corresponding to the first head and shoulder region, and the first training is performed. All the first training samples in the sample set are input into the regression model to train the regression model, and the trained regression model is determined as the tracker corresponding to the first head and shoulder region.
  • the process of calculating the tracker corresponding to the other head and shoulder regions in the tracking sequence set is the same as the process of calculating the tracker corresponding to the first head and shoulder region, and only the position of the first training sample set is different, so no further details are provided here.
  • the first training sample set corresponding to the head and shoulder area is acquired in the previous frame image based on the position of the head and shoulder area in the previous frame image when calculating the tracker corresponding to the head and shoulder area
  • the positions of the same head and shoulder regions are different, and therefore, the trackers corresponding to the same head and shoulder regions in different frame graphics are different.
  • the response value calculated by the corresponding tracker of each frame is more accurate, so that the tracking determination of the tracking target is more accurate.
  • the response value of the candidate tracking target corresponding to the first head and shoulder region is calculated according to the tracker corresponding to the first head and shoulder region, and each candidate tracking target corresponding to the first head and shoulder region is substituted and The tracker corresponding to the first head and shoulder area is used to obtain the response value of each candidate target corresponding to the first head and shoulder area.
  • Calculating the response value of the candidate tracking target corresponding to the other head and shoulder regions in the tracking sequence set is the same as the above-mentioned process of calculating the response value of the candidate tracking target corresponding to the first head and shoulder region, and only changing the tracker, that is, the tracker and the head The shoulder area corresponds, so it will not be described here.
  • Step 123 Determine a candidate tracking target with the largest response value among the candidate tracking targets corresponding to each of the head and shoulder regions as a tracking target of the corresponding head and shoulder region in the current frame image.
  • the first head and shoulder regions in the tracking sequence set are described, and the response values of the candidate tracking targets corresponding to the first head and shoulder regions are respectively compared, and the candidate tracking target with the largest response value is selected.
  • the tracking target of the first head and shoulder area in the current frame image is determined, that is, the candidate tracking target is the first head and shoulder area in the current frame image. It should be noted that the determination of the tracking target in the current frame image of the other head and shoulder regions in the tracking sequence set is the same as the tracking target of the first head and shoulder region in the current frame image, and therefore, no description is made herein. .
  • Step 124 Update each head and shoulder region in the tracking sequence set according to a tracking target of each of the head and shoulder regions in the current frame image.
  • each of the head and shoulder regions in the set of replacement tracking sequences corresponding to the tracking targets corresponding to the head and shoulder regions in the tracking sequence set in the current frame image is used to update each head and shoulder in the tracking sequence set. region.
  • the head and shoulder recognition model may not be able to detect the head and shoulder region in the background image of the obstruction, based on
  • the head and shoulder regions in the current frame of the pair of head and shoulder recognition models are detected to track the head and shoulder regions in the tracking sequence set
  • the head and shoulder regions in the tracking sequence may not be detected, thereby causing the head and shoulder regions. Missed inspection.
  • all the head and shoulder regions in the tracking sequence set can be tracked by the multi-target tracking algorithm, thereby avoiding the missed detection of the head and shoulder regions, improving the accuracy of detecting the head and shoulder regions, thereby improving the accuracy.
  • the accuracy of the pedestrian count increases the accuracy of the pedestrian flow funnel.
  • step S2 a head and shoulder region in the current frame image is acquired according to a head and shoulder recognition model, and the tracking sequence set is updated according to a head and shoulder region in the current frame image.
  • the current frame image is input into the head and shoulder recognition model to obtain a head and shoulder region in the current frame image, and the tracking sequence set is updated according to the head and shoulder regions in the current frame image. , the new head and shoulder area is added to the tracking sequence set, and the original head and shoulder area in the tracking sequence is updated.
  • the head and shoulder recognition model the head and shoulder area can be quickly obtained and the cost is low, thereby increasing the counting speed of the head and shoulder area and reducing the counting cost, thereby increasing the speed of generating the pedestrian flow funnel and reducing the cost of generating the pedestrian flow funnel.
  • the updating the tracking sequence set according to the head and shoulder area in the current frame image may include: calculating a similarity between each head and shoulder area in the current frame image and each head and shoulder area in the tracking sequence set Degree; updating the set of tracking sequences according to the similarity.
  • Calculating the similarity between each head and shoulder area in the current frame image and each head and shoulder area in the tracking sequence set may include: calculating each head and shoulder area in the current frame image according to the following formula: Tracking the similarity of each head and shoulder area in the sequence set:
  • sim(Q i , Q j ) is the similarity between the i-th head and shoulder region Q i in the current frame image and the j-th head and shoulder region Q j in the tracking target set
  • Q i area is The area of the i-th head and shoulder region Q i in the current frame image
  • Q j area is the area of the j-th head and shoulder region Q j in the set of tracking sequences.
  • the similarity between the first head and shoulder regions in the current frame image and each of the head and shoulder regions in the tracking sequence set is calculated as an example, and the area of the first head and shoulder region in the current frame image and the tracking sequence set are obtained.
  • the area of each head and shoulder region is combined with the above formula to calculate the similarity between the first head and shoulder regions in the current frame image and the head and shoulder regions in the tracking sequence set, respectively. It should be noted that the above steps may be repeated to calculate the similarity between the other head and shoulder regions in the current frame image and the head and shoulder regions in the tracking sequence set.
  • the updating the set of tracking sequences according to the similarity may include:
  • Step S21 Match each head and shoulder area in the current frame image with each head and shoulder area in the tracking sequence set according to the similarity and a similarity threshold.
  • the similarity threshold may be 0.5, but the present exemplary embodiment is not particularly limited thereto.
  • the following describes the matching process of the first head and shoulder regions in the current frame image, first obtaining the maximum similarity between the first head and shoulder regions and the head and shoulder regions in the tracking sequence set, and determining whether the maximum similarity is greater than the similarity.
  • the threshold value is determined to be successful when the maximum similarity is greater than the similarity threshold, that is, the first head and shoulder area and the head and shoulder area in the corresponding tracking sequence set are the same head and shoulder area.
  • the matching is considered to be a failure, that is, the head and shoulder area is a new head and shoulder area.
  • Step S22 If the matching is successful, replace the head and shoulder regions in the tracking sequence set corresponding to the matching head and shoulder regions in the current frame image.
  • the first head and shoulder region in the current frame image is taken as an example.
  • the maximum similarity between the head and shoulder regions in the first head and shoulder region and the tracking sequence set is greater than the similarity threshold, The head and shoulder regions in the set of tracking sequences are replaced by the first head and shoulder regions.
  • the replacing the head-shoulder region in the tracking sequence set corresponding to the successful matching head-shoulder region in the current frame image may include separately calculating the current frame image. Matching a confidence of a successful head and shoulder region and a corresponding head and shoulder region in the set of tracking sequences; a confidence level of a successful header and shoulder region in the current frame image is greater than a corresponding one of the corresponding tracking sequence sets In the confidence of the head and shoulder regions, the head and shoulder regions in the set of tracking sequences are replaced with corresponding matching head and shoulder regions in the current frame image.
  • the calculation formula of the confidence is:
  • conf(obj) is the confidence of obj
  • obj is the head and shoulder area in the current frame image or the head and shoulder area in the tracking sequence set
  • area(obj) is the area of obj
  • Score(obj) is The head-to-shoulder recognition model calculates the class attribution probability of obj, where ⁇ is 0.5 and B is 1000.
  • the above process is described by taking the first head and shoulder region in the current frame image and the third head and shoulder region in the tracking sequence set successfully as an example, and calculating the first head and shoulder region in the current frame image by using the above confidence formula. Confidence and the confidence of the third head and shoulder region in the set of tracking sequences; the current frame is used when the confidence of the first head and shoulder region in the current frame image is greater than the confidence of the third head and shoulder region in the tracking sequence set The first head and shoulder area in the image replaces the third head and shoulder area in the tracking sequence set.
  • Step S23 if the matching fails, adding a header and shoulder region of the matching failure in the current frame image to the tracking sequence set.
  • the first head and shoulder area in the current frame image is taken as an example.
  • the maximum similarity between the head and shoulder areas in the first head and shoulder area and the tracking sequence set is not greater than the similarity threshold.
  • the first head and shoulder area is added to the tracking sequence set to become a new tracking target.
  • the shoulder and shoulder regions in the current frame image are acquired by the shoulder recognition model, and the similarity between each head and shoulder region in the current frame image and each of the head and shoulder regions in the tracking sequence set is calculated, and according to The similarity updates the tracking sequence set, and the detected new head and shoulder area is added to the tracking sequence set, and the original head and shoulder area in the tracking sequence set is updated, thereby improving the head and shoulder area detection.
  • Accuracy which improves the accuracy of the count, avoids the missed detection of the tracking target, and thus improves the accuracy of the pedestrian flow funnel.
  • step S3 analyzing a motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrian, and generating a row based on the counting result of the pedestrian when the current frame image is the last frame image. Human flow funnel.
  • the motion trajectory of each head and shoulder region in the tracking sequence set is analyzed before the pedestrian is counted.
  • Step S31 in the tracking sequence set, when there is a head-shoulder region in the tracking state that is not updated from the head-shoulder region acquired in each frame image in the image of the first preset frame number, the first pre-pre- In the image in which the number of frames is not updated, the head and shoulder regions are not changed from the tracking state to the abnormal state.
  • the image of the first preset number of frames may be a continuous 5 frame image or a continuous 10 frame image, which is not specifically limited in this exemplary embodiment.
  • Step S31 is described by taking a continuous five-frame image as an example, and the head and shoulder regions corresponding to the five frames of images are respectively acquired according to the head and shoulder recognition model, and none of the head and shoulder regions corresponding to the five frames of images are in the tracking sequence set.
  • One of the head and shoulder areas in the head and shoulder area of the tracking state is successfully matched, that is, the head and shoulder area is determined to be an abnormal head and shoulder area, and the head and shoulder area is changed from the tracking state to the abnormal state.
  • Step S32 when there is a head-shoulder region in an abnormal state that is not successfully matched in the image of the second preset frame number in the tracking sequence set, deleting the second preset frame in the tracking sequence set.
  • the number of images matches the successful head and shoulder areas in an abnormal state.
  • the image of the second predetermined number of frames may be a continuous 5 frame image or a continuous 10 frame image, which is not specifically limited in this exemplary embodiment.
  • Step S32 is described by taking a continuous five-frame image as an example, and the head and shoulder regions corresponding to the five frames of images are respectively acquired according to the head and shoulder recognition model, and none of the head and shoulder regions corresponding to the five frames of images are in the tracking sequence set.
  • One of the head and shoulder regions in an abnormal state matches the successful head and shoulder region, that is, the head and shoulder region is identified as the misdetected head and shoulder region, and the head and shoulder region is deleted from the tracking sequence set.
  • the analyzing the motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrians may include: analyzing motion trajectories of each head and shoulder region in the tracking sequence set; The trajectory of the area is combined with a virtual count line to count pedestrians.
  • the motion trajectory of each head and shoulder region may be generated based on the positions in different frame images of the head and shoulder regions in the tracking sequence set.
  • a virtual count line may be set and counted by pedestrians in different directions by judging whether the motion trajectory of the head and shoulder area crosses the virtual count line and the direction across the virtual count line.
  • the process of counting pedestrians in different directions by determining whether the motion track of the head and shoulder region crosses the virtual count line and the direction crossing the virtual count line includes: mask mask MASK can be used to represent the area on both sides of the virtual count line (eg, 2), such as the first area and the second area, on the basis of which, according to the pixel values of the motion trajectories of the head and shoulder regions, it is determined whether each of the head and shoulder regions crosses from the first region to the second region; or Whether each head and shoulder area spans from the second area to the first area.
  • mask mask MASK can be used to represent the area on both sides of the virtual count line (eg, 2), such as the first area and the second area, on the basis of which, according to the pixel values of the motion trajectories of the head and shoulder regions, it is determined whether each of the head and shoulder regions crosses from the first region to the second region; or Whether each head and shoulder area spans from the second area to the first area.
  • the first shoulder area crosses the first area to the second area
  • the number of people in the out cell is increased by one
  • the shoulder area is from the second area.
  • crossing the first area the number of people entering the cell is increased by one.
  • the head and shoulder regions that have been counted may be marked.
  • the pedestrians are counted according to the motion trajectory of each of the head and shoulder regions and a virtual count line, and the counting method is simple, easy to implement, and high in counting accuracy.
  • step S2 After counting the pedestrian according to the motion trajectory of each head and shoulder region, it is judged whether the current frame image is the last frame image, and when the current frame image is not the last frame image, step S2 is repeated, and step S3 is performed to continue to the pedestrian.
  • Count A pedestrian flow funnel is generated based on the counting result of the pedestrian when the current frame image is the last frame image. For example, you can analyze the flow of people at different time periods to generate a pedestrian flow funnel, so that staff can analyze business process problems based on the pedestrian flow funnel to control macro data.
  • the method further includes: a gender age recognition model identifying age ranges and genders of each of the head and shoulder regions in the set of tracking sequences; and generating based on the counting results of the pedestrians and combining the age segments and genders of the head and shoulder regions in the tracking sequence set The pedestrian flow funnel.
  • constructing the gender age recognition model may include: obtaining a mark gender and an age group in the LFW data set and the social networking site. And a third training sample set; the gender age network is trained by using the third training sample set to obtain the gender age recognition model, wherein the gender age network comprises three convolutional layers and three fully connected layers.
  • a gender age network including 3 volume base layers and 3 full connection layers is established; a third training sample set is acquired in the LFW data set social platform, and each of the third training sample sets is The face of the third training sample is cropped at the center. For example, a third training sample having a size of 256*256 is cropped to a third training sample having a size of 227*227, and the cropped
  • the training sample identifies the gender and age group.
  • the gender is identified as male or female, and the age is identified according to the age group.
  • the age group can be divided into 8 stages for identification, respectively 0 to 3, 4 ⁇ 7, 8 to 14, 15 to 24, 25 to 37, 38 to 47, 48 to 59, 60+.
  • the division of the age group is not limited to this, the developer can set according to the demand; the positive too Gaussian distribution with the standard deviation of 0.01 and the mean value of 0 is used to initialize the weight value in the gender-age network; The method and the collection dropout train the initialized gender age network to obtain a gender age recognition model.
  • the dropout ratio is 0.5.
  • the age segment and gender of each head and shoulder region in each tracking sequence set are identified based on the gender age recognition model. It should be noted that only the age segment of the head and shoulder region in the tracking sequence set is added for the first time. Gender is identified. In addition, since the same gender age recognition model is used for age segment identification and gender recognition, the gender age recognition model output is 2 when the age is recognized, and the gender age recognition model output is 8 when the gender is recognized.
  • the pedestrian flow funnel is generated based on the counting result of the pedestrian and the age and gender of each head and shoulder region in the tracking sequence set.
  • the pedestrian flow funnel can be analyzed by analyzing the flow rate and gender ratio of different time periods and the age ratio, so that the staff can analyze the macro-data from multiple dimension business process problems according to the pedestrian flow funnel.
  • a pedestrian flow funnel generating device is further provided.
  • the pedestrian traffic funnel generating device 100 may include a tracking update module 101, a detection update module 102, and a technology generation module 103. :
  • the tracking update module 101 may be configured to acquire a current frame image, and track and update the head and shoulder regions in the tracking sequence set in the current frame image according to the multi-target tracking algorithm;
  • the detection update module 102 may be configured to acquire a head and shoulder area in the current frame image according to a head and shoulder recognition model, and update the tracking sequence set according to a head and shoulder area in the current frame image;
  • the technology generating module 103 may be configured to analyze a motion trajectory of each head and shoulder region in the tracking sequence set to count the pedestrian, and when the current frame image is the last frame image, based on the counting result of the pedestrian Generate a pedestrian flow funnel.
  • each pedestrian flow funnel generating device module in the above has been described in detail in the corresponding pedestrian traffic funnel generating method, and therefore will not be described herein.
  • an electronic device capable of implementing the above method is also provided.
  • FIG. 4 An electronic device 600 according to such an embodiment of the present disclosure is described below with reference to FIG. 4 is merely an example and should not impose any limitation on the function and scope of use of the embodiments of the present disclosure.
  • electronic device 600 is embodied in the form of a general purpose computing device.
  • the components of the electronic device 600 may include, but are not limited to, the at least one processing unit 610, the at least one storage unit 620, the bus 630 connecting the different system components (including the storage unit 620 and the processing unit 610), and the display unit 640.
  • the storage unit stores program code, which can be executed by the processing unit 610, such that the processing unit 610 performs various exemplary embodiments according to the present disclosure described in the "Exemplary Method" section of the present specification.
  • the processing unit 610 may perform step S1 shown in FIG.
  • Step S2 acquiring a head and shoulder area in the current frame image according to the head and shoulder recognition model, and updating the tracking sequence set according to the head and shoulder area in the current frame image;
  • Step S3 analyzing the tracking sequence set The trajectory of each of the head and shoulder regions is used to count the pedestrians, and when the current frame image is the last frame image, a pedestrian flow funnel is generated based on the counting result of the pedestrian.
  • the storage unit 620 can include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 6201 and/or a cache storage unit 6202, and can further include a read only storage unit (ROM) 6203.
  • RAM random access storage unit
  • ROM read only storage unit
  • the storage unit 620 can also include a program/utility 6204 having a set (at least one) of the program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more applications, other program modules, and program data, Implementations of the network environment may be included in each or some of these examples.
  • Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local area using any of a variety of bus structures. bus.
  • the electronic device 600 can also communicate with one or more external devices 700 (eg, a keyboard, pointing device, Bluetooth device, etc.), and can also communicate with one or more devices that enable the user to interact with the electronic device 600, and/or with The electronic device 600 is enabled to communicate with any device (e.g., router, modem, etc.) that is in communication with one or more other computing devices. This communication can take place via an input/output (I/O) interface 650. Also, electronic device 600 can communicate with one or more networks (eg, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through network adapter 660. As shown, network adapter 660 communicates with other modules of electronic device 600 via bus 630.
  • network adapter 660 communicates with other modules of electronic device 600 via bus 630.
  • the technical solution according to an embodiment of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network.
  • a non-volatile storage medium which may be a CD-ROM, a USB flash drive, a mobile hard disk, etc.
  • a number of instructions are included to cause a computing device (which may be a personal computer, server, terminal device, or network device, etc.) to perform a method in accordance with an embodiment of the present disclosure.
  • a computer readable storage medium having stored thereon a program product capable of implementing the above method of the present specification.
  • various aspects of the present disclosure may also be embodied in the form of a program product comprising program code for causing said program product to run on a terminal device The terminal device performs the steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Method" section of the present specification.
  • a program product 800 for implementing the above method which may employ a portable compact disk read only memory (CD-ROM) and includes program code, and may be at a terminal device, is illustrated in accordance with an embodiment of the present disclosure.
  • CD-ROM portable compact disk read only memory
  • the program product of the present disclosure is not limited thereto, and in this document, the readable storage medium may be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.
  • the program product can employ any combination of one or more readable media.
  • the readable medium can be a readable signal medium or a readable storage medium.
  • the readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive lists) of readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.
  • the computer readable signal medium may include a data signal that is propagated in the baseband or as part of a carrier, carrying readable program code. Such propagated data signals can take a variety of forms including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the readable signal medium can also be any readable medium other than a readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
  • Program code embodied on a readable medium can be transmitted using any suitable medium, including but not limited to wireless, wireline, optical cable, RF, etc., or any suitable combination of the foregoing.
  • Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language, such as Java, C++, etc., including conventional procedural Programming language—such as the "C" language or a similar programming language.
  • the program code can execute entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the remote computing device on the user computing device, or entirely on the remote computing device or server. Execute on.
  • the remote computing device can be connected to the user computing device via any kind of network, including a local area network (LAN) or wide area network (WAN), or can be connected to an external computing device (eg, provided using an Internet service) Businesses are connected via the Internet).
  • LAN local area network
  • WAN wide area network
  • Businesses are connected via the Internet.

Landscapes

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

Abstract

本公开涉及数据处理技术领域，尤其涉及一种行人流量漏斗生成方法及装置、存储介质、电子设备。该方法可以包括获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。本公开提高了检测头肩区域的准确率，避免了头肩区域的漏检，进而提高了行人计数的准确率，从而提高了行人流量漏斗的准确率。

Description

行人流量漏斗生成方法及装置、存储介质、电子设备
相关申请的交叉引用
本申请要求于2017年10月24日递交的、名称为《行人流量漏斗生成方法及装置、存储介质、电子设备》的中国专利申请第201711002265.5号的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本公开涉及数据处理技术领域，尤其涉及一种行人流量漏斗生成方法及装置、存储介质、电子设备。
背景技术
随着社会的不断进步，视频监控系统的应用范围越来越广。在超市、商场、体育馆以及机场车站等场所的出入口常安装有监控摄像机，以便保安人员和管理者对这些场所的出入口进行监控。另一方面，超市、商场、体育馆以及机场车站等场所进出的人流量对于上述场所的经营者或管理者来说有着重要的意义，其中，人流量是指按一定方向流动的人数，本文中特指按进入/离开两个方向流动的人数。
现有的视频监控中，人流量统计主要是通过监控人员人工清点来实现，但存在准确率难以保证的问题，同时人工成本较高。
因此，需要提供一种行人流量漏斗生成方法，以更加快速且准确的获取人流量数据和性别年龄统计数据，进而基于准确的人流量数据和性别年龄统计数据生成更加准确的行人流量漏斗。
需要说明的是，在上述背景技术部分公开的信息仅用于加强对本公开的背景的理解，因此可以包括不构成对本领域普通技术人员已知的现有技术的信息。
公开内容
本公开的目的在于提供一种行人流量漏斗生成方法及装置、存储介质、电子设备，进而至少在一定程度上克服由于相关技术的限制和缺陷而导致的一个或者多个问题。
根据本公开的一个方面，提供一种行人流量漏斗生成方法，包括：
获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
在本公开的一种示例性实施例中，在所述获取当前帧图像之前还包括：
获取第一帧图像，并根据所述头肩识别模型获取所述第一帧图像中的头肩区域，并通过所述头肩区域对所述跟踪序列集合进行初始化。
在本公开的一种示例性实施例中，所述方法还包括：
根据一性别年龄识别模型识别所述跟踪序列集合中的各头肩区域的年龄段和性别。
在本公开的一种示例性实施例中，所述基于所述行人的计数结果生成一行人流量漏斗包括：
基于所述行人的计数结果并结合所述跟踪序列集合中的各头肩区域的年龄段和性别生成所述行人流量漏斗。
在本公开的一种示例性实施例中，所述根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新包括：
根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新。
在本公开的一种示例性实施例中，所述根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新包括：
分别基于所述跟踪序列集合中的各头肩区域在上一帧图像中的位置在所述当前帧图像中获取与所述跟踪序列集合中的各头肩区域对应的候选跟踪目标；
分别根据与各所述头肩区域对应的跟踪器对应的计算与各所述头肩区域对应的候选跟踪目标的响应值；
将各所述头肩区域对应的候选跟踪目标中的响应值最大的候选跟踪目标确定为对应的头肩区域在所述当前帧图像中的跟踪目标；以及
根据各所述头肩区域在所述当前帧图像中的跟踪目标对应的更新所述跟踪序列集合中的各头肩区域。
在本公开的一种示例性实施例中，所述方法还包括：计算各所述头肩区域对应的跟踪器，包括：
分别基于所述跟踪序列集合中的各头肩区域在所述上一帧图像中的位置在所述上一帧图像中获取与所述跟踪序列集合中的各头肩区域对应的第一训练样本集；
分别根据与各所述头肩区域对应的第一训练样本集训练一回归模型，以得到与各所述头肩区域对应的跟踪器。
在本公开的一种示例性实施例中，所述方法还包括根据卷积神经网络生成所述头肩识别模型，包括：
根据ImageNet分类数据集训练MobileNet网络，以得到所述MobileNet网络的权重 值；
在所述MobileNet网络之上增加一预设层数的卷基层以得到头肩检测网络，其中，所述卷基层的大小逐层递减；
获取标记出所述头肩区域的第二训练样本集，并根据所述第二训练样本集对保留有所述MobileNet网络的权重值的所述头肩检测网络进行训练，以得到所述头肩识别模型。
在本公开的一种示例性实施例中，所述根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新包括：
计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度；
根据所述相似度对所述跟踪序列集合进行更新。
在本公开的一种示例性实施例中，所述计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度包括：
根据下式计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度：
Figure PCTCN2018111713-appb-000001
其中，sim(Q i,Q j)为所述当前帧图像中的第i个头肩区域Q i与所述跟踪目标集合中的第j个头肩区域Q j的相似度，Q iarea为所述当前帧图像中的第i个头肩区域Q i的面积，Q jarea为所述跟踪序列集合中的第j个头肩区域Q j的面积。
在本公开的一种示例性实施例中，所述根据所述相似度对所述跟踪序列集合进行更新包括：
根据所述相似度与一相似度阈值，分别将所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域进行匹配；
若匹配成功，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域；
若匹配失败，将所述当前帧图像中的匹配失败的头肩区域添加至所述跟踪序列集合中。
在本公开的一种示例性实施例中，所述用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域包括：
分别计算所述当前帧图像中的匹配成功的头肩区域和对应的所述跟踪序列集合中的头肩区域的置信度；
在所述当前帧图像中的匹配成功的头肩区域的置信度大于对应的所述跟踪序列集合中的头肩区域的置信度时，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述 跟踪序列集合中的头肩区域。
在本公开的一种示例性实施例中，所述置信度的计算公式为：
Figure PCTCN2018111713-appb-000002
其中：conf(obj)为obj的置信度，obj为所述当前帧图像中的头肩区域或所述跟踪序列集合中的头肩区域，area(obj)为obj的面积，Score(obj)为所述头肩识别模型计算的obj的类别归属概率，α取0.5，B取1000。
在本公开的一种示例性实施例中，在所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数之前还包括：
在所述跟踪序列集合中存在在第一预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于跟踪状态的头肩区域时，将在第一预设帧数的图像中均未被从每帧图像中获取的头肩区域更新头肩区域从跟踪状态改为非正常状态；
在所述跟踪序列集合中存在在第二预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于非正常状态的头肩区域时，在跟踪序列集合中删除在第二预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于非正常状态的头肩区域。
在本公开的一种示例性实施例中，所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数包括：
分析所述跟踪序列集合中各头肩区域的运动轨迹；
根据各头肩区域的运动轨迹并结合一虚拟计数线对行人进行计数。
在本公开的一种示例性实施例中，所述方法还包括：构建所述性别年龄识别模型，包括：
在LFW数据集和社交网站中获取标记性别和年龄段的第三训练样本集；
利用所述第三训练样本集对性别年龄网络进行训练，以得到所述性别年龄识别模型，其中，所述性别年龄网络包括3个卷积层和3各全连接层。
根据本公开的一个方面，提供一种行人流量漏斗生成装置，包括：
跟踪更新模块，用于获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
检测更新模块，用于根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
技术生成模块，用于分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
根据本公开的一个方面，提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现上述任意一项所述的行人流量漏斗生成方法。
根据本公开的一个方面，提供一种电子设备，包括：
处理器；以及
存储器，用于存储所述处理器的可执行指令；
其中，所述处理器配置为经由执行所述可执行指令来执行上述中任意一项所述的行人流量漏斗生成方法。
本公开一种示例实施例提供的行人流量漏斗生成方法及装置、存储介质、电子设备。首先获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新，然后根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新，最后分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。一方面，通过多目标跟踪算法并结合头肩识别模型可以避免头肩区域的漏检，提高了检测头肩区域的准确率，进而提高了行人计数的准确率，从而提高了行人流量漏斗的准确率；另一方面，通过头肩识别模型可以快速的获取头肩区域且成本低，从而提高了头肩区域的计数速度并降低了计数成本，进而也提高了生成行人流量漏斗的速度并降低了生成行人流量漏斗的成本。
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，并不能限制本公开。
附图说明
通过参照附图来详细描述其示例性实施例，本公开的上述和其它特征及优点将变得更加明显。显而易见地，下面描述中的附图仅仅是本公开的一些实施例，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的附图。在附图中：
图1为本公开一种行人流量漏斗生成方法的流程图；
图2为本公开一示例性实施例提供的掩膜版图的示意图；
图3为本公开一种行人流量漏斗生成方法的框图；
图4为本公开示一示例性实施例中的电子设备的模块示意图；
图5为本公开示一示例性实施例中的程序产品示意图。
具体实施方式
现在将参考附图更全面地描述示例实施例。然而，示例实施例能够以多种形式实施，且不应被理解为限于在此阐述的实施例；相反，提供这些实施例使得本公开将全面和完整，并将示例实施例的构思全面地传达给本领域的技术人员。在图中相同的附图标记表示相同或类似的部分，因而将省略对它们的重复描述。
此外，所描述的特征、结构或特性可以以任何合适的方式结合在一个或更多实施例 中。在下面的描述中，提供许多具体细节从而给出对本公开的实施例的充分理解。然而，本领域技术人员将意识到，可以实践本公开的技术方案而没有所述特定细节中的一个或更多，或者可以采用其它的方法、组元、材料、装置、步骤等。在其它情况下，不详细示出或描述公知结构、方法、装置、实现、材料或者操作以避免模糊本公开的各方面。
附图中所示的方框图仅仅是功能实体，不一定必须与物理上独立的实体相对应。即，可以采用软件形式来实现这些功能实体，或在一个或多个软件硬化的模块中实现这些功能实体或功能实体的一部分，或在不同网络和/或处理器装置和/或微控制器装置中实现这些功能实体。
相关视频监控技术中，人流量统计主要是通过监控人员人工清点来实现。这种人工统计人流量的方法在监控时间短、人流量稀疏的情况下比较可靠，但由于人眼生物特性的限制，当监控时间较长，人流量密集时，统计的准确性将大大下降(即出现漏检)，而且人工统计的方式需要耗费大量的人力成本。在此基础上，若要统计人流量的性别和年龄，会更进一步的降低统计的效率和准确率，且增加成了本。此外，由于人流量和性别年龄的统计准确率低，因此，基于不准确的人流量和性别年龄的统计数据得到的流量漏斗的准确度也不高，从而使得统计人员通过该流量漏斗无法得到到更加准确的分析数据。
本示例性实施例中首先公开了一种行人流量漏斗生成方法，参照图1所示，所述行人流量漏斗生成方法可以包括以下步骤：
步骤S1、获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
步骤S2、根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
步骤S3、分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
根据本示例性实施例中的行人流量漏斗生成方法一方面，通过多目标跟踪算法并结合头肩识别模型可以避免头肩区域的漏检，提高了检测头肩区域的准确率，进而提高了行人计数的准确率，从而提高了行人流量漏斗的准确率；另一方面，通过头肩识别模型可以快速的获取头肩区域且成本低，从而提高了头肩区域的计数速度并降低了计数成本，进而也提高了生成行人流量漏斗的速度并降低了生成行人流量漏斗的成本。
下面，将参照图1，对本示例性实施例中的行人流量漏斗生成方法作进一步说明。
步骤S1、获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新。
在本示例性实施例中，在所述获取当前帧图像之前还可以包括：获取第一帧图像，并根据所述头肩识别模型获取所述第一帧图像中的头肩区域，并通过所述头肩区域对所 述跟踪序列集合进行初始化。在本示例性实施例中，可以实时接收监控设备传送的视频资料，并在视频资料中截取第一帧图像。例如，在统计商场营业期间内的人流量和性别年龄段时，可以获取商场营业开始时间的图像作为第一帧图像，例如，从八点开始获取图像。在获取到第一帧图像时，根据头肩识别模型对第一帧图像中的头肩区域进行识别；在识别出头肩区域之后，根据识别出的头肩区域对跟踪序列集合进行初始化，即将识别出的头肩区域加入跟踪序列集合中。需要说明的是，在通过头肩识别模型识别第一帧图像中的头肩区域时，还可以通过头肩识别模型计算出每个头肩区域的Score值，即每个头肩区域的类别归属概率。
在本示例性实施例中，可以根据卷积神经网络生成所述头肩识别模型，具体的，根据卷积神经网络生成所述头肩识别模型的过程包括以下步骤：
步骤S111，根据ImageNet分类数据集训练MobileNet网络，以得到所述MobileNet网络的权重值。
在本示例性实施例中，所述ImageNet是目前世界上图像识别最大的数据库。所述MobileNet网络为一种基于一个流线型的架构，使用深度可分离的卷积轻量级的深层神经网络，其可以将标准卷积分解成一个深度卷积和一个点卷积，即将每个卷积核应用到每一个通道，而1×1卷积用来组合通道卷积的输出。已经证明，这种分解可以有效减少计算量，降低模型大小，网络性能较好同时速度也非常快。在步骤S11中可以将ImageNet分类数据集中的每个标记过头肩区域的图像输入至MobileNet网络中，以得到MobileNet网络的权重值。
步骤S112，在所述MobileNet网络之上增加一预设层数的卷基层以得到头肩检测网络，其中，所述卷基层的大小逐层递减。
在本示例性实施例中，所述预设层数可以根据具体的需求进行设置，例如，可以为3层，还可以为6层，本示例性实施例对此不作特殊限定。所述卷基层的大小逐层递减，以利用大小逐层递减的卷基层完成多维度的头肩区域的预测。
步骤S113，获取标记出所述头肩区域的第二训练样本集，并根据所述第二训练样本集对保留有所述MobileNet网络的权重值的所述头肩检测网络进行训练，以得到所述头肩识别模型。
在本示例性实施例中，为了保证第二训练样本集的多样性，以提高头肩识别模型对头肩区域的识别的准确率，可以通过图像获取模块获取不同角度以及不同背景(树枝遮挡、建筑物遮挡等背景)中的头肩图像，并对每个头肩图像中的头肩区域进行标记，以得到标记出头肩区域的第二训练样本集。基于第二训练样本集并利用迁移率学习方法对保留有所述MobileNet网络的权重值的所述头肩检测网络进行训练，以得到头肩识别模型。具体的，将第二训练样本集中的标记出头肩区域的图像输入至头肩检测网络，以得到头肩检测网络中的权重值，在头肩检测网络的loss基本稳定不在下降，且平均值为0.8 时，该头肩检测网络即为头肩识别模型。
在本示例性实施例中，在步骤S1中的跟踪序列集合为在当前帧图像的上一帧图形中经过更新的跟踪序列集合。需要说明的是，在当前帧图像的上一帧图像为第一帧图像时，此时的跟踪序列集合为在第一帧图像中经过初始或后的跟踪序列集合。
在步骤S1中，所述根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新可以包括：根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新。具体的，所述根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新可以包括：
步骤S121，分别基于所述跟踪序列集合中的各头肩区域在上一帧图像中的位置在所述当前帧图像中获取与所述跟踪序列集合中的各头肩区域对应的候选跟踪目标。
在本示例性实施例中，可以依次获取跟踪序列集合中的各头肩区域对应的候选跟踪目标，其中，每个头肩区域对应的候选跟踪目标可以包括多个。下面，以获取跟踪序列集合中的第一个头肩区域对应的候选跟踪目标为例进行说明，在当前帧图像中以第一个头肩区域在上一帧图像中的位置为中心，在预设范围内，获取多个候选跟踪目标，该多个候选跟踪目标即为第一头肩区域对应的候选跟踪目标，所述预设范围可以由开发商进行设置。在获取跟踪序列集合中的其他头肩区域对应的候选跟踪目标时，其获取原则与获取第一头肩区域对应的候选跟踪目标相同，仅获取位置发生了变化，因此此处不再赘述。
步骤122，分别根据与各所述头肩区域对应的跟踪器对应的计算与各所述头肩区域对应的候选跟踪目标的响应值。
在本示例性实施例中，先对计算各所述头肩区域对应的跟踪器的过程进行说明。所述计算各所述头肩区域对应的跟踪器可以包括：分别基于所述跟踪序列集合中的各头肩区域在所述上一帧图像中的位置在所述上一帧图像中获取与所述跟踪序列集合中的各头肩区域对应的第一训练样本集；分别根据与各所述头肩区域对应的第一训练样本集训练一回归模型，以得到与各所述头肩区域对应的跟踪器。
在本示例性实施例中，以计算跟踪序列集合中的第一头肩区域对应的跟踪器为例进行说明。在上一帧图像中以第一头肩区域在上一帧图像中的位置为中心，获取多个第一训练样本，以组成第一头肩区域对应的第一训练样本集，将第一训练样本集中的所有第一训练样本输入回归模型，以对回归模型进行训练，将该训练后的回归模型确定为第一头肩区域对应的跟踪器。计算跟踪序列集合中的其他头肩区域对应的跟踪器的过程与上述计算第一头肩区域对应的跟踪器的过程相同，仅获取第一训练样本集的位置不同，因此此处不再赘述。
由上可知，由于在计算头肩区域对应的跟踪器时，基于头肩区域在上一帧图像中的 位置在上一帧图像中获取与头肩区域对应的第一训练样本集，又由于在不同帧图形中，同一头肩区域的位置不同，因此，对于不同帧图形中的同一头肩区域对应的跟踪器是不同的。基于此，通过每一帧对应的跟踪器计算得出的响应值更加准确，从而使得跟踪目标的跟踪确定更加准确。
基于此，以根据与第一头肩区域对应的跟踪器计算与第一头肩区域对应的候选跟踪目标的响应值为例进行说明，分别将第一头肩区域对应的各候选跟踪目标代入与第一头肩区域对应的跟踪器中，以得到第一头肩区域对应的各候选目标的响应值。计算跟踪序列集合中的其他头肩区域对应的候选跟踪目标的响应值与上述计算第一头肩区域对应的候选跟踪目标的响应值的过程相同，仅仅改变了跟踪器，即跟踪器要与头肩区域对应，因此此处不再赘述。
步骤123，将各所述头肩区域对应的候选跟踪目标中的响应值最大的候选跟踪目标确定为对应的头肩区域在所述当前帧图像中的跟踪目标。
在本示例性实施例中，以跟踪序列集合中的第一头肩区域进行说明，分别将第一头肩区域对应的各候选跟踪目标的响应值进行比对，将响应值最大的候选跟踪目标确定为第一头肩区域在当前帧图像中的跟踪目标，即，该候选跟踪目标为当前帧图像中的第一头肩区域。需要说明的是，跟踪序列集合中的其他头肩区域在当前帧图像中的跟踪目标的确定与上述第一头肩区域在当前帧图像中的跟踪目标的确定方式相同，因此，此处不作赘述。
步骤124，根据各所述头肩区域在所述当前帧图像中的跟踪目标对应的更新所述跟踪序列集合中的各头肩区域。
在本示例性实施例中，用当前帧图像中与跟踪序列集合中的各头肩区域对应的跟踪目标对应的替换跟踪序列集合中的各头肩区域，以更新跟踪序列集合中的各头肩区域。
综上所述，在头肩区域从开阔的背景中走进有树枝等遮挡物遮挡的背景中时，头肩识别模型可能无法在有遮挡物的背景图形中检测到该头肩区域，基于此，在仅仅通过头肩识别模型对的当前帧中的头肩区域进行检测以实现对跟踪序列集合中的头肩区域进行跟踪时，可以无法检测跟踪序列中的头肩区域，从而造成头肩区域的漏检。而在本示例性实施例中，通过多目标跟踪算法可以对跟踪序列集合中的所有头肩区域进行跟踪，避免了头肩区域的漏检，提高了检测头肩区域的准确率，进而提高了行人计数的准确率，从而提高了行人流量漏斗的准确率。
在步骤S2中，根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新。
在本示例性实施例中，将当前帧图像输入头肩识别模型中，以获得当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新，即将新的头肩区域加入跟踪序列集合中，并对跟踪序列中的原有的头肩区域进行更新。通过头 肩识别模型可以快速的获取头肩区域且成本低，从而提高了头肩区域的计数速度并降低了计数成本，进而也提高了生成行人流量漏斗的速度并降低了生成行人流量漏斗的成本。
所述根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新可以包括：计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度；根据所述相似度对所述跟踪序列集合进行更新。
所述计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度可以包括：根据下式计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度：
Figure PCTCN2018111713-appb-000003
其中，sim(Q i,Q j)为所述当前帧图像中的第i个头肩区域Q i与所述跟踪目标集合中的第j个头肩区域Q j的相似度，Q iarea为所述当前帧图像中的第i个头肩区域Q i的面积，Q jarea为所述跟踪序列集合中的第j个头肩区域Q j的面积。
以计算当前帧图像中的第一个头肩区域分别与跟踪序列集合中的各头肩区域的相似度为例进行说明，获取当前帧图像中的第一头肩区域的面积以及跟踪序列集合中的各头肩区域的面积，结合上述公式计算当前帧图像中的第一头肩区域分别与跟踪序列集合中的各头肩区域的相似度。需要说明的是，重复上述步骤可以计算当前帧图像中的其他头肩区域分别与跟踪序列集合中的各头肩区域的相似度的。
所述根据所述相似度对所述跟踪序列集合进行更新可以包括：
步骤S21，根据所述相似度与一相似度阈值，分别将所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域进行匹配。
在本示例性实施例中，所述相似度阈值可以为0.5，但本示例性实施例对此不作特殊限定。下面以当前帧图像中的第一头肩区域的匹配过程进行说明，首先获取第一头肩区域与跟踪序列集合中的各头肩区域的最大相似度，并判断该最大相似度是否大于相似度阈值，在判断该最大相似度大于相似度阈值时，认为匹配成功，即该第一头肩区域与对应的跟踪序列集合中的头肩区域为同一个头肩区域。在判断该最大相似度不大于相似度阈值时，认为匹配失败，即该头肩区域为一个新的头肩区域。
步骤S22，若匹配成功，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域。
在本示例性实施例中，以当前帧图像中的第一头肩区域为例进行说明，在第一头肩区域与跟踪序列集合中的各头肩区域的最大相似度大于相似度阈值时，用该第一头肩区域对应的替换跟踪序列集合中的头肩区域。
为了更进一步的增加匹配的精度，所述用所述当前帧图像中的匹配成功的头肩区域 对应的替换所述跟踪序列集合中的头肩区域可以包括：分别计算所述当前帧图像中的匹配成功的头肩区域和对应的所述跟踪序列集合中的头肩区域的置信度；在所述当前帧图像中的匹配成功的头肩区域的置信度大于对应的所述跟踪序列集合中的头肩区域的置信度时，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域。
在本示例性实施例中，所述置信度的计算公式为：
Figure PCTCN2018111713-appb-000004
其中：conf(obj)为obj的置信度，obj为所述当前帧图像中的头肩区域或所述跟踪序列集合中的头肩区域，area(obj)为obj的面积，Score(obj)为所述头肩识别模型计算的obj的类别归属概率，α取0.5，B取1000。
以当前帧图像中的第一头肩区域与跟踪序列集合中的第三个头肩区域匹配成功为例对上述过程进行说明，通过上述置信度公式分别计算当前帧图像中的第一头肩区域的置信度与跟踪序列集合中的第三个头肩区域的置信度；在当前帧图像中的第一头肩区域的置信度大于跟踪序列集合中的第三个头肩区域的置信度时，用当前帧图像中的第一头肩区域替换跟踪序列集合中的第三个头肩区域。
步骤S23，若匹配失败，将所述当前帧图像中的匹配失败的头肩区域添加至所述跟踪序列集合中。
在本示例性实施例中，以当前帧图像中的第一头肩区域为例进行说明，在第一头肩区域与跟踪序列集合中的各头肩区域的最大相似度不大于相似度阈值时，将该第一头肩区域加入跟踪序列集合中，以成为新的跟踪目标。
由上可知，通过肩识别模型获取当前帧图像中的头肩区域，并通过计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度，以及根据所述相似度对所述跟踪序列集合进行更新，可以将检测到的新的头肩区域加入跟踪序列集合中，并对跟踪序列集合中原有的头肩区域进行更新，提高了头肩区域检测的准确性，从而提高了计数的真确率，避免了跟踪目标的漏检，进而提高了行人流量漏斗的准确率。
在步骤S3中，分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
在本示例性实施例中，为了提供跟目标的准确性，进而提高行人计数的准确率，在所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数之前还可以包括：
步骤S31、在所述跟踪序列集合中存在在第一预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于跟踪状态的头肩区域时，将在第一预设帧数的图像中均未 被从每帧图像中获取的头肩区域更新头肩区域从跟踪状态改为非正常状态。
在本示例性实施例中，所述第一预设帧数的图像可以为连续的5帧图像或连续的10帧图像，本示例性实施例对此不作特殊限定。以连续的5帧图像为例对步骤S31进行说明，根据头肩识别模型分别获取上述5帧图像对应的头肩区域，在上述5帧图像对应的头肩区域中均没有与跟踪序列集合中的其中一个处于跟踪状态头肩区域匹配成功的头肩区域，即认定该头肩区域为不正常的头肩区域，将该头肩区域从跟踪状态改为非正常状态。
步骤S32、在所述跟踪序列集合中存在在第二预设帧数的图像中均未匹配成功的处于非正常状态的头肩区域时，在所述跟踪序列集合中删除在第二预设帧数的图像中均匹配成功的处于非正常状态的头肩区域。
在本示例性实施例中，所述第二预设帧数的图像可以为连续的5帧图像或连续的10帧图像，本示例性实施例对此不作特殊限定。以连续的5帧图像为例对步骤S32进行说明，根据头肩识别模型分别获取上述5帧图像对应的头肩区域，在上述5帧图像对应的头肩区域中均没有与跟踪序列集合中的其中一个处于非正常状态的头肩区域匹配成功的头肩区域，即认定该头肩区域为误检测头肩区域，将该头肩区域从跟踪序列集合中删除。
在步骤S3中，所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数可以包括：分析所述跟踪序列集合中各头肩区域的运动轨迹；根据各头肩区域的运动轨迹并结合一虚拟计数线对行人进行计数。
在本示例性实施例中，可以根据跟踪序列集合中的各头肩区域的不同帧图像中位置，生成各头肩区域的运动轨迹。为了方便为计数，在本示例性实施例中，可以设置一虚拟计数线，并通过判断头肩区域的运动轨迹是否跨越该虚拟计数线以及跨越虚拟计数线的方向随不同方向的行人进行计数。具体的，通过判断头肩区域的运动轨迹是否跨越该虚拟计数线以及跨越虚拟计数线的方向对不同方向的行人进行计数的过程包括：可以使用掩膜版图MASK表示虚拟计数线两边的区域(如图2所示)，如第一区域和第二区域，在此基础上，根据各头肩区域的运动轨迹的像素值，判断各头肩区域是否从第一区域跨越到第二区域；或者判断各头肩区域是否从第二区域跨越至第一区域。
例如，第一区域为以小区内，第二区域为小区外，在一头肩区域从第一区域跨域到第二区域时，在出小区的人数上加1，在一头肩区域从第二区域跨越到第一区域时，在进入小区的人数上加1。
需要说明的是，为了防止重复计数，可以对已经记过数的头肩区域进行标记。由上可知，根据所述各头肩区域的运动轨迹和一虚拟计数线对所述行人进行计数，计数方法简单，易于实现，计数准确率高。
在根据各头肩区域的运动轨迹对行人进行计数后，判断当前帧图像是否为最后一帧图像，在当前帧图像不为最后一帧图像时，重复步骤S2，以及步骤S3，以对行人继续进 行计数。在当前帧图像为最后一帧图像时基于所述行人的计数结果生成一行人流量漏斗。例如，可以分析不同时段的人流量，以生成行人流量漏斗，以使工作人员可以根据该行人流量漏斗分析业务流程问题等，以对宏观数据进行把控。
为了使得成行人流量漏斗可以涵盖更多的维度的数据，以使工作人员可以根据该行人流量漏斗从多个维度分析业务流程问题等，以对宏观数据进行把控，所述方法还包括：根据一性别年龄识别模型识别所述跟踪序列集合中的各头肩区域的年龄段和性别；以及基于所述行人的计数结果并结合所述跟踪序列集合中的各头肩区域的年龄段和性别生成所述行人流量漏斗。
在本示例性实施例中，首先对构建所述性别年龄识别模型的过程进行说明，其中，构建所述性别年龄识别模型可以包括：在LFW数据集和社交网站中获取标记性别和年龄段的第三训练样本集；利用所述第三训练样本集对性别年龄网络进行训练，以得到所述性别年龄识别模型，其中，所述性别年龄网络包括3个卷积层和3各全连接层。
在本示例性实施例中，建立一个包括3各卷基层和3各全连接层的性别年龄网络；在LFW数据集合社交平台中获取第三训练样本集，并以第三训练样本集中的每个第三训练样本的人脸为中心进行裁剪，例如，以人脸为中心将大小为256*256大小的第三训练样本裁剪为大小为227*227大小的第三训练样本，以及对经过裁剪后的训练样本进行性别和年龄段的标识，其中，将性别标识为男或女，将年龄根据年龄段进行标识，例如，将年龄段可以分为8各阶段进行标识，分别为0～3、4～7、8～14、15～24、25～37、38～47、48～59、60+。需要说明的是，年龄段的划分不限于此，开发商可以根据需求进行设置；采用标准差为0.01，均值为0的正太高斯分布对性别年龄网络中的权重值进行初始化；通过随机梯度下降的方法并集合dropout对所述初始化后的性别年龄网络进行训练以得到性别年龄识别模型。其中dropout比例采用0.5。
基于该性别年龄识别模型对每个跟踪序列集合中的各头肩区域的年龄段和性别进行识别，需要说明的是，只需要对第一次加入跟踪序列集合中的头肩区域的年龄段和性别进行识别。此外由于年龄段识别和性别识别采用同一个性别年龄识别模型，因此，识别年龄时，性别年龄识别模型输出为2，在识别性别时，性别年龄识别模型输出为8。
在此基础上，在识别出头肩区域的性别和年龄后，基于所述行人的计数结果并结合所述跟踪序列集合中的各头肩区域的年龄段和性别生成所述行人流量漏斗。例如，可以分析不同时间段的人流量和性别比例以及年龄比例得到行人流量漏斗，以使工作人员可以根据该行人流量漏斗分析从多个维度业务流程问题等，以对宏观数据进行把控。
需要说明的是，尽管在附图中以特定顺序描述了本公开中方法的各个步骤，但是，这并非要求或者暗示必须按照该特定顺序来执行这些步骤，或是必须执行全部所示的步骤才能实现期望的结果。附加的或备选的，可以省略某些步骤，将多个步骤合并为一个步骤执行，以及/或者将一个步骤分解为多个步骤执行等。
在本公开的示例性实施例中，还提供了行人流量漏斗生成装置，如图3所示，所述行人流量漏斗生成装置100可以包括跟踪更新模块101、检测更新模块102、技术生成模块103其中：
跟踪更新模块101可以用于获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
检测更新模块102可以用于根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
技术生成模块103可以用于分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
上述中各行人流量漏斗生成装置模块的具体细节已经在对应的行人流量漏斗生成方法中进行了详细的描述，因此此处不再赘述。
应当注意，尽管在上文详细描述中提及了用于执行的设备的若干模块或者单元，但是这种划分并非强制性的。实际上，根据本公开的实施方式，上文描述的两个或更多模块或者单元的特征和功能可以在一个模块或者单元中具体化。反之，上文描述的一个模块或者单元的特征和功能可以进一步划分为由多个模块或者单元来具体化。
在本公开的示例性实施例中，还提供了一种能够实现上述方法的电子设备。
所属技术领域的技术人员能够理解，本公开的各个方面可以实现为系统、方法或程序产品。因此，本公开的各个方面可以具体实现为以下形式，即：完全的硬件实施方式、完全的软件实施方式(包括固件、微代码等)，或硬件和软件方面结合的实施方式，这里可以统称为“电路”、“模块”或“系统”。
下面参照图4来描述根据本公开的这种实施方式的电子设备600。图4显示的电子设备600仅仅是一个示例，不应对本公开实施例的功能和使用范围带来任何限制。
如图4所示，电子设备600以通用计算设备的形式表现。电子设备600的组件可以包括但不限于：上述至少一个处理单元610、上述至少一个存储单元620、连接不同系统组件(包括存储单元620和处理单元610)的总线630、显示单元640。
其中，所述存储单元存储有程序代码，所述程序代码可以被所述处理单元610执行，使得所述处理单元610执行本说明书上述“示例性方法”部分中描述的根据本公开各种示例性实施方式的步骤。例如，所述处理单元610可以执行如图1中所示的步骤S1、获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；步骤S2、根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；步骤S3、分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
存储单元620可以包括易失性存储单元形式的可读介质，例如随机存取存储单元(RAM)6201和/或高速缓存存储单元6202，还可以进一步包括只读存储单元(ROM)6203。
存储单元620还可以包括具有一组(至少一个)程序模块6205的程序/实用工具6204，这样的程序模块6205包括但不限于：操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。
总线630可以为表示几类总线结构中的一种或多种，包括存储单元总线或者存储单元控制器、外围总线、图形加速端口、处理单元或者使用多种总线结构中的任意总线结构的局域总线。
电子设备600也可以与一个或多个外部设备700(例如键盘、指向设备、蓝牙设备等)通信，还可与一个或者多个使得用户能与该电子设备600交互的设备通信，和/或与使得该电子设备600能与一个或多个其它计算设备进行通信的任何设备(例如路由器、调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口650进行。并且，电子设备600还可以通过网络适配器660与一个或者多个网络(例如局域网(LAN)，广域网(WAN)和/或公共网络，例如因特网)通信。如图所示，网络适配器660通过总线630与电子设备600的其它模块通信。应当明白，尽管图中未示出，可以结合电子设备600使用其它硬件和/或软件模块，包括但不限于：微代码、设备驱动器、冗余处理单元、外部磁盘驱动阵列、RAID系统、磁带驱动器以及数据备份存储系统等。
通过以上的实施方式的描述，本领域的技术人员易于理解，这里描述的示例实施方式可以通过软件实现，也可以通过软件结合必要的硬件的方式来实现。因此，根据本公开实施方式的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性存储介质(可以是CD-ROM，U盘，移动硬盘等)中或网络上，包括若干指令以使得一台计算设备(可以是个人计算机、服务器、终端装置、或者网络设备等)执行根据本公开实施方式的方法。
在本公开的示例性实施例中，还提供了一种计算机可读存储介质，其上存储有能够实现本说明书上述方法的程序产品。在一些可能的实施方式中，本公开的各个方面还可以实现为一种程序产品的形式，其包括程序代码，当所述程序产品在终端设备上运行时，所述程序代码用于使所述终端设备执行本说明书上述“示例性方法”部分中描述的根据本公开各种示例性实施方式的步骤。
参考图5所示，描述了根据本公开的实施方式的用于实现上述方法的程序产品800，其可以采用便携式紧凑盘只读存储器(CD-ROM)并包括程序代码，并可以在终端设备，例如个人电脑上运行。然而，本公开的程序产品不限于此，在本文件中，可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
所述程序产品可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以为但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了可读程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。可读信号介质还可以是可读存储介质以外的任何可读介质，该可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于无线、有线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言的任意组合来编写用于执行本公开操作的程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、C++等，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。在涉及远程计算设备的情形中，远程计算设备可以通过任意种类的网络，包括局域网(LAN)或广域网(WAN)，连接到用户计算设备，或者，可以连接到外部计算设备(例如利用因特网服务提供商来通过因特网连接)。
此外，上述附图仅是根据本公开示例性实施例的方法所包括的处理的示意性说明，而不是限制目的。易于理解，上述附图所示的处理并不表明或限制这些处理的时间顺序。另外，也易于理解，这些处理可以是例如在多个模块中同步或异步执行的。
本领域技术人员在考虑说明书及实践这里公开的发明后，将容易想到本公开的其他实施例。本申请旨在涵盖本公开的任何变型、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公开的一般性原理并包括本公开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，本公开的真正范围和精神由权利要求指出。
应当理解的是，本公开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。本公开的范围仅由所附的权利要求来限。

Claims (19)

  1. 一种行人流量漏斗生成方法，其特征在于，包括：
    获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
    根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
    分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
  2. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，在所述获取当前帧图像之前还包括：
    获取第一帧图像，并根据所述头肩识别模型获取所述第一帧图像中的头肩区域，并通过所述头肩区域对所述跟踪序列集合进行初始化。
  3. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，所述方法还包括：
    根据一性别年龄识别模型识别所述跟踪序列集合中的各头肩区域的年龄段和性别。
  4. 根据权利要求2所述的行人流量漏斗生成方法，其特征在于，所述基于所述行人的计数结果生成一行人流量漏斗包括：
    基于所述行人的计数结果并结合所述跟踪序列集合中的各头肩区域的年龄段和性别生成所述行人流量漏斗。
  5. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，所述根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新包括：
    根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新。
  6. 根据权利要求5所述的行人流量漏斗生成方法，其特征在于，所述根据核相关滤波跟踪算法在所述当前帧图像中分别对所述跟踪序列集合中的头肩区域进行跟踪和更新包括：
    分别基于所述跟踪序列集合中的各头肩区域在上一帧图像中的位置在所述当前帧图像中获取与所述跟踪序列集合中的各头肩区域对应的候选跟踪目标；
    分别根据与各所述头肩区域对应的跟踪器对应的计算与各所述头肩区域对应的候选跟踪目标的响应值；
    将各所述头肩区域对应的候选跟踪目标中的响应值最大的候选跟踪目标确定为对应的头肩区域在所述当前帧图像中的跟踪目标；以及
    根据各所述头肩区域在所述当前帧图像中的跟踪目标对应的更新所述跟踪序列集合中的各头肩区域。
  7. 根据权利要求6所述的行人流量漏斗生成方法，其特征在于，所述方法还包括： 计算各所述头肩区域对应的跟踪器，包括：
    分别基于所述跟踪序列集合中的各头肩区域在所述上一帧图像中的位置在所述上一帧图像中获取与所述跟踪序列集合中的各头肩区域对应的第一训练样本集；
    分别根据与各所述头肩区域对应的第一训练样本集训练一回归模型，以得到与各所述头肩区域对应的跟踪器。
  8. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，所述方法还包括根据卷积神经网络生成所述头肩识别模型，包括：
    根据ImageNet分类数据集训练MobileNet网络，以得到所述MobileNet网络的权重值；
    在所述MobileNet网络之上增加一预设层数的卷基层以得到头肩检测网络，其中，所述卷基层的大小逐层递减；
    获取标记出所述头肩区域的第二训练样本集，并根据所述第二训练样本集对保留有所述MobileNet网络的权重值的所述头肩检测网络进行训练，以得到所述头肩识别模型。
  9. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，所述根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新包括：
    计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度；
    根据所述相似度对所述跟踪序列集合进行更新。
  10. 根据权利要求9所述的行人流量漏斗生成方法，其特征在于，所述计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度包括：
    根据下式计算所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域的相似度：
    Figure PCTCN2018111713-appb-100001
    其中，sim(Q i,Q j)为所述当前帧图像中的第i个头肩区域Q i与所述跟踪目标集合中的第j个头肩区域Q j的相似度，Q iarea为所述当前帧图像中的第i个头肩区域Q i的面积，Q jarea为所述跟踪序列集合中的第j个头肩区域Q j的面积。
  11. 根据权利要求9所述的行人流量漏斗生成方法，其特征在于，所述根据所述相似度对所述跟踪序列集合进行更新包括：
    根据所述相似度与一相似度阈值，分别将所述当前帧图像中的各头肩区域与所述跟踪序列集合中的各头肩区域进行匹配；
    若匹配成功，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域；
    若匹配失败，将所述当前帧图像中的匹配失败的头肩区域添加至所述跟踪序列集合中。
  12. 根据权利要求11所述的行人流量漏斗生成方法，其特征在于，所述用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域包括：
    分别计算所述当前帧图像中的匹配成功的头肩区域和对应的所述跟踪序列集合中的头肩区域的置信度；
    在所述当前帧图像中的匹配成功的头肩区域的置信度大于对应的所述跟踪序列集合中的头肩区域的置信度时，用所述当前帧图像中的匹配成功的头肩区域对应的替换所述跟踪序列集合中的头肩区域。
  13. 根据权利要求12所述的行人流量漏斗生成方法，其特征在于，所述置信度的计算公式为：
    Figure PCTCN2018111713-appb-100002
    其中：conf(obj)为obj的置信度，obj为所述当前帧图像中的头肩区域或所述跟踪序列集合中的头肩区域，area(obj)为obj的面积，Score(obj)为所述头肩识别模型计算的obj的类别归属概率，α取0.5，B取1000。
  14. 根据权利要求11所述的行人流量漏斗生成方法，其特征在于，在所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数之前还包括：
    在所述跟踪序列集合中存在在第一预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于跟踪状态的头肩区域时，将在第一预设帧数的图像中均未被从每帧图像中获取的头肩区域更新头肩区域从跟踪状态改为非正常状态；
    在所述跟踪序列集合中存在在第二预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于非正常状态的头肩区域时，在跟踪序列集合中删除在第二预设帧数的图像中均未被从每帧图像中获取的头肩区域更新的处于非正常状态的头肩区域。
  15. 根据权利要求1所述的行人流量漏斗生成方法，其特征在于，所述分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数包括：
    分析所述跟踪序列集合中各头肩区域的运动轨迹；
    根据各头肩区域的运动轨迹并结合一虚拟计数线对行人进行计数。
  16. 根据权利要求3所述的行人流量漏斗生成方法，其特征在于，所述方法还包括：构建所述性别年龄识别模型，包括：
    在LFW数据集和社交网站中获取标记性别和年龄段的第三训练样本集；
    利用所述第三训练样本集对性别年龄网络进行训练，以得到所述性别年龄识别模型，其中，所述性别年龄网络包括3个卷积层和3各全连接层。
  17. 一种行人流量漏斗生成装置，其特征在于，包括：
    跟踪更新模块，用于获取当前帧图像，根据多目标跟踪算法在所述当前帧图像中对跟踪序列集合中的头肩区域进行跟踪和更新；
    检测更新模块，用于根据头肩识别模型获取所述当前帧图像中的头肩区域，并根据所述当前帧图像中的头肩区域对所述跟踪序列集合进行更新；
    技术生成模块，用于分析所述跟踪序列集合中各头肩区域的运动轨迹以对所述行人进行计数，并在所述当前帧图像为最后一帧图像时，基于所述行人的计数结果生成一行人流量漏斗。
  18. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1～16中任意一项所述的行人流量漏斗生成方法。
  19. 一种电子设备，其特征在于，包括：
    处理器；以及
    存储器，用于存储所述处理器的可执行指令；
    其中，所述处理器配置为经由执行所述可执行指令来执行权利要求1～16中任意一项所述的行人流量漏斗生成方法。
PCT/CN2018/111713 2017-10-24 2018-10-24 行人流量漏斗生成方法及装置、存储介质、电子设备 WO2019080881A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2020520032A JP7270617B2 (ja) 2017-10-24 2018-10-24 歩行者流量ファネル生成方法及び装置、プログラム、記憶媒体、電子機器
US16/758,958 US11210795B2 (en) 2017-10-24 2018-10-24 Pedestrian flow funnel generation method and apparatus, storage medium and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711002265.5 2017-10-24
CN201711002265.5A CN109697499B (zh) 2017-10-24 2017-10-24 行人流量漏斗生成方法及装置、存储介质、电子设备

Publications (1)

Publication Number Publication Date
WO2019080881A1 true WO2019080881A1 (zh) 2019-05-02

Family

ID=66227762

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/111713 WO2019080881A1 (zh) 2017-10-24 2018-10-24 行人流量漏斗生成方法及装置、存储介质、电子设备

Country Status (4)

Country Link
US (1) US11210795B2 (zh)
JP (1) JP7270617B2 (zh)
CN (1) CN109697499B (zh)
WO (1) WO2019080881A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368650A (zh) * 2020-02-18 2020-07-03 北京交通大学 一种基于轨迹数据的人群拥挤行为识别方法

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11341650B2 (en) * 2020-03-19 2022-05-24 At&T Intellectual Property I, L.P. Method for accelerating three-dimensional object segmentation with point cloud simplifications
CN112434566B (zh) * 2020-11-04 2024-05-07 深圳云天励飞技术股份有限公司 客流统计方法、装置、电子设备及存储介质
CN112380960A (zh) * 2020-11-11 2021-02-19 广东电力信息科技有限公司 一种人群计数方法、装置、设备和存储介质
CN112270381B (zh) * 2020-11-16 2022-06-03 电子科技大学 基于深度学习的人流量检测方法
CN112464917A (zh) * 2021-01-06 2021-03-09 希望银蕨智能科技有限公司 一种非接触式测温系统的人数统计方法
CN112906590A (zh) * 2021-03-02 2021-06-04 东北农业大学 一种基于FairMOT的多目标跟踪人流量监测方法
CN113673342A (zh) * 2021-07-19 2021-11-19 浙江大华技术股份有限公司 行为检测方法、电子装置和存储介质
CN114372994B (zh) * 2022-01-10 2022-07-22 北京中电兴发科技有限公司 视频浓缩中背景图像的生成方法
CN114677651B (zh) * 2022-05-30 2022-09-27 山东极视角科技有限公司 一种基于低画质低帧率视频的客流统计方法及相关装置
CN116330658B (zh) * 2023-01-12 2024-05-28 珠海视熙科技有限公司 基于深度图像的目标跟踪方法、装置、系统及摄像设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101877058A (zh) * 2010-02-10 2010-11-03 杭州海康威视软件有限公司 人流量统计的方法及系统
CN103021059A (zh) * 2012-12-12 2013-04-03 天津大学 一种基于视频监控的公交客流计数方法
CN103425967A (zh) * 2013-07-21 2013-12-04 浙江大学 一种基于行人检测和跟踪的人流监控方法
CN106203513A (zh) * 2016-07-08 2016-12-07 浙江工业大学 一种基于行人头肩多目标检测及跟踪的统计方法

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7492924B2 (en) * 2003-07-18 2009-02-17 Intel Corporation Face tracking
JP2009211311A (ja) 2008-03-03 2009-09-17 Canon Inc 画像処理装置及び方法
CN101777114B (zh) * 2009-01-08 2013-04-24 北京中星微电子有限公司 视频监控智能分析系统和方法及头肩检测跟踪系统和方法
CN103049787B (zh) * 2011-10-11 2015-11-25 汉王科技股份有限公司 一种基于头肩特征的人数统计方法和系统
US9367733B2 (en) * 2012-11-21 2016-06-14 Pelco, Inc. Method and apparatus for detecting people by a surveillance system
KR102474837B1 (ko) * 2015-09-14 2022-12-07 주식회사 한화 전경 추출 방법 및 장치
JP2017097510A (ja) * 2015-11-20 2017-06-01 ソニー株式会社 画像処理装置と画像処理方法およびプログラム
US9911198B2 (en) * 2015-12-17 2018-03-06 Canon Kabushiki Kaisha Method, system and apparatus for matching moving targets between camera views
JP6688611B2 (ja) 2016-01-06 2020-04-28 パナソニックｉ−ＰＲＯセンシングソリューションズ株式会社 動線分析システム及び動線分析方法
US10402633B2 (en) * 2016-05-23 2019-09-03 Intel Corporation Human detection in high density crowds
CN106530340B (zh) * 2016-10-24 2019-04-26 深圳市商汤科技有限公司 一种指定物体跟踪方法
CN107066990B (zh) * 2017-05-04 2019-10-11 厦门美图之家科技有限公司 一种目标跟踪方法及移动设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101877058A (zh) * 2010-02-10 2010-11-03 杭州海康威视软件有限公司 人流量统计的方法及系统
CN103021059A (zh) * 2012-12-12 2013-04-03 天津大学 一种基于视频监控的公交客流计数方法
CN103425967A (zh) * 2013-07-21 2013-12-04 浙江大学 一种基于行人检测和跟踪的人流监控方法
CN106203513A (zh) * 2016-07-08 2016-12-07 浙江工业大学 一种基于行人头肩多目标检测及跟踪的统计方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368650A (zh) * 2020-02-18 2020-07-03 北京交通大学 一种基于轨迹数据的人群拥挤行为识别方法
CN111368650B (zh) * 2020-02-18 2024-03-12 北京交通大学 一种基于轨迹数据的人群拥挤行为识别方法

Also Published As

Publication number Publication date
JP2020537240A (ja) 2020-12-17
JP7270617B2 (ja) 2023-05-10
CN109697499A (zh) 2019-04-30
CN109697499B (zh) 2021-09-07
US20200349720A1 (en) 2020-11-05
US11210795B2 (en) 2021-12-28

Similar Documents

Publication Publication Date Title
WO2019080881A1 (zh) 行人流量漏斗生成方法及装置、存储介质、电子设备
WO2022134344A1 (zh) 目标检测方法、系统、设备及存储介质
Zhang et al. Deep convolutional neural networks for forest fire detection
US20200342271A1 (en) Pedestrian re-identification method, computer device and readable medium
Myrvoll et al. Counting public transport passenger using WiFi signatures of mobile devices
EP3633615A1 (en) Deep learning network and average drift-based automatic vessel tracking method and system
WO2022134120A1 (zh) 基于目标活动预测的停车场管控方法、装置及电子设备
CN105740615A (zh) 利用手机轨迹追踪传染源和预测传染病流行趋势的方法
US20220180534A1 (en) Pedestrian tracking method, computing device, pedestrian tracking system and storage medium
Rezaee et al. An autonomous UAV-assisted distance-aware crowd sensing platform using deep ShuffleNet transfer learning
CN108986143B (zh) 一种视频中目标检测跟踪方法
WO2023273041A1 (zh) 车路协同中目标检测方法、装置和路侧设备
CN114186069B (zh) 基于多模态异构图注意力网络的深度视频理解知识图谱构建方法
US20230041943A1 (en) Method for automatically producing map data, and related apparatus
WO2022134576A1 (zh) 红外视频时序行为定位方法、装置、设备及存储介质
WO2023082679A1 (zh) 复杂地下空间轨迹融合的视频人员重识别方法
CN109376736A (zh) 一种基于深度卷积神经网络的视频小目标检测方法
CN112634329A (zh) 一种基于时空与或图的场景目标活动预测方法及装置
Gündüz et al. A new YOLO-based method for social distancing from real-time videos
CN113780145A (zh) 精子形态检测方法、装置、计算机设备和存储介质
Zhang et al. A Multiple Instance Learning and Relevance Feedback Framework for Retrieving Abnormal Incidents in Surveillance Videos.
CN115482436B (zh) 图像筛选模型的训练方法、装置以及图像筛选方法
US20220415051A1 (en) Object re-identification
Xudong et al. Pedestrian detection and tracking with deep mutual learning
CN112541403A (zh) 一种利用红外摄像头的室内人员跌倒检测方法

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: 18871289

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020520032

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 19/08/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18871289

Country of ref document: EP

Kind code of ref document: A1