WO2019206265A1 - 行人重识别方法和装置 - Google Patents

行人重识别方法和装置 Download PDF

Info

Publication number
WO2019206265A1
WO2019206265A1 PCT/CN2019/084489 CN2019084489W WO2019206265A1 WO 2019206265 A1 WO2019206265 A1 WO 2019206265A1 CN 2019084489 W CN2019084489 W CN 2019084489W WO 2019206265 A1 WO2019206265 A1 WO 2019206265A1
Authority
WO
WIPO (PCT)
Prior art keywords
pedestrian
identification
identified
historical
feature
Prior art date
Application number
PCT/CN2019/084489
Other languages
English (en)
French (fr)
Inventor
吴江旭
张伟华
张洪光
彭刚林
孔磊锋
武跃峰
雍兴辉
李凡
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 EP19792513.4A priority Critical patent/EP3761221B1/en
Priority to US17/044,578 priority patent/US11804059B2/en
Publication of WO2019206265A1 publication Critical patent/WO2019206265A1/zh

Links

Images

Classifications

    • 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
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • 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
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport 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
    • 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
    • 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 image recognition, and in particular, to a pedestrian recognition method and apparatus.
  • the Pedestrian Re-Identification (Re-ID) technology tracks the pedestrian trajectory by using the same pedestrian image collected by the same camera at different time periods and the image of the same pedestrian collected by different cameras.
  • the present disclosure provides a pedestrian re-identification scheme capable of effectively suppressing external noise interference and improving the success rate of pedestrian recognition.
  • a pedestrian re-identification method comprising: detecting a pedestrian to be identified in a specified video frame; extracting a captured video frame within a specified time period before a shooting time of the specified video frame All pedestrians appearing as historical pedestrians, wherein the end time of the specified time period is the shooting time of the specified video frame; calculating the characteristic distance between the pedestrian to be identified and the historical pedestrian; and extracting the predetermined number according to the order of the feature distance from small to large The historical pedestrian identification corresponding to the feature distance; clustering the extracted historical pedestrian identifications to re-identify the pedestrians to be identified.
  • clustering the extracted historical pedestrian identification comprises: clustering the extracted historical pedestrian identifications to classify the same historical pedestrian identification into the same cluster set; counting each cluster The number of identifiers in the collection; the historical pedestrian identity in the cluster collection with the largest number of identifiers is used as the identifier of the pedestrian to be identified.
  • the method further includes: calculating, in a predetermined number of feature distances, an average value of the feature distances corresponding to each historical pedestrian identifier; and determining, according to the calculation result, the pedestrian to be identified And determining whether the average value of the corresponding feature distance is a minimum value; and determining that the average value of the feature distance corresponding to the identifier of the pedestrian to be identified is a minimum value, determining that the identity of the pedestrian to be identified matches the pedestrian to be identified.
  • the predetermined number of values is increased; determining whether the predetermined number of current values is greater than the first threshold; In the case that the current value is not greater than the first threshold, the step of extracting the historical pedestrian identification corresponding to the predetermined number of feature distances in the order of the feature distance from small to large is performed.
  • a new historical pedestrian identification is assigned to the pedestrian to be identified if the predetermined number of current values is greater than the first threshold.
  • the method further includes: counting, for each historical pedestrian identification, the number of corresponding feature distances; determining whether the statistical result exceeds the second threshold; In the case of the second threshold, the maximum feature distance in the corresponding feature distance is deleted; and then the step of extracting the historical pedestrian identity corresponding to the predetermined number of feature distances in the order of the feature distance from small to large is performed.
  • a pedestrian re-identification apparatus comprising: a detection module configured to detect a pedestrian to be identified in a specified video frame; and a historical pedestrian extraction module configured to specify a video frame During the specified time period before the shooting time, all pedestrians appearing in the captured video frame are extracted as historical pedestrians, wherein the end time of the specified time period is the shooting time of the specified video frame; the feature distance calculation module is configured to calculate the to-be-identified The feature extraction distance between the pedestrian and the historical pedestrian; the identifier extraction module is configured to extract the historical pedestrian identification corresponding to the predetermined number of feature distances according to the order of the feature distance from the smallest to the largest; the identification module is configured to extract the historical pedestrian identification Clustering is performed to identify the pedestrians to be re-identified.
  • the identification module is configured to cluster the extracted historical pedestrian identifications to group the same historical pedestrian identifications into the same cluster set, and to count the number of identifiers in each cluster set, The historical pedestrian identification in the cluster collection with the largest number of identifications is used as the identifier of the pedestrian to be identified.
  • the pedestrian re-identification device further includes an average value calculation module configured to calculate, after the identification module identifies the pedestrian to be identified according to the clustering result, calculate, in a predetermined number of feature distances, each historical pedestrian identification An average value of the feature distances; the identification module is further configured to determine, according to the calculation result of the average value calculation module, whether the average value of the feature distances corresponding to the identification of the pedestrian to be identified is a minimum value, corresponding to the identifier of the pedestrian to be identified In the case where the average value of the feature distances is the minimum value, it is determined that the identification of the pedestrian to be identified matches the pedestrian to be identified.
  • the identification module is further configured to increase a predetermined number of values if the average value of the feature distances corresponding to the identification of the pedestrian to be identified is not the minimum value, and determine whether the predetermined number of current values is greater than A threshold, in a case that the predetermined number of current values is not greater than the first threshold, the indication identifier extraction module performs an operation of extracting the historical pedestrian identifier corresponding to the predetermined number of feature distances in an order of the feature distance from small to large.
  • the identification module is further configured to assign a new historical pedestrian identification to the pedestrian to be identified if the predetermined number of current values is greater than the first threshold.
  • the pedestrian re-identification device further includes a feature deletion module configured to: after the feature distance calculation module calculates the feature distance of the pedestrian to be recognized and the historical pedestrian, the number of corresponding feature distances is counted for each historical pedestrian identification, Determining whether the statistical result exceeds the second threshold, and if the statistical result exceeds the second threshold, deleting the maximum feature distance in the corresponding feature distance, and then instructing the identifier extraction module to execute the predetermined number according to the feature distance from small to large. The operation of the corresponding pedestrian identification corresponding to the feature distance.
  • a pedestrian re-identification apparatus comprising: a memory configured to store an instruction; a processor coupled to the memory, the processor being configured to perform an implementation based on the instruction stored in the memory as described above A method related to an embodiment.
  • a computer readable storage medium stores computer instructions that, when executed by a processor, implement a method as recited in any of the above embodiments.
  • FIG. 1 is an exemplary flowchart of a pedestrian re-identification method according to an embodiment of the present disclosure
  • FIG. 2 is an exemplary flowchart of a pedestrian re-identification method according to another embodiment of the present disclosure
  • FIG. 3 is an exemplary block diagram of a pedestrian re-identification apparatus according to an embodiment of the present disclosure
  • FIG. 4 is an exemplary block diagram of a pedestrian re-identification apparatus according to another embodiment of the present disclosure.
  • FIG. 5 is an exemplary block diagram of a pedestrian re-identification apparatus according to still another embodiment of the present disclosure.
  • the inventors have found through research that in the related art, whether a pedestrian is a recurring pedestrian is determined by using feature extraction and feature distance.
  • the pedestrian feature to be identified in the current video frame is extracted, and the pedestrian appearing in the video frame within 2 minutes before the current video frame shooting time is taken as the historical pedestrian.
  • the characteristic distance between the pedestrian to be identified and each historical pedestrian it is determined according to the distance threshold whether the pedestrian to be identified is the same person as a certain one of the historical pedestrians.
  • the pedestrian to be identified detected from the current video frame is a1.
  • the history library there are 16 historical pedestrians appearing in the video frame within 2 minutes before the current video frame shooting time. Of the 16 people, 5 were identified as pedestrian 1 (corresponding ID 1), 5 were identified as pedestrian 2 (corresponding ID 2), and 6 were identified as pedestrian 3 (corresponding ID 3) ). It should be noted that since the camera performs continuous capture, the same pedestrian will appear in multiple video frames.
  • Table 1 shows the characteristic distances of the pedestrians a1 to be identified and the pedestrians 1 in the historical pedestrians.
  • Table 2 shows the characteristic distances of the pedestrians a1 to be identified and the pedestrians 2 in the historical pedestrians.
  • ID2 ID2 ID2 ID2 ID2 A1 19.9999 21.879 19.8341 24.1748 23.2484
  • Table 3 shows the characteristic distances of the pedestrians a1 to be identified and the pedestrians 3 in the historical pedestrians.
  • ID3 ID3 ID3 ID3 ID3 ID3 A1 2.97169 2.94814 4.43732 2.67215 15.59878 16.04216
  • the characteristic distance between the pedestrian a1 to be identified and each pedestrian 3 in the historical pedestrian is small. Due to factors such as posture, illumination, and shooting angle, the pedestrian feature distance between the pedestrian a1 to be identified and the pedestrian in the historical pedestrian is abrupt, as shown in Table 3. In this case, the average distance of the feature distance between the pedestrian a1 to be identified and the pedestrian in the historical pedestrian is:
  • the preset distance threshold is 7.0
  • the result is significantly larger than the distance threshold.
  • the average value of the feature distances of the pedestrian a1 to be recognized and the pedestrian 2 and the pedestrian 3 also exceeds the distance threshold. Therefore, the pedestrian a1 to be identified is regarded as a pedestrian different from the pedestrian 1, the pedestrian 2 and the pedestrian 3, in which case a new pedestrian identification is assigned to the pedestrian a1 to be identified, thereby causing the pedestrian to fail to recognize again. .
  • the present disclosure provides a pedestrian re-identification scheme capable of effectively suppressing external noise interference and improving the success rate of pedestrian recognition.
  • FIG. 1 is an exemplary flowchart of a pedestrian re-identification method according to an embodiment of the present disclosure.
  • the method steps of the present embodiment can be performed by a pedestrian re-identification device.
  • step 101 a pedestrian to be identified in the specified video frame is detected.
  • step 102 all pedestrians appearing in the captured video frame are extracted as historical pedestrians within a specified time period before the shooting time of the specified video frame.
  • the end time of the specified time period is the shooting time of the specified video frame.
  • the specified time period has a length of 2 minutes, and the end time of the specified time period is the shooting time of the specified video frame.
  • the feature distances of the pedestrian to be identified and the historical pedestrian are calculated.
  • step 104 the historical pedestrian identification corresponding to the predetermined number of feature distances is extracted in the order of the feature distances from small to large.
  • the history library there are 10 historical pedestrians with a logo of 6, and 6 historical pedestrians with a logo of 7.
  • the pedestrian is to be identified as a pedestrian 6. Since the average distances of other historical pedestrians in the historical library and the pedestrians to be identified are large, they are not discussed here.
  • the average value of the characteristic distance between the pedestrian to be identified and each pedestrian 6 is:
  • the average of the characteristic distances of the pedestrian to be identified 7 is:
  • the feature distance between the pedestrian to be recognized and the pedestrian 7 is smaller, and thus the pedestrian to be identified is re-identified as the pedestrian 7.
  • the present disclosure performs an overall analysis of feature distances.
  • the historical pedestrian identification corresponding to the first five feature distances is extracted from the feature distances of the pedestrians to be recognized and the pedestrians 6 and the pedestrians 7 in descending order of the feature distances.
  • the extracted historical pedestrian identifications are clustered to re-identify the pedestrians to be identified.
  • the extracted historical pedestrian identifications are clustered to group the same historical pedestrian identifications into the same cluster set. By counting the number of identifiers in each cluster set, the historical pedestrian identifier in the cluster set with the largest number of identifiers is used as the identifier of the pedestrian to be identified.
  • the historical pedestrian identification is divided into two sets by clustering processing.
  • the first set corresponds to historical pedestrian 6, with 3 historical pedestrian signs.
  • the second set corresponds to historical pedestrian 7, with 2 historical pedestrian signs.
  • the historical pedestrian identification 6 involved in the first set is assigned to the pedestrian to be identified. That is, through the above processing, it is determined that the identity of the pedestrian to be identified is the historical pedestrian 6.
  • the overall analysis of the feature distances of the pedestrians and the historical pedestrians is performed, thereby effectively suppressing external noise interference and improving the pedestrian recognition success rate.
  • Step 2 is an exemplary flowchart of a pedestrian re-identification method according to another embodiment of the present disclosure.
  • the method steps of the present embodiment are performed by a pedestrian re-identification device.
  • Steps 201-205 are the same as steps 101-105 in the above embodiment.
  • the pedestrian to be identified in the specified video frame is detected.
  • step 202 all pedestrians appearing in the captured video frame are extracted as historical pedestrians during a specified time period prior to the shooting time of the specified video frame.
  • the feature distances of the pedestrian to be identified and the historical pedestrian are calculated.
  • step 204 the historical pedestrian identification corresponding to the predetermined number of feature distances is extracted in the order of the feature distances from small to large.
  • the extracted historical pedestrian identifications are clustered to re-identify the pedestrians to be identified.
  • an average of the feature distances corresponding to each historical pedestrian identification is calculated over a predetermined number of feature distances.
  • step 207 based on the calculation result, it is determined whether the average value of the feature distances corresponding to the identification of the pedestrian to be identified is the minimum value.
  • step 208 is performed. If the average value of the feature distances corresponding to the identification of the pedestrian to be identified is not the minimum value, step 209 is performed.
  • step 208 it is determined that the identity of the pedestrian to be identified matches the pedestrian to be identified, and the confirmation indicates that the pedestrian recognition is successful.
  • the average of the three feature distances corresponding to the historical pedestrian identification 6 is:
  • the average of the two feature distances corresponding to the historical pedestrian identification 7 is:
  • the identifier assigned to the pedestrian to be identified is also the identifier 6. This indicates that pedestrian recognition is successful.
  • step 209 a predetermined number of values are incremented.
  • step 210 it is determined whether the predetermined number of current values is greater than the first threshold.
  • step 204 is performed. If the predetermined number of current values is greater than the first threshold, step 211 is performed.
  • a new historical pedestrian identification is assigned to the pedestrian to be identified.
  • the historical pedestrian identification assigned to the pedestrian to be identified is the identification 6, the historical pedestrian identification 7 corresponds to the smallest average value of the feature distance. That means that pedestrian redistribution is not successful.
  • N the value of the predetermined number N, more history samples can be used in the process of re-recognition processing. For example, the value of N can be doubled.
  • the pedestrian to be identified is not included in the historical pedestrian. In this case, the pedestrian to be identified is assigned a new historical pedestrian identity.
  • step 203 the number of corresponding feature distances is counted for each historical pedestrian identification. Determine if the statistical result exceeds the second threshold. If the statistical result exceeds the second threshold, the maximum feature distance of the corresponding feature distance is deleted, and then step 204 is performed. If the statistical result does not exceed the second threshold, the pedestrian distance recognition process is directly performed using the obtained feature distance.
  • a pedestrian is in a certain area, so there will be more information in the history library with the same historical pedestrian identification in a certain period of time. Due to factors such as pedestrian attitude, illumination, and shooting angle, there is a case where the characteristic distance deviation is large. By deleting the maximum feature distance in the corresponding feature distance, the external noise can be effectively filtered.
  • the pedestrian to be identified is a historical pedestrian 6, there are 21 pieces of information in the history library corresponding to the historical pedestrian identification 6, exceeding a predetermined threshold (for example, a predetermined threshold of 20).
  • a predetermined threshold for example, a predetermined threshold of 20.
  • the feature distance having the largest value is deleted, and the remaining 20 feature distances are used for corresponding processing. This effectively eliminates external noise interference.
  • FIG. 3 is an exemplary block diagram of a pedestrian re-identification apparatus according to an embodiment of the present disclosure.
  • the pedestrian re-identification device includes a detection module 31, a historical pedestrian extraction module 32, a feature distance calculation module 33, an identification extraction module 34, and an identification module 35.
  • the detection module 31 is configured to detect a pedestrian to be identified in a specified video frame.
  • the historical pedestrian extraction module 32 is configured to extract all pedestrians appearing in the captured video frame as historical pedestrians within a specified time period prior to the shooting time of the specified video frame.
  • the end time of the specified time period is the shooting time of the specified video frame.
  • the specified time period is 2 minutes, and the end time of the specified time period is the shooting time of the specified video frame.
  • the feature distance calculation module 33 is configured to calculate a feature distance of the pedestrian to be identified and the historical pedestrian.
  • the identifier extraction module 34 is configured to extract historical pedestrian identifiers corresponding to a predetermined number of feature distances in descending order of feature distances.
  • the identification module 35 is configured to cluster the extracted historical pedestrian identifications in order to re-identify the pedestrians to be identified.
  • the identification module 35 is configured to cluster the extracted historical pedestrian identifications to group the same historical pedestrian identifications into the same cluster set, and to count the number of identifications in each cluster set.
  • the historical pedestrian identification in the cluster set with the largest number of identifications is used as the identifier of the pedestrian to be identified.
  • historical pedestrian identification is divided into 2 sets by clustering processing.
  • the first set corresponds to historical pedestrian 6, with 3 historical pedestrian signs.
  • the second set corresponds to historical pedestrian 7, with 2 historical pedestrian signs.
  • the historical pedestrian identification 6 involved in the first set is assigned to the pedestrian to be identified. That is, it is determined that the identity of the pedestrian to be identified is historical pedestrian 6.
  • the pedestrian re-identification device provided by the above embodiment of the present disclosure, by analyzing the feature distance of the pedestrian and the historical pedestrian, the external noise interference can be effectively suppressed, and the pedestrian recognition success rate can be improved.
  • FIG. 4 is an exemplary block diagram of a pedestrian re-identification apparatus according to another embodiment of the present disclosure. 4 differs from FIG. 3 in that, in the embodiment shown in FIG. 4, the pedestrian re-identification device further includes an average value calculation module 36.
  • the average value calculation module 36 is configured to calculate an average value of the feature distances corresponding to each historical pedestrian identification among the predetermined number of feature distances after the identification module 35 identifies the pedestrians to be identified according to the clustering result. .
  • the identification module 35 is further configured to determine, according to the calculation result of the average value calculation module 36, whether the average value of the feature distances corresponding to the identification of the pedestrian to be identified is a minimum value, and the feature distance corresponding to the identification of the pedestrian to be identified In the case where the average value is the minimum value, it is determined that the identification of the pedestrian to be identified matches the pedestrian to be identified.
  • the average of the three feature distances corresponding to the historical pedestrian identification 6 is 2.08278, and the average of the two feature distances corresponding to the historical pedestrian identification 7 is 3.054895. Since the average distance of the feature distance corresponding to the historical pedestrian identification 6 is the smallest, the identifier assigned to the pedestrian to be identified is also the identifier 6. This indicates that pedestrian recognition is successful.
  • the identification module 35 is further configured to increase a predetermined number of values if the average of the feature distances corresponding to the identification of the pedestrian to be identified is not the minimum value, and determine whether the predetermined number of current values is greater than The first threshold, in the case that the predetermined number of current values is not greater than the first threshold, the indication identifier extraction module 34 performs an operation of extracting the historical pedestrian identification corresponding to the predetermined number of feature distances in descending order of the feature distance.
  • the identification module 35 is further configured to assign a new historical pedestrian identification to the pedestrian to be identified if the predetermined number of current values is greater than the first threshold.
  • the predetermined number of values is further expanded to select more historical samples for identification. If the sample size is expanded, if the number of samples exceeds the first threshold, the pedestrian recognition cannot be successfully achieved, indicating that the pedestrian to be identified is not included in the historical pedestrian. In this case, the pedestrian to be identified is assigned a new historical pedestrian identity.
  • the pedestrian re-identification device further includes a feature deletion module 37.
  • the feature deletion module 37 is configured to calculate the number of corresponding feature distances for each historical pedestrian identification after the feature distance calculation module 33 calculates the feature distance of the pedestrian to be recognized by the historical pedestrian, and determine whether the statistical result exceeds the second threshold. If the result exceeds the second threshold, the maximum feature distance of the corresponding feature distance is deleted, and then the indication extraction module 34 performs an operation of extracting the historical pedestrian identification corresponding to the predetermined number of feature distances according to the order of the feature distance from small to large. .
  • a pedestrian is in a certain area, so there will be more information in the history library with the same historical pedestrian identification in a certain period of time. Due to factors such as pedestrian attitude, illumination, and shooting angle, there is a case where the characteristic distance deviation is large. By deleting the maximum feature distance in the corresponding feature distance, the external noise can be effectively filtered.
  • the pedestrian to be identified is a historical pedestrian 6, there are 21 pieces of information in the history library corresponding to the historical pedestrian identification 6, exceeding a predetermined threshold (for example, a predetermined threshold of 20).
  • a predetermined threshold for example, a predetermined threshold of 20.
  • the feature distance having the largest value is deleted, and the remaining 20 feature distances are used for corresponding processing. This effectively eliminates external noise interference.
  • FIG. 5 is an exemplary block diagram of a pedestrian re-identification apparatus according to still another embodiment of the present disclosure.
  • the pedestrian re-identification means includes a memory 51 and a processor 52.
  • Memory 51 is used to store instructions
  • processor 52 is coupled to memory 51
  • processor 52 is configured to perform the methods involved in any of the embodiments of FIG. 1 or FIG. 2 based on instructions stored in the memory.
  • the pedestrian re-identification device further includes a communication interface 53 for performing information interaction with other devices.
  • the apparatus further includes a bus 54, the processor 52, the communication interface 53, and the memory 51 accomplishing communication with each other via the bus 54.
  • the memory 51 may include a high speed RAM memory, and may also include a non-volatile memory such as at least one disk memory.
  • the memory 51 can also be a memory array.
  • the memory 51 may also be partitioned, and the blocks may be combined into a virtual volume according to certain rules.
  • processor 52 can be a central processing unit CPU, or can be an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present disclosure.
  • the present disclosure also relates to a computer readable storage medium storing computer instructions that, when executed by a processor, implement the method of any of the embodiments of FIG. 1 or FIG.
  • the functional unit modules described above may be implemented as a general purpose processor, a Programmable Logic Controller (PLC), a digital signal processor (for example) for performing the functions described in this disclosure ( Digital Signal Processor (DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistors Logic device, discrete hardware component, or any suitable combination thereof.
  • PLC Programmable Logic Controller
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
  • the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Human Computer Interaction (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Traffic Control Systems (AREA)
  • Image Analysis (AREA)

Abstract

本公开提供一种行人重识别方法和装置。行人重识别装置检测出指定视频帧中的待识别行人，在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人，计算待识别行人与历史行人的特征距离，按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识，对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。本公开能够有效抑制外界噪声干扰、提高行人重识别成功率。

Description

行人重识别方法和装置
相关申请的交叉引用
本申请是以CN申请号为201810383810.8，申请日为2018年4月26日的申请为基础，并主张其优先权，该CN申请的公开内容在此作为整体引入本申请中。
技术领域
本公开涉及图像识别领域，特别涉及一种行人重识别方法和装置。
背景技术
行人重识别(Pedestrian Re-Identification，简称：Re-ID)技术，是通过利用同一摄像头在不同时间段采集的同一行人的图像，以及不同摄像头采集到的同一行人的图像，对行人轨迹进行跟踪。
发明内容
发明人通过研究发现，在实际场景下，行人的姿态、光照条件和拍摄角度等因素都会对基于行人外貌特征的行人重识别方案造成较大影响，最终导致行人重识别失败。
为此，本公开提供一种能够有效抑制外界噪声干扰、提高行人重识别成功率的行人重识别方案。
根据本公开实施例的第一方面，提供一种行人重识别方法，包括：检测出指定视频帧中的待识别行人；在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人，其中指定时间段的结束时间为指定视频帧的拍摄时间；计算待识别行人与历史行人的特征距离；按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识；对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。
在一些实施例中，对提取出的历史行人标识进行聚类包括：对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中；统计每个聚类集合中的标识数量；将标识数量最多的聚类集合中的历史行人标识作为待识别行人的标识。
在一些实施例中，在对待识别行人进行重识别后，还包括：在预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值；根据计算结果，判断与待识别行人的标识相对应的特征距离的平均值是否为最小值；在与待识别行人的标识相对应的特征距离的平均值为最小值的情况下，判定待识别行人的标识与待识别行人相匹配。
在一些实施例中，在与待识别行人的标识相对应的特征距离的平均值不是最小值的情况下，增大预定数量的数值；判断预定数量的当前值是否大于第一门限；在预定数量的当前值不大于第一门限的情况下，执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的步骤。
在一些实施例中，在预定数量的当前值大于第一门限的情况下，给待识别行人分配新的历史行人标识。
在一些实施例中，在计算待识别行人与历史行人的特征距离后，还包括：对于每个历史行人标识，统计对应特征距离的数量；判断统计结果是否超过第二门限；在统计结果超过第二门限的情况下，将对应特征距离中的最大特征距离删除；然后执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的步骤。
根据本公开实施例的第二方面，提供一种行人重识别装置，包括：检测模块，被配置为检测出指定视频帧中的待识别行人；历史行人提取模块，被配置为在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人，其中指定时间段的结束时间为指定视频帧的拍摄时间；特征距离计算模块，被配置为计算待识别行人与历史行人的特征距离；标识提取模块，被配置为按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识；识别模块，被配置为对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。
在一些实施例中，识别模块被配置为对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中，统计每个聚类集合中的标识数量，将标识数量最多的聚类集合中的历史行人标识作为待识别行人的标识。
在一些实施例中，行人重识别装置还包括平均值计算模块，被配置为在识别模块根据聚类结果对待识别行人进行识别后，在预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值；识别模块还被配置为根据平均值计算模块的计算结果，判断与待识别行人的标识相对应的特征距离的平均值是否为最小值，在与待识别 行人的标识相对应的特征距离的平均值为最小值的情况下，判定待识别行人的标识与待识别行人相匹配。
在一些实施例中，识别模块还被配置为在与待识别行人的标识相对应的特征距离的平均值不是最小值的情况下，增大预定数量的数值，判断预定数量的当前值是否大于第一门限，在预定数量的当前值不大于第一门限的情况下，指示标识提取模块执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
在一些实施例中，识别模块还被配置为在预定数量的当前值大于第一门限的情况下，给待识别行人分配新的历史行人标识。
在一些实施例中，行人重识别装置还包括特征删除模块，被配置为在特征距离计算模块计算待识别行人与历史行人的特征距离后，对于每个历史行人标识，统计对应特征距离的数量，判断统计结果是否超过第二门限，在统计结果超过第二门限的情况下，将对应特征距离中的最大特征距离删除，然后指示标识提取模块执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
根据本公开实施例的第三方面，提供一种行人重识别装置，包括：存储器，被配置为存储指令；处理器，耦合到存储器，处理器被配置为基于存储器存储的指令执行实现如上述任一实施例涉及的方法。
根据本公开实施例的第四方面，提供一种计算机可读存储介质，其中，计算机可读存储介质存储有计算机指令，指令被处理器执行时实现如上述任一实施例涉及的方法。
通过以下参照附图对本公开的示例性实施例的详细描述，本公开的其它特征及其优点将会变得清楚。
附图说明
为了更清楚地说明本公开实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，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性的前提下，还可以根据这些附图获得其他的附图。
图1为本公开一个实施例的行人重识别方法的示例性流程图；
图2为本公开另一个实施例的行人重识别方法的示例性流程图；
图3为本公开一个实施例的行人重识别装置的示例性框图；
图4为本公开另一个实施例的行人重识别装置的示例性框图；
图5为本公开又一个实施例的行人重识别装置的示例性框图。
具体实施方式
下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。以下对至少一个示例性实施例的描述实际上仅仅是说明性的，决不作为对本公开及其应用或使用的任何限制。基于本公开中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
除非另外具体说明，否则在这些实施例中阐述的部件和步骤的相对布置、数字表达式和数值不限制本公开的范围。
同时，应当明白，为了便于描述，附图中所示出的各个部分的尺寸并不是按照实际的比例关系绘制的。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论，但在适当情况下，所述技术、方法和设备应当被视为授权说明书的一部分。
在这里示出和讨论的所有示例中，任何具体值应被解释为仅仅是示例性的，而不是作为限制。因此，示例性实施例的其它示例可以具有不同的值。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步讨论。
发明人通过研究发现，在相关技术中，通过使用特征提取和特征距离来判定一个行人是否为重复出现的行人。首先，将当前视频帧中的待识别行人特征提取出来，再将当前视频帧拍摄时间之前的2分钟内的视频帧中出现的行人作为历史行人。通过计算待识别行人与每个历史行人的特征距离，根据距离阈值来判断待识别行人是否与历史行人中的某个人是同一个人。
例如，设从当前视频帧中检测到的待识别行人为a1。在历史库中，在当前视频帧拍摄时间之前的2分钟内的视频帧中出现的历史行人共有16人。在这16人中，有5人被识别为行人1(相应的标识为ID1)，5人被识别为行人2(相应的标识为ID2)，6人被识别为行人3(相应的标识为ID3)。需要说明的是，由于摄像头会进行连续抓拍，因此同一行人会出现在多个视频帧中。
表1为待识别行人a1与历史行人中各行人1的特征距离。
  ID1 ID1 ID1 ID1 ID1
a1 7.02946 6.72801 7.04967 7.2829 9.28402
表1
表2为待识别行人a1与历史行人中各行人2的特征距离。
  ID2 ID2 ID2 ID2 ID2
a1 19.9999 21.879 19.8341 24.1748 23.2484
表2
表3为待识别行人a1与历史行人中各行人3的特征距离。
  ID3 ID3 ID3 ID3 ID3 ID3
a1 2.97169 2.94814 4.43732 2.67215 15.59878 16.04216
表3
由于待识别行人a1实质上就是行人3，因此待识别行人a1与历史行人中各行人3的特征距离较小。由于姿态、光照、拍摄角度等因素变化的原因，导致待识别行人a1与历史行人中标记为行人3的行人特征距离会发生突变，如表3所示。在这种情况下，待识别行人a1与历史行人中标记为行人3的特征距离平均值为：
(2.9716+2.94814+4.43732+2.67215+15.59878+16.04216)/6＝7.445
若预先设置的距离阈值为7.0，则这个结果明显大于该距离阈值。此外，待识别行人a1与行人2和行人3的特征距离平均值也超过该距离阈值。由此，会将待识别行人a1视为是与行人1、行人2和行人3均不相同的行人，在这种情况下会给待识别行人a1分配新的行人标识，从而导致行人重识别失败。
为此，本公开提供一种能够有效抑制外界噪声干扰、提高行人重识别成功率的行人重识别方案。
图1为本公开一个实施例的行人重识别方法的示例性流程图。在一些实施例中，本实施例的方法步骤可由行人重识别装置执行。
在步骤101，检测出指定视频帧中的待识别行人。
在步骤102，在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中 出现的全部行人以作为历史行人。
这里需要说明的是，指定时间段的结束时间为指定视频帧的拍摄时间。例如，指定时间段的长度为2分钟，该指定时间段的结束时间即为指定视频帧的拍摄时间。
在步骤103，计算待识别行人与历史行人的特征距离。
在步骤104，按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识。
例如，在历史库中，标识为6的历史行人有10人，标识为7的历史行人有6人。在该实施例中，设待识别行人为行人6。由于历史库中的其他历史行人与待识别行人的特征距离的平均值较大，因此这里不再讨论。
待识别行人与各行人6的特征距离的平均值为：
(7.18478+6.57366+6.71023+5.65011+4.6309+3.73646+5.03583+3.01242+1.57515
+1.66077)/10＝4.57703
待识别行人与行人7的特征距离的平均值为：
(3.7303+2.37949+3.94587+4.21519+5.34866+6.1209)/6＝4.29007
由于特征受到干扰，待识别行人与行人7的特征距离更小，因此会将待识别行人重识别为行人7。
为此，本公开对特征距离进行整体分析。
在一些实施例中，如表4所示，按照特征距离从小到大的顺序，从待识别行人与各行人6和各行人7的特征距离中，提取出前5个特征距离对应的历史行人标识。
特征距离值 历史行人ID
1.57515 ID6
1.66077 ID6
2.37949 ID7
3.01242 ID6
3.7303 ID7
表4
在步骤105，对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。
在一些实施例中，通过对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中。通过统计每个聚类集合中的标识数量，将标识数量最多的聚类 集合中的历史行人标识作为待识别行人的标识。
如表4所示，通过聚类处理，将历史行人标识分为2个集合。第一个集合对应历史行人6，有3个历史行人标识。第二个集合对应历史行人7，有2个历史行人标识。在这种情况下，将第一个集合中涉及的历史行人标识6分配给待识别行人。即，通过上述处理，确定待识别行人的身份为历史行人6。
在本公开上述实施例提供的行人重识别方法中，通过对待识别行人和历史行人的特征距离进行整体分析，从而能够有效抑制外界噪声干扰，提高行人重识别成功率。
图2为本公开另一个实施例的行人重识别方法的示例性流程图。在一些实施例中，本实施例的方法步骤由行人重识别装置执行。步骤201-205与上述实施例中的步骤101-105相同。
在步骤201，检测出指定视频帧中的待识别行人。
在步骤202，在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人。
在步骤203，计算待识别行人与历史行人的特征距离。
在步骤204，按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识。
在步骤205，对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。
在步骤206，在预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值。
在步骤207，根据计算结果，判断与待识别行人的标识相对应的特征距离的平均值是否为最小值。
若与待识别行人的标识相对应的特征距离的平均值为最小值，执行步骤208。若与待识别行人的标识相对应的特征距离的平均值不是最小值，执行步骤209。
在步骤208，判定待识别行人的标识与待识别行人相匹配，确认表明行人重识别成功。
如表4所示，与历史行人标识6相对应的3个特征距离平均值为：
(1.57515+1.66077+3.01242)/3＝2.08278
与历史行人标识7相对应的2个特征距离平均值为：
(2.37949+3.7303)/2＝3.054895
由于历史行人标识6对应的特征距离平均值最小，而分配给待识别行人的标识也为标识6。由此表明行人重识别成功。
在步骤209，增大预定数量的数值。
在步骤210，判断预定数量的当前值是否大于第一门限。
若预定数量的当前值不大于第一门限的情况下，执行步骤204。若预定数量的当前值大于第一门限，则执行步骤211。
在步骤211，给待识别行人分配新的历史行人标识。
在上述实施例中，若分配给待识别行人的历史行人标识为标识6，而历史行人标识7对应的特征距离平均值最小。即表明行人重分配并不成功。在这种情况下，通过扩大预定数量N的数值，以便在重新进行识别处理的过程中，能够使用更多的历史样本。例如，可将N值扩大一倍。
在上述实施例中，若通过调整预定数量N，在预定数量N超过第一门限的情况下，仍无法得到成功的行人重识别结果，则会认为该待识别行人未包括在历史行人中。在这种情况下，会给该待识别行人分配新的历史行人标识。
在一些实施例中，在步骤203后，对于每个历史行人标识，统计对应特征距离的数量。判断统计结果是否超过第二门限。在统计结果超过第二门限的情况下，将对应特征距离中的最大特征距离删除，然后再执行步骤204。若统计结果未超过第二门限的情况下，则直接利用得到的特征距离进行行人重识别处理。
例如，行人在某个区域徘徊，因此在某个时间段内，在历史库中具有同一历史行人标识的信息会较多。由于行人姿态、光照、拍摄角度等因素不同，因此会存在特征距离偏差较大的情况。通过将对应特征距离中的最大特征距离删除，可有效过滤外界噪声。
在一些实施例中，若待识别行人为历史行人6，历史库中与历史行人标识6相对应的信息有21条，超过预定门限(例如，预定门限为20条)。在这种情况下，在待识别行人与历史行人标识6的特征距离中，将数值最大的特征距离删除，利用剩余的20个特征距离进行相应处理。由此有效消除外界噪声干扰。
图3为本公开一个实施例的行人重识别装置的示例性框图。如图3所示，行人重识别装置包括检测模块31、历史行人提取模块32、特征距离计算模块33、标识提取模块34和识别模块35。
如图3所示，检测模块31被配置为检测出指定视频帧中的待识别行人。
历史行人提取模块32被配置为在指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人。指定时间段的结束时间为指定视频帧的拍摄时间。
在一些实施例中，指定时间段为2分钟，则该指定时间段的结束时间即为指定视频帧的拍摄时间。
特征距离计算模块33被配置为计算待识别行人与历史行人的特征距离。
标识提取模块34被配置为按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识。
识别模块35被配置为对提取出的历史行人标识进行聚类，以便对待识别行人进行重识别。
在一些实施例中，识别模块35被配置为对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中，统计每个聚类集合中的标识数量，将标识数量最多的聚类集合中的历史行人标识作为待识别行人的标识。
在一些实施例中，如表4所示，通过聚类处理，将历史行人标识分为2个集合。第一个集合对应历史行人6，有3个历史行人标识。第二个集合对应历史行人7，有2个历史行人标识。在这种情况下，将第一个集合中涉及的历史行人标识6分配给待识别行人。即确定待识别行人的身份为历史行人6。
在本公开上述实施例提供的行人重识别装置中，通过对待识别行人和历史行人的特征距离进行整体分析，从而能够有效抑制外界噪声干扰，提高行人重识别成功率。
图4为本公开另一个实施例的行人重识别装置的示例性框图。图4与图3的不同之处在于，在图4所示实施例中，行人重识别装置还包括平均值计算模块36。
如图4所示，平均值计算模块36被配置为在识别模块35根据聚类结果对待识别行人进行识别后，在预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值。
识别模块35还被配置为根据平均值计算模块36的计算结果，判断与待识别行人的标识相对应的特征距离的平均值是否为最小值，在与待识别行人的标识相对应的特征距离的平均值为最小值的情况下，判定待识别行人的标识与待识别行人相匹配。
在一些实施例中，如表4所示，与历史行人标识6相对应的3个特征距离平均值为2.08278，与历史行人标识7相对应的2个特征距离平均值为3.054895。由于历史行人标识6对应的特征距离平均值最小，而分配给待识别行人的标识也为标识6。由此表明行人重识别成功。
在一些实施例中，识别模块35还被配置为在与待识别行人的标识相对应的特征距离的平均值不是最小值的情况下，增大预定数量的数值，判断预定数量的当前值是否大于 第一门限，在预定数量的当前值不大于第一门限的情况下，指示标识提取模块34执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
在一些实施例中，识别模块35还被配置为在预定数量的当前值大于第一门限的情况下，给待识别行人分配新的历史行人标识。
例如，若行人重识别不成功，则进一步扩大预定数量的数值，以便选取更多的历史样本进行识别。若扩大样本规模，在样本数量超过第一门限的情况下，仍无法成功实现行人重识别，则表明该待识别行人未包括在历史行人中。在这种情况下，会给该待识别行人分配新的历史行人标识。
在一些实施例中，如图4所示，行人重识别装置还包括特征删除模块37。
特征删除模块37被配置为在特征距离计算模块33计算待识别行人与历史行人的特征距离后，对于每个历史行人标识，统计对应特征距离的数量，判断统计结果是否超过第二门限，在统计结果超过第二门限的情况下，将对应特征距离中的最大特征距离删除，然后指示标识提取模块34执行按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
例如，行人在某个区域徘徊，因此在某个时间段内，在历史库中具有同一历史行人标识的信息会较多。由于行人姿态、光照、拍摄角度等因素不同，因此会存在特征距离偏差较大的情况。通过将对应特征距离中的最大特征距离删除，可有效过滤外界噪声。
在一些实施例中，若待识别行人为历史行人6，历史库中与历史行人标识6相对应的信息有21条，超过预定门限(例如，预定门限为20条)。在这种情况下，在待识别行人与历史行人标识6的特征距离中，将数值最大的特征距离删除，利用剩余的20个特征距离进行相应处理。由此有效消除外界噪声干扰。
图5为本公开又一个实施例的行人重识别装置的示例性框图。如图5所示，行人重识别装置包括存储器51和处理器52。
存储器51用于存储指令，处理器52耦合到存储器51，处理器52被配置为基于存储器存储的指令执行实现如图1或图2中任一实施例涉及的方法。
如图5所示，该行人重识别装置还包括通信接口53，用于与其它设备进行信息交互。同时，该装置还包括总线54，处理器52、通信接口53、以及存储器51通过总线54完成相互间的通信。
存储器51可以包含高速RAM存储器，也可还包括非易失性存储器(non-volatile memory)，例如至少一个磁盘存储器。存储器51也可以是存储器阵列。存储器51还可能 被分块，并且块可按一定的规则组合成虚拟卷。
此外，处理器52可以是一个中央处理器CPU，或者可以是专用集成电路ASIC，或者是被配置成实施本公开实施例的一个或多个集成电路。
本公开同时还涉及一种计算机可读存储介质，其中计算机可读存储介质存储有计算机指令，指令被处理器执行时实现如图1或图2中任一实施例涉及的方法。
在一些实施例中，在上面所描述的功能单元模块可以实现为用于执行本公开所描述功能的通用处理器、可编程逻辑控制器(Programmable Logic Controller，简称：PLC)、数字信号处理器(Digital Signal Processor，简称：DSP)、专用集成电路(Application Specific Integrated Circuit，简称：ASIC)、现场可编程门阵列(Field-Programmable Gate Array，简称：FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件或者其任意适当组合。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。
本公开的描述是为了示例和描述起见而给出的，而并不是无遗漏的或者将本公开限于所公开的形式。很多修改和变化对于本领域的普通技术人员而言是显然的。选择和描述实施例是为了更好说明本公开的原理和实际应用，并且使本领域的普通技术人员能够理解本公开从而设计适于特定用途的带有各种修改的各种实施例。

Claims (14)

  1. 一种行人重识别方法，包括：
    检测出指定视频帧中的待识别行人；
    在所述指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人，其中所述指定时间段的结束时间为所述指定视频帧的拍摄时间；
    计算所述待识别行人与所述历史行人的特征距离；
    按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识；
    对提取出的历史行人标识进行聚类，以便对所述待识别行人进行重识别。
  2. 根据权利要求1所述的行人重识别方法，对提取出的历史行人标识进行聚类包括：
    对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中；
    统计每个聚类集合中的标识数量；
    将标识数量最多的聚类集合中的历史行人标识作为所述待识别行人的标识。
  3. 根据权利要求2所述的行人重识别方法，在对所述待识别行人进行重识别后，还包括：
    在所述预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值；
    根据计算结果，判断与所述待识别行人的标识相对应的特征距离的平均值是否为最小值；
    在与所述待识别行人的标识相对应的特征距离的平均值为最小值的情况下，判定所述待识别行人的标识与所述待识别行人相匹配。
  4. 根据权利要求3所述的行人重识别方法，还包括，
    在与所述待识别行人的标识相对应的特征距离的平均值不是最小值的情况下，增 大所述预定数量的数值；
    判断所述预定数量的当前值是否大于第一门限；
    在所述预定数量的当前值不大于第一门限的情况下，执行所述按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的步骤。
  5. 根据权利要求4所述的行人重识别方法，还包括，
    在所述预定数量的当前值大于第一门限的情况下，给所述待识别行人分配新的历史行人标识。
  6. 根据权利要求1-5中任一项所述的行人重识别方法，在计算所述待识别行人与所述历史行人的特征距离后，还包括：
    对于每个历史行人标识，统计对应特征距离的数量；
    判断统计结果是否超过第二门限；
    在统计结果超过第二门限的情况下，将所述对应特征距离中的最大特征距离删除；
    然后执行所述按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的步骤。
  7. 一种行人重识别装置，包括：
    检测模块，被配置为检测出指定视频帧中的待识别行人；
    历史行人提取模块，被配置为在所述指定视频帧的拍摄时间之前的指定时间段内，提取出所拍摄视频帧中出现的全部行人以作为历史行人，其中所述指定时间段的结束时间为所述指定视频帧的拍摄时间；
    特征距离计算模块，被配置为计算所述待识别行人与所述历史行人的特征距离；
    标识提取模块，被配置为按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识；
    识别模块，被配置为对提取出的历史行人标识进行聚类，以便对所述待识别行人进行重识别。
  8. 根据权利要求7所述的行人重识别装置，其中，
    识别模块被配置为对提取出的历史行人标识进行聚类，以便将相同的历史行人标识划归到同一聚类集合中，统计每个聚类集合中的标识数量，将标识数量最多的聚类集合中的历史行人标识作为所述待识别行人的标识。
  9. 根据权利要求8所述的行人重识别装置，还包括：
    平均值计算模块，被配置为在识别模块根据聚类结果对所述待识别行人进行识别后，在所述预定数量个特征距离中，计算每个历史行人标识所对应特征距离的平均值；
    识别模块还被配置为根据平均值计算模块的计算结果，判断与所述待识别行人的标识相对应的特征距离的平均值是否为最小值，在与所述待识别行人的标识相对应的特征距离的平均值为最小值的情况下，判定所述待识别行人的标识与所述待识别行人相匹配。
  10. 根据权利要求9所述的行人重识别装置，其中，
    识别模块还被配置为在与所述待识别行人的标识相对应的特征距离的平均值不是最小值的情况下，增大所述预定数量的数值，判断所述预定数量的当前值是否大于第一门限，在所述预定数量的当前值不大于第一门限的情况下，指示标识提取模块执行所述按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
  11. 根据权利要求10所述的行人重识别装置，其中，
    识别模块还被配置为在所述预定数量的当前值大于第一门限的情况下，给所述待识别行人分配新的历史行人标识。
  12. 根据权利要求7-11中任一项所述的行人重识别装置，还包括：
    特征删除模块，被配置为在特征距离计算模块计算所述待识别行人与所述历史行人的特征距离后，对于每个历史行人标识，统计对应特征距离的数量，判断统计结果是否超过第二门限，在统计结果超过第二门限的情况下，将所述对应特征距离中的最大特征距离删除，然后指示标识提取模块执行所述按照特征距离从小到大的顺序，提取出预定数量个特征距离对应的历史行人标识的操作。
  13. 一种行人重识别装置，包括：
    存储器，被配置为存储指令；
    处理器，耦合到存储器，处理器被配置为基于存储器存储的指令执行实现如权利要求1-6中任一项的方法。
  14. 一种计算机可读存储介质，其中，计算机可读存储介质存储有计算机指令，指令被处理器执行时实现如权利要求1-6中任一项的方法。
PCT/CN2019/084489 2018-04-26 2019-04-26 行人重识别方法和装置 WO2019206265A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP19792513.4A EP3761221B1 (en) 2018-04-26 2019-04-26 Pedestrian re-identification method and apparatus
US17/044,578 US11804059B2 (en) 2018-04-26 2019-04-26 Pedestrian re-identification method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810383810.8A CN110414294B (zh) 2018-04-26 2018-04-26 行人重识别方法和装置
CN201810383810.8 2018-04-26

Publications (1)

Publication Number Publication Date
WO2019206265A1 true WO2019206265A1 (zh) 2019-10-31

Family

ID=68294845

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/084489 WO2019206265A1 (zh) 2018-04-26 2019-04-26 行人重识别方法和装置

Country Status (4)

Country Link
US (1) US11804059B2 (zh)
EP (1) EP3761221B1 (zh)
CN (1) CN110414294B (zh)
WO (1) WO2019206265A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110874576A (zh) * 2019-11-14 2020-03-10 西安工程大学 一种基于典型相关分析融合特征的行人再识别方法
CN112052819A (zh) * 2020-09-15 2020-12-08 浙江智慧视频安防创新中心有限公司 一种行人重识别方法、装置、设备及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368115B (zh) * 2020-03-03 2023-09-29 杭州海康威视系统技术有限公司 数据聚类方法、装置、聚类服务器及存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107153824A (zh) * 2017-05-22 2017-09-12 中国人民解放军国防科学技术大学 基于图聚类的跨视频行人重识别方法
CN107153817A (zh) * 2017-04-29 2017-09-12 深圳市深网视界科技有限公司 行人重识别数据标注方法和装置
US20180060647A1 (en) * 2016-08-31 2018-03-01 Fuji Xerox Co., Ltd. Image processing apparatus, non-transitory computer readable medium, and image processing method

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10664705B2 (en) 2014-09-26 2020-05-26 Nec Corporation Object tracking apparatus, object tracking system, object tracking method, display control device, object detection device, and computer-readable medium
AU2014277853A1 (en) * 2014-12-22 2016-07-07 Canon Kabushiki Kaisha Object re-identification using self-dissimilarity
CN105160319B (zh) * 2015-08-31 2018-10-16 电子科技大学 一种在监控视频下实现行人再识别的方法
CN107346409B (zh) * 2016-05-05 2019-12-17 华为技术有限公司 行人再识别方法和装置
US10275683B2 (en) * 2017-01-19 2019-04-30 Cisco Technology, Inc. Clustering-based person re-identification

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180060647A1 (en) * 2016-08-31 2018-03-01 Fuji Xerox Co., Ltd. Image processing apparatus, non-transitory computer readable medium, and image processing method
CN107153817A (zh) * 2017-04-29 2017-09-12 深圳市深网视界科技有限公司 行人重识别数据标注方法和装置
CN107153824A (zh) * 2017-05-22 2017-09-12 中国人民解放军国防科学技术大学 基于图聚类的跨视频行人重识别方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110874576A (zh) * 2019-11-14 2020-03-10 西安工程大学 一种基于典型相关分析融合特征的行人再识别方法
CN112052819A (zh) * 2020-09-15 2020-12-08 浙江智慧视频安防创新中心有限公司 一种行人重识别方法、装置、设备及存储介质

Also Published As

Publication number Publication date
US11804059B2 (en) 2023-10-31
EP3761221C0 (en) 2023-12-13
EP3761221A1 (en) 2021-01-06
US20210201009A1 (en) 2021-07-01
EP3761221B1 (en) 2023-12-13
EP3761221A4 (en) 2021-12-22
CN110414294B (zh) 2023-09-26
CN110414294A (zh) 2019-11-05

Similar Documents

Publication Publication Date Title
WO2019206265A1 (zh) 行人重识别方法和装置
CN108154171B (zh) 一种人物识别方法、装置及电子设备
CN109635686B (zh) 结合人脸与外观的两阶段行人搜索方法
CN105631418B (zh) 一种人数统计的方法和装置
Rivlin et al. A real-time system for classification of moving objects
CN109426785B (zh) 一种人体目标身份识别方法及装置
Li et al. Shot boundary detection based on multilevel difference of colour histograms
WO2016015621A1 (zh) 人脸图片人名识别方法和系统
KR101508310B1 (ko) 영상 감시 시스템에서 다중 이동 물체를 추적하는 방법 및 장치
CN106295532B (zh) 一种视频图像中的人体动作识别方法
US11132577B2 (en) System and a method for efficient image recognition
TWI704505B (zh) 人臉辨識系統、建立人臉辨識之資料之方法及其人臉辨識之方法
JP7311310B2 (ja) 情報処理装置、情報処理方法及びプログラム
WO2014193220A2 (en) System and method for multiple license plates identification
CN111680543A (zh) 动作识别方法、装置及电子设备
WO2017219450A1 (zh) 一种信息处理方法、装置及移动终端
CN110765903A (zh) 行人重识别方法、装置及存储介质
CN113191180A (zh) 目标跟踪方法、装置、电子设备及存储介质
KR20200036079A (ko) 적응적 비최대억제 방법을 이용하는 딥러닝기반 영상객체 탐지를 위한 장치 및 방법
JP2019079494A (ja) オブジェクトを描写するための候補である可能性が高い１つまたは複数の候補デジタル画像を突き止める方法
Gupta et al. People detection and counting using YOLOv3 and SSD models
CN111079757A (zh) 服饰属性识别方法、装置及电子设备
JP2015219648A5 (zh)
CN106778675B (zh) 一种视频图像中目标对象的识别方法和装置
CN106557523B (zh) 代表性图像选择方法和设备以及对象图像检索方法和设备

Legal Events

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

Ref document number: 19792513

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019792513

Country of ref document: EP

Effective date: 20200929

NENP Non-entry into the national phase

Ref country code: DE