JP3062383B2 - People recognition method - Google Patents
People recognition methodInfo
- Publication number
- JP3062383B2 JP3062383B2 JP6010056A JP1005694A JP3062383B2 JP 3062383 B2 JP3062383 B2 JP 3062383B2 JP 6010056 A JP6010056 A JP 6010056A JP 1005694 A JP1005694 A JP 1005694A JP 3062383 B2 JP3062383 B2 JP 3062383B2
- Authority
- JP
- Japan
- Prior art keywords
- person
- area
- arm
- intruding
- rectangle
- Prior art date
- Legal status (The legal status 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 status listed.)
- Expired - Lifetime
Links
Landscapes
- Image Processing (AREA)
- Closed-Circuit Television Systems (AREA)
- Burglar Alarm Systems (AREA)
- Image Analysis (AREA)
Description
【０００１】[0001]
【産業上の利用分野】本発明は、例えばテレビジョンカ
メラ（以下、ＴＶカメラという）等による撮像画像から
侵入者を検出する侵入者監視システムにおいて、人物を
認識する人物認識方法に関するものである。BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a person recognition method for recognizing a person in an intruder monitoring system for detecting an intruder from an image picked up by, for example, a television camera (hereinafter referred to as a TV camera).
【０００２】[0002]
【従来の技術】従来、このような分野の技術としては、
例えば次のような文献に記載されるものがあった。 文献；１９９０年電子情報通信学会春季全国大会講演論
文集文冊 ６Ｄ−４３６、佐久間 喜郎、伊東 潔、増
田 功著「微分画像のフレーム間差分による侵入物体検
出」Ｐ．６−４３８ 従来、侵入者監視システムにおける人物認識方法として
は、前記文献に記載されるように、侵入者の有無を確認
するための入力画像と背景画像をＴＶカメラで画像入力
し、前記画像入力と背景画像との差分二値化画像から変
化領域を求め、前記変化領域の特徴量から侵入者の有無
を判断し、侵入者有りの場合には警報を発するようにな
っている。ここで、変化領域の特徴量から侵入者の有無
を判断する方法としては、変化領域の面積と外接四角形
をパラメータとして侵入者の有無を判断している。つま
り、変化領域の面積がある値以上で、かつ外接四角形の
縦横比（外接四角形の縦の長さ／横の長さ）がある範囲
以内であるものを、侵入者として判断するようになって
いた。2. Description of the Related Art Conventionally, techniques in such a field include:
For example, there is one described in the following literature. References: 1990 IEICE Spring National Conference Lecture Paper 6D-436, Yoshiro Sakuma, Kiyoshi Ito, Isao Masuda, "Detection of Intruding Objects by Inter-frame Difference of Differential Images", p. Conventionally, as a person recognizing method in an intruder monitoring system, as described in the above-mentioned document, an input image for confirming the presence or absence of an intruder and a background image are input by a TV camera, and the image input is performed. A change area is obtained from a binary image of the difference between the image and the background image, the presence or absence of an intruder is determined from the feature amount of the change area, and an alarm is issued if an intruder is present. Here, as a method of determining the presence or absence of an intruder from the feature amount of the change area, the presence or absence of an intruder is determined using the area of the change area and a circumscribed rectangle as parameters. That is, an area in which the area of the change area is equal to or larger than a certain value and whose aspect ratio of the circumscribed rectangle (vertical length / horizontal length of the circumscribed rectangle) is within a certain range is determined as an intruder. Was.
【０００３】[0003]
【発明が解決しようとする課題】しかしながら、従来の
人物認識方法では、次のような課題があった。侵入者を
ＴＶカメラで画像入力し、その画像に基づき侵入者の有
無を検出する場合、ＴＶカメラの設置場所から遠くにい
る侵入者は、画像上で変化領域の面積が小さくなる。こ
れに対し、ＴＶカメラの設置場所の近くでは、小さなも
のが動いても変化領域の面積が大きくなる。そのため、
ＴＶカメラの設置場所から遠い位置にいる侵入者を認識
しようとして、変化領域の面積における閾値の設定を低
くすると、ＴＶカメラの近くで人以外の小さな物体が動
くことによって生じる変化領域との区別ができずに、誤
報を発してしまうことがある。また、もう一つのパラメ
ータである外接四角形の縦横比は、人が通常に歩行する
動作を横から捉えた場合、足と腕を閉じたときは開いた
ときに比べ約２〜３倍ほども大きくなる。そのため、縦
横比だけから侵入者を判断する場合も、誤認識が多くな
る。図２は、人物の通常歩行時のパターンであり、図３
は、腕を挙げたときの歩行例を示す図である。図３のよ
うに、侵入人物の腕の開きが著しい場合、外接四角形の
縦横比が大きく変化して人物認識率が低下する。このよ
うに、変化領域の面積と外接四角形の縦横比のみから侵
入者の有無を判断すると、多くの誤報を発するおそれが
あり、これを解決することが困難であった。本発明は、
前記従来技術が持っていた課題として、変化領域の面積
と外接四角形の縦横比から侵入者の有無を判断する方法
によって生じる誤認識の点を解決し、歩行時に発生する
人物特有の形状を利用することによって人物を正確に認
識し、誤報を少なくする侵入者監視システムにおける人
物認識方法を提供するものである。However, the conventional person recognition method has the following problems. When an intruder is input as an image using a TV camera and the presence or absence of the intruder is detected based on the image, the intruder far from the installation location of the TV camera has a smaller change area on the image. On the other hand, near the installation location of the TV camera, the area of the change area increases even if a small object moves. for that reason,
If an attempt is made to recognize an intruder located far from the installation location of the TV camera and the threshold value in the area of the change area is set low, it is difficult to distinguish the change area caused by the movement of a small object other than a person near the TV camera. Failure to do so can cause false alarms. In addition, the aspect ratio of the circumscribed rectangle, which is another parameter, is about 2-3 times larger when a person normally walks from the side when the feet and arms are closed than when they are opened. Become. Therefore, when an intruder is determined only from the aspect ratio, erroneous recognition increases. FIG. 2 shows a pattern of a person during normal walking, and FIG.
FIG. 3 is a diagram showing an example of walking when an arm is raised. As shown in FIG. 3, when the arm of the intruder is significantly open, the aspect ratio of the circumscribed rectangle changes greatly, and the person recognition rate decreases. As described above, judging the presence or absence of an intruder only from the area of the change region and the aspect ratio of the circumscribed rectangle may cause many false reports, and it has been difficult to solve this. The present invention
The problem of the prior art is to solve the problem of misrecognition caused by the method of determining the presence or absence of an intruder from the area of the change area and the aspect ratio of the circumscribed rectangle, and to use a person-specific shape generated when walking. Accordingly, it is an object of the present invention to provide a person recognizing method in an intruder monitoring system for accurately recognizing a person and reducing false reports.
【０００４】[0004]
【課題を解決するための手段】本発明は、前記課題を解
決するために、予め記憶された侵入物体の存在しないと
きの背景画像と逐次入力される入力画像とから侵入物体
の存在によって生じる背景画像に対する該入力画像の変
化領域を抽出する差分処理を施し、その変化領域に基づ
いて侵入物体が人物であることを認識する人物認識方法
において、次のような処理を行うようにしている。即
ち、変化領域の外接四角形を縦に３等分に分割し、侵入
人物の頭領域を含む第１の四角形と胴領域を含む第２の
四角形と足領域を含む第３の四角形を設定する変化領域
分割処理と、その各第１、２及び３の四角形での変化領
域を横軸にそれぞれ投影して得られる投影ヒストグラム
の第１の特徴量と第３の四角形内の足領域の横幅に対す
る外接四角形の横幅の比とに基づき侵入人物の腕の開閉
状態を判定する腕開閉判定処理とを施す。そして、侵入
人物が腕を閉じていると判断した場合は、各第１、２及
び第３の四角形の投影ヒストグラムの第２の特徴量に基
づき、侵入人物を認識する閉腕時人物認識処理を行う。
また、侵入人物が腕を開いていると判定した場合は、各
第１及び第２の四角形の横幅を前記第３の四角形内の足
領域の横幅に限定して抽出したこれら各第１、２及び第
３の四角形の投影ヒストグラムの第３の特徴量に基づ
き、侵入人物を認識する開腕時人物認識処理を行うよう
にしている。第２の発明は、第１の発明における第１の
特徴量を、各第１、２及び第３の四角形での変化領域を
横軸にそれぞれ投影して得られる投影ヒストグラムの標
準偏差としている。SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a background image generated by the presence of an intruding object from a previously stored background image in the absence of an intruding object and an input image sequentially inputted. The following process is performed in a person recognition method for performing a difference process for extracting a change region of the input image from the image and recognizing that the intruding object is a person based on the change region. That is, the circumscribed rectangle of the change area is vertically divided into three equal parts, and the first rectangle including the head area of the invading person, the second rectangle including the torso area, and the third rectangle including the foot area are set. Region dividing processing, and circumscribing the first feature amount of the projection histogram obtained by projecting the first, second and third quadrangular change regions on the horizontal axis and the lateral width of the foot region within the third quadrangle An arm open / close determination process for determining the open / closed state of the intruder's arm based on the ratio of the width of the rectangle is performed. When it is determined that the intruding person has his arms closed, the closed-arm person recognition process for recognizing the intruding person is performed based on the second feature amounts of the first, second, and third rectangular projection histograms. Do.
If it is determined that the intruding person has his arms open, the first and second rectangles are extracted by limiting the width of each of the first and second rectangles to the width of the foot region in the third rectangle. Based on the third feature amount of the third rectangular projection histogram, a person recognition process at the time of arm opening to recognize an intruding person is performed. According to a second aspect of the present invention, the first feature amount in the first aspect is a standard deviation of a projection histogram obtained by projecting the first, second, and third rectangular change areas on the horizontal axis.
【０００５】[0005]
【作用】本発明によれば、以上のように人物認識方法を
構成したので、変化領域分割処理は、侵入物体によって
変動する変化領域に対して、その侵入物体が人物である
場合に人物固有の特徴を認識し易くするために、外接四
角形を３等分する。腕開閉判定処理は、各３個の四角形
での変化領域を横軸にそれぞれ投影して得られる投影ヒ
ストグラムの第１の特徴量と該第３の四角形内の足領域
の横幅に対する前記外接四角形の横幅の比とに基づき侵
入人物の腕の開閉状態を判定する。そして、閉腕時人物
認識処理は、腕開閉判定処理の結果において侵入人物が
腕を閉じていると判断した場合に、各第１、２及び第３
の四角形の投影ヒストグラムの第２の特徴量に基づき、
侵入人物を認識する。また、侵入人物が腕を開いている
と判定した場合に開腕時人物認識処理が、各第１及び第
２の四角形の横幅を前記第３の四角形内の足領域の横幅
に限定して抽出した各第１、２及び第３の四角形の投影
ヒストグラムの第３の特徴量に基づき、侵入人物を認識
する。従って、前記課題を解決できるのである。According to the present invention, since the person recognition method is configured as described above, the changing area division processing is performed on a changing area that fluctuates due to an intruding object when the intruding object is a person. The circumscribed rectangle is divided into three equal parts so that the features can be easily recognized. The arm opening / closing determination processing includes the first feature amount of a projection histogram obtained by projecting each of the three quadrangular change regions on the horizontal axis and the width of the circumscribed quadrangle with respect to the horizontal width of the foot region in the third quadrangle. The open / closed state of the arm of the intruder is determined based on the width ratio. Then, the person recognition process at the time of arm closing is performed in each of the first, second, and third cases when it is determined that the intruding person has closed his arms in the result of the arm opening / closing determination process.
Based on the second feature of the rectangular projection histogram of
Recognize intruders. Further, when it is determined that the intruding person has the arms open, the person recognition process at the time of arm opening extracts the width of each of the first and second rectangles limited to the width of the foot region in the third rectangle. The intruding person is recognized based on the third feature amounts of the first, second, and third rectangular projection histograms. Therefore, the above problem can be solved.
【０００６】[0006]
【実施例】図１は、本発明の実施例を示すもので、例え
ば侵入者監視システムにおける人物認識方法の処理手順
を示す図である。この人物認識方法は、ＴＶカメラ等の
画像入力装置によって背景画像及び入力画像を入力し、
それらを背景画像メモリ１及び入力画像メモリ２に格納
する画像入力処理を行い、該入力した背景画像及び入力
画像に対して差分二値化処理３、変化領域切出し処理
４、変化領域分割処理５、投影ヒストグラム作成処理
６、特徴量抽出処理７を施す。ここで、図１の人物認識
方法は、侵入物体が人物であるらしいことを推定する人
物推定処理８を行う。侵入物体が人物らしい推定された
場合、侵入人物が腕を開いているか閉じているかを判定
する腕開閉判定処理９が実施され、腕を開いていると判
定した場合、特徴量再抽出処理１０と、開腕時人物認識
処理１１とが施され、一方、腕開閉判定処理９の結果で
侵入人物が腕を閉じていると判定した場合、閉腕時人物
認識処理１２が施される。人物推定処理８の推定結果で
侵入物体が人物でないと判明した結果、開腕時人物認識
処理１１の処理結果、または閉腕時人物認識処理１２の
処理結果が、出力処理１３で出力される。以下、これら
各処理を、図を参照しつつ説明する。FIG. 1 shows an embodiment of the present invention, and is a diagram showing a processing procedure of a person recognizing method in an intruder monitoring system, for example. In this person recognition method, a background image and an input image are input by an image input device such as a TV camera,
An image input process for storing them in the background image memory 1 and the input image memory 2 is performed, and the input background image and the input image are subjected to a difference binarization process 3, a changing region extracting process 4, a changing region dividing process 5, A projection histogram creation process 6 and a feature amount extraction process 7 are performed. Here, the person recognition method of FIG. 1 performs a person estimation process 8 for estimating that the intruding object seems to be a person. When the intruding object is estimated to be a person, an arm opening / closing determination process 9 for determining whether the intruding person has an open or closed arm is performed, and when it is determined that the arm is open, a feature re-extraction process 10 is performed. When the arm opening / closing determination processing 9 determines that the intruding person has closed his arms, the arm recognition processing 12 is performed. As a result of the estimation result of the person estimation processing 8, it is determined that the intruding object is not a person, and the processing result of the open-arm person recognition processing 11 or the processing result of the closed-arm person recognition processing 12 is output in the output processing 13. Hereinafter, each of these processes will be described with reference to the drawings.
【０００７】メモリ１，２への画像入力処理 図４は背景画像の例、図５は入力画像の例を示す図であ
る。まず、ＴＶカメラ等の画像入力装置を用い、図４に
示すように、侵入物体が存在しない背景だけを撮影し、
その背景画像を予め背景画像メモリ１の中に格納してお
く。そして、背景画像を撮影したのと同じ位置で撮影
し、その撮影した図５に示すような入力画像を入力画像
メモリ２の中に格納する。差分二値化処理３ 画像入力処理後、差分二値化処理３において、入力画像
中における背景画像からの変化を検出するために、背景
画像メモリ１と入力画像メモリ２からそれぞれ入力され
る背景画像と入力画像とで、その差の絶対値を計算して
ある適当な閾値を用いて二値化し、差分二値化画像を求
める。この差分二値化画像では、入力画像中で背景画像
から変化した領域の画素の値が１、変化していない領域
では画素の値が０に設定され、変化領域として抽出され
ることになる。しかし、侵入物体によっては背景画像と
局所的に同じ濃度を持っていることがある。その場合の
差分二値化画像は、変化領域が切れ切れになる。そのた
め、それぞれの変化領域は別個に認識され、本来の一つ
の侵入物体しては認識されなくなる。そこで、次の変化
領域切出し処理４を行う。[0007] The image input processing Figure 4 to the memory 1 and 2 example of the background image, FIG. 5 is a diagram showing an example of an input image. First, using an image input device such as a TV camera, as shown in FIG. 4, only the background where no intruding object exists is photographed.
The background image is stored in the background image memory 1 in advance. Then, the background image is photographed at the same position where the background image was photographed, and the photographed input image as shown in FIG. 5 is stored in the input image memory 2. Difference binarization processing 3 After the image input processing, in the difference binarization processing 3, the background images input from the background image memory 1 and the input image memory 2, respectively, in order to detect a change from the background image in the input image. And the input image, the absolute value of the difference is calculated and binarized using an appropriate threshold to obtain a differential binarized image. In this difference binarized image, the value of the pixel in the area that has changed from the background image in the input image is set to 1, and the value of the pixel is set to 0 in the area that has not changed, and is extracted as a changed area. However, some intruding objects may have the same local density as the background image. In this case, in the difference binary image, the change area is cut off. Therefore, each change area is recognized separately, and is not recognized as an original one intruding object. Therefore, the following change area extraction processing 4 is performed.
【０００８】変化領域切出し処理４ 図６は、差分二値化画像の例を示す図である。図６のよ
うな差分二値画像に対し、変化領域切出し処理４では、
一つの侵入物体を表す変化領域を一つにまとめるため
に、例えば差分二値化画像を数回膨張した後、同じ回数
だけ収縮する。膨張とは、値０の画素の回りに値１の画
素がある場合、その値０の画素を１に変換するものであ
る。つまり、変化領域の回りを一画素分膨らますのであ
る。収縮はその逆で、変化領域の回りを一画素分小さく
する。この作業によって、一つの侵入物体を表すと思わ
れる変化領域は、一つの領域として抽出される。そし
て、このようにして得られた変化領域を外接四角形で囲
み、この外接四角形に囲まれた領域を差分二値化画像か
ら切出す。この切出された領域から、侵入物体が人物か
否かを判断する場合、外接四角形の縦横比だけから判断
しようとすると、腕の開閉状態により外接四角形の縦横
比は大きく変化し、正確に人を認識できない。しかし、
この外接四角形の縦横比の変化はある規制性がある。そ
れは、外接四角形の縦横比が小さくなる（すなわち、外
接四角形が横に広がる）のは、人が腕と足とを開いた場
合であり、縦横比が大きくなるのは腕と足を閉じた場合
である。ただし、歩行時における腕の開き方は人によっ
て個人差があるので、腕の開き具合と縦横比の変化とが
常に関係しているとはいえない。一方、足の開閉と縦横
比の変化は常に上で述べたような関係がある。そこで、
外接四角形の縦横比と変化領域中で足に相当する部分の
形状とのこのような規則性を捉えることができれば、侵
入物体が人であるか否かの認識をし易くなる。そのた
め、次の変化領域分割処理５を行う。[0008] change region extraction processing 4 FIG. 6 is a diagram showing an example of a differential binary image. For the difference binary image as shown in FIG.
In order to combine the change areas representing one intruding object into one, for example, the differential binarized image is expanded several times and then contracted the same number of times. The dilation means that, when there is a pixel having a value of 1 around a pixel having a value of 0, the pixel having a value of 0 is converted to 1. In other words, it expands by one pixel around the change area. Shrinkage is the opposite, making the area around the change area smaller by one pixel. By this operation, a change area that is considered to represent one intruding object is extracted as one area. Then, the thus obtained change region is surrounded by a circumscribed rectangle, and the region surrounded by the circumscribed rectangle is cut out from the differential binarized image. When judging whether or not the intruding object is a person from the cut-out area, if it is attempted to judge only from the aspect ratio of the circumscribed rectangle, the aspect ratio of the circumscribed rectangle greatly changes depending on the open / closed state of the arm, and the person is accurately determined Can not recognize. But,
The change in the aspect ratio of the circumscribed rectangle has a certain regulating property. That is, the aspect ratio of the circumscribed rectangle decreases (that is, the circumscribed rectangle expands horizontally) when a person opens his / her arms and legs, and the aspect ratio increases when the arms and legs are closed. It is. However, since the manner of opening the arms during walking varies from person to person, the degree of opening of the arms and the change in the aspect ratio are not always related. On the other hand, the opening / closing of the foot and the change in the aspect ratio always have the relationship described above. Therefore,
If such regularity of the aspect ratio of the circumscribed rectangle and the shape of the portion corresponding to the foot in the change area can be grasped, it becomes easy to recognize whether or not the intruding object is a person. Therefore, the following change area division processing 5 is performed.
【０００９】変化領域分割処理５ 図７は、外接四角形の分割と足領域の横幅を示す図であ
り、変化領域の外接四角形と侵入人物の頭領域を含む第
１の四角形と胴領域を含む第２の四角形と足領域を含む
第３の四角形が示されている。変化領域分割処理５で
は、図７に示すように、変化領域の外接四角形を縦に３
等分する。これは、人を表す変化領域において、縦横比
変化の原因となる人の腕と足の部分を別々の領域に分割
し、変化領域の形状を捉え易くするためのものである。
そして、次の投影ヒストグラム作成処理６へ進む。投影ヒストグラム作成処理 ６ 図８は、投影ヒストグラムの例を示し、図７の各第１〜
３の四角形の投影ヒストグラムが示されている。投影ヒ
ストグラム作成処理６は、各第１〜３の四角形での変化
領域の度数を横軸に投影した投影ヒストグラムをそれぞ
れ作成する。特徴量抽出処理７ 特徴量抽出処理７では、外接四角形の縦横比、各投影ヒ
ストグラムの平均値及び投影ヒストグラムの標準偏差等
を用いて、侵入物体の特徴量を計算する。[0009] change area dividing process 5 FIG. 7 is a diagram showing the width of the divided and the foot area of the circumscribed rectangle, first comprises a first square and a torso region including the head area of the circumscribed quadrangle and intrusion person change region Two rectangles and a third rectangle including a foot region are shown. In the change area dividing process 5, as shown in FIG.
Divide equally. This is for dividing a human arm and a foot portion which causes a change in the aspect ratio in a change region representing a person into separate regions so that the shape of the change region can be easily grasped.
Then, the process proceeds to the next projection histogram creation processing 6. Projection Histogram Creation Process 6 FIG. 8 shows an example of the projection histogram,
A three square projection histogram is shown. The projection histogram creation processing 6 creates projection histograms in which the frequencies of the change areas in the first to third rectangles are projected on the horizontal axis. Feature amount extraction process 7 In feature amount extraction process 7, the feature amount of the intruding object is calculated using the aspect ratio of the circumscribed rectangle, the average value of each projection histogram, the standard deviation of the projection histogram, and the like.
【００１０】人物推定処理８ 人物推定処理８では、外接四角形の縦横比等の特徴量が
予め設定された範囲に在れば、侵入物体は、「人らし
い」と推定しする。また、特徴量が範囲を外れている場
合には、「人ではない」と推定する。侵入物体が「人ら
しい」と推定された場合、次の腕開閉判定処理９が行わ
れる。腕開閉判定処理９ 腕開閉判定処理９は、各第１〜３の四角形における投影
ヒストグラムの第１の特徴量である標準偏差と、図７中
に示される第３の四角形内の足領域の横幅（足領域にお
ける画素の値が１の領域の最大幅）に対する外接四角形
の横幅の比とから、侵入人物が腕を開いているか否かを
判定する。本実施例では、腕の開きの大きさを手の先の
位置によって場合分けし、それぞれの場合に閾値を設け
ている。その場合分けには各第１の四角形と第２の四角
形の投影ヒストグラムの標準偏差ＳＤ１，ＳＤ２を用い
ている。標準偏差ＳＤ１＞ＳＤ２のときは、手の先が第
１の四角形内にある（腕が頭領域の高さまで広く開かれ
ている）と判定し、標準偏差がＳＤ１≦ＳＤ２のとき、
手の先が第２の四角形にある（腕が胴領域の高さ程度で
開かれている）と判定する。侵入人物の腕が開かれてい
ると判定した場合と、腕が閉じていると判定した場合と
で以降の処理が別れる。[0010] In the figures estimating process 8 person estimation process 8, if the range of the feature amount, such as the aspect ratio of the circumscribed rectangle is set in advance, the intruding object is estimated as "likely human". When the feature amount is out of the range, it is estimated that “the person is not a person”. When the intruding object is estimated to be “human”, the next arm opening / closing determination processing 9 is performed. Arm opening / closing determination processing 9 Arm opening / closing determination processing 9 includes a standard deviation, which is a first feature amount of the projection histogram in each of the first to third rectangles, and a width of the foot region in the third rectangle shown in FIG. It is determined from the ratio of the width of the circumscribed rectangle to (the maximum width of the region where the value of the pixel in the foot region is 1) whether or not the intruding person has his arms open. In this embodiment, the size of the opening of the arm is classified according to the position of the tip of the hand, and a threshold is provided for each case. In this case, the standard deviations SD1 and SD2 of the projection histograms of the first rectangle and the second rectangle are used. When the standard deviation SD1> SD2, it is determined that the tip of the hand is within the first rectangle (the arm is wide open to the height of the head region), and when the standard deviation is SD1 ≦ SD2,
It is determined that the tip of the hand is in the second rectangle (the arm is open at about the height of the torso region). Subsequent processing is divided into a case where it is determined that the arm of the intruder is open and a case where it is determined that the arm is closed.
【００１１】特徴量再抽出処理１０ 腕開閉判定処理９で侵入人物が腕を開いていると判定し
た場合、特徴量再抽出処理１０が行われる。特徴量再抽
出処理１０では、すべての変化領域（頭、胴、足領域）
を足領域の横幅の範囲内に限定して第３の特徴量を計算
する。この第３の特徴量は、例えば、その範囲限定され
た場合の外接四角形の縦横比と各第１〜３の四角形での
投影ヒストグラムの平均値と標準偏差等である。図９
は、足領域の横幅に限定した変化領域を示している。即
ち、特徴量再抽出処理は、図９のように、最も動きの大
きい腕の部分を除くことで、人の特徴量の評価を小さい
範囲に限定する。これにより、侵入物体を人物として認
識するための特徴量を厳密に設定することが可能とな
る。開腕時人物認識処理１１ 開腕時人物認識処理１１は、特徴量再抽出処理１０の計
算結果の特徴量にもとづいて侵入物体が人物であるか否
かが求められる。[0011] If the entering person by the feature amount extracted again for 10 arms closing determination process 9 is determined to be open arms, the feature quantity extracted again for 10 performed. In the feature amount re-extraction process 10, all the change areas (head, torso, and foot areas)
Is limited to the range of the width of the foot region, and the third feature value is calculated. The third feature amount is, for example, the aspect ratio of the circumscribed rectangle and the average value and standard deviation of the projection histograms of the first to third rectangles when the range is limited. FIG.
Indicates a change region limited to the width of the foot region. That is, in the feature re-extraction process, as shown in FIG. 9, the evaluation of the feature of a person is limited to a small range by excluding the arm portion where the movement is the largest. Thus, it is possible to strictly set a feature amount for recognizing an intruding object as a person. Arm-opening person recognition process 11 In arm-opening person recognition process 11, it is determined whether or not the intruding object is a person based on the feature amount calculated by the feature amount re-extraction process 10.
【００１２】閉腕時人物認識処理１２ 腕開閉判定処理９で侵入人物が腕を閉じていると判定さ
れた場合、閉腕時人物認識処理１２が行われ、侵入物体
が人物であるか否かが求められる。この場合は、侵入人
物が歩行状態と考えられるので、閉腕時人物認識処理１
２では、まず、第１〜３の四角形から、例えば次の
（１）〜（６）のような第２の特徴量を求める。 （１）；各第１〜３の四角形中のヒストグラムのブロッ
ク数 （２）；各ブロックの平均値 （３）；各ブロックの平均値の第１〜３の四角形間の差 （４）；各ブロックにおける極大値数 （５）；各ブロックにおける面積の割合 （６）；第３の四角形における極大値の位置 ここで、ブロックとは各投影ヒストグラム中で度数がゼ
ロの所で区切られる度数が１以上の一塊の部分を指して
いる。これらの第２の特徴量と外接四角形の縦横比との
関係から、変化領域の判定を行い、侵入物体が人物であ
るのか否かが判定される。Arm Recognition Process 12 When it is determined in the arm opening / closing determination process 9 that the intruding person has his arms closed, the arm closing person recognition process 12 is performed to determine whether the intruding object is a person. Is required. In this case, since the intruding person is considered to be in the walking state, the person recognition process 1 when the arm is closed
In step 2, first, for example, second feature values such as the following (1) to (6) are obtained from the first to third rectangles. (1); the number of blocks of the histogram in each of the first to third rectangles (2); the average value of each block (3); the difference between the first to third rectangles of the average value of each block (4); Number of local maxima in block (5); Ratio of area in each block (6); Location of local maxima in third rectangle Here, a block has a frequency of 1 at a frequency of zero in each projection histogram. It points to the above lump. A change area is determined from the relationship between the second feature amount and the aspect ratio of the circumscribed rectangle, and it is determined whether the intruding object is a person.
【００１３】出力処理１３ 出力処理１３では、人物推定処理８で侵入物体が人物で
ないと判明した結果、開腕時人物認識処理１１の処理結
果、または閉腕時人物認識処理１２の処理結果が、出力
処理１３で出力される。以上のように、本実施例では、
侵入人物の腕の開閉状態を第１の四角形及び第２の四角
形における投影ヒストグラムと足領域の横幅に対する外
接四角形の横幅の比とを用いて腕開閉判定処理９で判定
し、それぞれの場合に対応して人物認識を行っている。
即ち、開腕時人物認識処理１１で、腕の開きが著しく大
きい場合に、腕の部分を人物認識の対象となる変化領域
から除いている。これにより、人物の特徴量の評価の範
囲が限定され、人物の認識率が向上する。また、腕の開
きが小さい場合、の閉腕時人物認識処理１２で、外接四
角形の縦横比と第１〜３の四角形の各変化領域を横軸上
へ投影して得られる投影ヒストグラムの特徴量である特
徴（１）〜（６）とから、人物を認識するようにしたの
で、正確に人物を認識でき、侵入者監視システムにおい
て誤報を少なくできる。なお、本発明は上記実施例に限
定されず、例えば、各特徴量として用いられた投影ヒス
トグラムの標準偏差は、分散を用いてもよい、また、閉
腕時人物認識処理１２における人物認識は、他の方法を
用いたり特徴（１）〜（６）を変更することも可能であ
る。 Output Processing 13 In the output processing 13, as a result of the person estimation processing 8 determining that the intruding object is not a person, the processing result of the open-arm human recognition processing 11 or the processing result of the closed-arm human recognition processing 12 is It is output in output processing 13. As described above, in this embodiment,
The open / closed state of the arm of the intruding person is determined by the arm open / close determination processing 9 using the projection histograms of the first and second rectangles and the ratio of the width of the circumscribed rectangle to the width of the foot region, and corresponds to each case. And perform person recognition.
That is, in the person recognition process 11 when the arm is opened, when the opening of the arm is extremely large, the arm portion is excluded from the change area to be subjected to the person recognition. Thereby, the range of evaluation of the feature amount of the person is limited, and the recognition rate of the person is improved. In the case where the arm is open slightly, the characteristic amount of the projection histogram obtained by projecting the aspect ratio of the circumscribed rectangle and the respective changed areas of the first to third rectangles on the horizontal axis in the closed-arm person recognition process 12 of FIG. Since the person is recognized from the following features (1) to (6), the person can be recognized accurately, and false reports can be reduced in the intruder monitoring system. Note that the present invention is not limited to the above-described embodiment. For example, the standard deviation of the projection histogram used as each feature amount may use a variance. It is also possible to use another method or change the features (1) to (6).
【００１４】[0014]
【発明の効果】以上詳細に説明したように、本発明によ
れば、侵入人物の腕の開閉状態を各投影ヒストグラムと
足領域の横幅に対する外接四角形の横幅の比とを用いて
腕開閉判定処理で判定し、判定結果に基づき人物認識を
行っている。即ち、開腕時人物認識処理で、腕の開きが
著しく大きい場合に、腕の部分を人物認識の対象となる
変化領域から除いている。これにより、人物の特徴量の
評価の範囲が限定され、人物の認識率が向上する。ま
た、腕の開きが小さい場合、閉腕時人物認識処理で、外
接四角形の縦横比と第１〜３の四角形の各変化領域を横
軸上へ投影して得られる投影ヒストグラムの特徴量から
人物認識するようにしたので、正確に人物を認識でき、
例えば、侵入者監視システムにおいて誤報を少なくでき
る。As described above in detail, according to the present invention, the open / closed state of the arm of the invading person is determined by using each projection histogram and the ratio of the width of the circumscribed rectangle to the width of the foot area. And the person is recognized based on the determination result. That is, in the person recognition process when the arm is opened, if the opening of the arm is extremely large, the arm portion is excluded from the change area to be subjected to the person recognition. Thereby, the range of evaluation of the feature amount of the person is limited, and the recognition rate of the person is improved. Further, when the opening of the arm is small, the person recognition processing is performed based on the feature amount of the projection histogram obtained by projecting the aspect ratio of the circumscribed rectangle and each of the first to third rectangles on the horizontal axis in the human recognition process when the arm is closed. Because it is made to recognize, the person can be accurately recognized,
For example, false alarms can be reduced in an intruder monitoring system.
【図１】本発明の実施例の人物認識方法の処理手順を示
す図である。FIG. 1 is a diagram showing a processing procedure of a person recognition method according to an embodiment of the present invention.
【図２】人物の通常歩行時のパターンを示す図である。FIG. 2 is a diagram showing a pattern when a person walks normally.
【図３】腕を挙げた時の歩行例を示す図である。FIG. 3 is a diagram illustrating an example of walking when an arm is raised.
【図４】背景画像の例を示す図である。FIG. 4 is a diagram illustrating an example of a background image.
【図５】入力画像の例を示す図である。FIG. 5 is a diagram illustrating an example of an input image.
【図６】差分二値画像の例を示す図である。FIG. 6 is a diagram illustrating an example of a differential binary image.
【図７】外接四角形の分割と足領域の横幅を示す図であ
る。FIG. 7 is a diagram illustrating division of a circumscribed rectangle and width of a foot region.
【図８】投影ヒストグラムを示す図である。FIG. 8 is a diagram showing a projection histogram.
【図９】足領域の横幅に限定した変化領域を示す図であ
る。FIG. 9 is a diagram showing a change region limited to the width of the foot region.
１ 背景画像メモリ ２ 入力画像メモリ ３ 差分二値化処理 ４ 変化領域切出し処理 ５ 変化領域分割処理 ６ 投影ヒストグラム作成処理 ７ 特徴量抽出処理 ８ 人物推定処理 ９ 腕開閉判定処理 １０ 特徴量再抽出処理 １１ 開腕時人物認識処理 １２ 閉腕時人物認識処理 １３ 出力処理 DESCRIPTION OF SYMBOLS 1 Background image memory 2 Input image memory 3 Difference binarization processing 4 Change area extraction processing 5 Change area division processing 6 Projection histogram creation processing 7 Feature amount extraction processing 8 Person estimation processing 9 Arm open / close determination processing 10 Feature amount reextraction processing 11 Arm Recognition Processing with Arm Open 12 Person Recognition Processing with Arm Closed 13 Output Processing
───────────────────────────────────────────────────── フロントページの続き (72)発明者 中川 聰 東京都港区虎ノ門１丁目７番12号 沖電 気工業株式会社内 (56)参考文献 特開 平６−20049（ＪＰ，Ａ) 特開 平５−334572（ＪＰ，Ａ) (58)調査した分野(Int.Cl.7，ＤＢ名) G06T 1/00 G06T 7/00 - 7/60 H04N 7/18 G08B 13/194 - 13/196 ──────────────────────────────────────────────────続 き Continuation of front page (72) Inventor Satoshi Nakagawa 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) References JP-A-6-20049 (JP, A) JP-A Hei 5-334572 (JP, A) (58) Fields investigated (Int. Cl. 7 , DB name) G06T 1/00 G06T 7 /00-7/60 H04N 7/18 G08B 13/194-13/196
Claims (2)
きの背景画像と逐次入力される入力画像とから侵入物体
の存在によって生じる該背景画像に対する該入力画像の
変化領域を抽出する差分処理を施し、該変化領域に基づ
いて該侵入物体が人物であることを認識する人物認識方
法において、 前記変化領域の外接四角形を縦に３等分に分割し、前記
侵入人物の頭領域を含む第１の四角形と胴領域を含む第
２の四角形と足領域を含む第３の四角形を設定する変化
領域分割処理と、 前記各第１、２及び３の四角形での変化領域を横軸にそ
れぞれ投影して得られる投影ヒストグラムの第１の特徴
量と該第３の四角形内の足領域の横幅に対する前記外接
四角形の横幅の比とに基づき前記侵入人物の腕の開閉状
態を判定する腕開閉判定処理とを施し、 前記侵入人物が腕を閉じていると判定した場合は、前記
各第１、２及び第３の四角形の投影ヒストグラムの第２
の特徴量に基づき、前記侵入人物と認識する閉腕時人物
認識処理を行い、 前記侵入人物が腕を開いていると判定した場合は、前記
第２の特徴量と異なって前記各第１及び第２の四角形の
横幅を前記第３の四角形内の足領域の横幅に限定して抽
出した該各第１、２及び第３の四角形の投影ヒストグラ
ムの第３の特徴量に基づき、該侵入人物を認識する開腕
時人物認識処理を行う、 ことを特徴する人物認識方法。1. A difference processing for extracting a change area of an input image with respect to a background image caused by the presence of an intruding object from a previously stored background image when no intruding object is present and an input image sequentially input. A person recognition method for recognizing that the intruding object is a person based on the changing area, wherein a circumscribed rectangle of the changing area is vertically divided into three equal parts, and a first area including a head area of the intruding person is included. A changing area dividing process for setting a second quadrangle including a quadrangle and a torso area and a third quadrangle including a foot area; and projecting the first, second and third quadrangular changing areas on the horizontal axis, respectively. Arm opening / closing determination processing for determining the open / closed state of the intruding person's arm based on the first feature amount of the obtained projection histogram and the ratio of the width of the circumscribed rectangle to the width of the foot region in the third rectangle. Alms, before If it is determined that the invading person has closed his arms, the second, third, and third rectangular projection histograms of the first, second, and third squares are used.
Based on the characteristic amount of the invading person, perform a person-recognition process at the time of arm closing to recognize the intruding person. When it is determined that the intruding person has his arms open, the first and second characteristic amounts differ from the second characteristic amount. The intruding person is determined based on the third feature amounts of the first, second, and third rectangle projection histograms extracted by limiting the width of the second rectangle to the width of the foot region in the third rectangle. Performing a person recognition process at the time of arm opening for recognizing a person.
び第３の四角形での変化領域を横軸にそれぞれ投影して
得られる投影ヒストグラムの標準偏差としたことを特徴
とする請求項１記載の人物認識方法。2. The method according to claim 1, wherein the first feature amount is a standard deviation of a projection histogram obtained by projecting the first, second, and third quadrangular change areas on a horizontal axis. The person recognition method according to claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP6010056A JP3062383B2 (en) | 1994-01-31 | 1994-01-31 | People recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP6010056A JP3062383B2 (en) | 1994-01-31 | 1994-01-31 | People recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JPH07220047A JPH07220047A (en) | 1995-08-18 |
| JP3062383B2 true JP3062383B2 (en) | 2000-07-10 |
Family
ID=11739739
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP6010056A Expired - Lifetime JP3062383B2 (en) | 1994-01-31 | 1994-01-31 | People recognition method |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP3062383B2 (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE69921237T2 (en) * | 1998-04-30 | 2006-02-02 | Texas Instruments Inc., Dallas | Automatic video surveillance system |
| US6404900B1 (en) * | 1998-06-22 | 2002-06-11 | Sharp Laboratories Of America, Inc. | Method for robust human face tracking in presence of multiple persons |
| KR100790860B1 (en) * | 2004-12-14 | 2008-01-03 | 삼성전자주식회사 | Human tracking apparatus and method thereof, and recording media and removing electronic system therefor |
| JP4710437B2 (en) * | 2005-07-01 | 2011-06-29 | トヨタ自動車株式会社 | Object search device |
-
1994
- 1994-01-31 JP JP6010056A patent/JP3062383B2/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| JPH07220047A (en) | 1995-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7215798B2 (en) | Method for forgery recognition in fingerprint recognition by using a texture classification of gray scale differential images | |
| JP3350296B2 (en) | Face image processing device | |
| US5715325A (en) | Apparatus and method for detecting a face in a video image | |
| KR100455294B1 (en) | Method for detecting user and detecting motion, and apparatus for detecting user within security system | |
| US6661907B2 (en) | Face detection in digital images | |
| KR101724658B1 (en) | Human detecting apparatus and method | |
| JP3636927B2 (en) | Face image processing device | |
| JP3695990B2 (en) | Face image processing device | |
| KR100624481B1 (en) | Method for tracking face based on template | |
| JP2975629B2 (en) | Image recognition device | |
| JPH07192112A (en) | Intruding object recognizing method | |
| CN112733814B (en) | Deep learning-based pedestrian loitering retention detection method, system and medium | |
| JP3062383B2 (en) | People recognition method | |
| KR20200060868A (en) | multi-view monitoring system using object-oriented auto-tracking function | |
| JP3088880B2 (en) | Person recognition device | |
| KR101985869B1 (en) | A livestock theft surveillance apparatus using morphological feature-based model and method thereof | |
| JP3417635B2 (en) | Intruding object recognition method | |
| JPH07129778A (en) | Image monitoring device | |
| JP3065838B2 (en) | People recognition method | |
| CN111462169B (en) | Mouse trajectory tracking method based on background modeling | |
| JP3065854B2 (en) | People recognition method | |
| JP3078203B2 (en) | Intruding object recognition device | |
| Alexandropoulos et al. | Real-time change detection for surveillance in public transportation | |
| JP6866182B2 (en) | Bed positioning device | |
| JP6847708B2 (en) | Bed positioning device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 Effective date: 20000411 |