JP3105336B2 - Moving object counting method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
The present invention relates to a moving object counting method for counting moving objects such as people, vehicles, and objects moving on a conveyor belt.

[0002]

2. Description of the Related Art Japanese Patent Application No. 2-2400000 discloses a method of counting moving objects in a moving image by image processing.
A moving object is processed from a two-dimensional image obtained by stacking one-dimensional slit image signals of a certain length obtained from one scanning line in the time direction without processing the moving image in frame units, based on a background image in which no object is previously captured. Is extracted, and the number of moving objects is counted by outputting the number of regions in the image.

[0003]

However, in this method, since the area of the extraction area is divided by a fixed fixed reference value and converted to an integer, the moving direction, the speed, the deformation, and the overlapping of a plurality of objects are determined. When the area changes, the counting error for one area increases, and the counting accuracy over a long period of time becomes unstable. Also, since the object area is extracted based on the difference from the once stored background image, the background image that should be regarded as the background is stored due to environmental changes, slight movement of the imaging system, and long-term suspension of moving objects. It will remain different, and subsequent accurate extraction will not be possible.

Further, in the extraction process, if the moving object and the background have similar image information, it may be difficult to separate the moving object and the background. It can be expected to improve the accuracy of.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned conventional problems and to provide a method for accurately counting a moving object group.

[0006]

FIG. 1 shows a principle diagram of the present invention. The basic operation principle of the present invention is described in Japanese Patent Application No. 2-2400.
It is based on the principle of counting moving objects in No. 00. In the case of the present invention, in order to absorb the counting error for each area, the number estimation is performed by the non-integer processing of the count output or the multi-step threshold processing, and the parameter serving as the reference of the count is reflected by reflecting the situation of the target moving image. The background image is determined semi-automatically, and the background image is automatically reconstructed in order to perform stable extraction processing even if the slit changes suddenly. Is arranged.

An image input unit 1-1 captures a moving object 1-9 moving on a background 1-8 at a fixed angle and a one-dimensional signal (slit) of digital image data corresponding to a measurement line 1-10. Image).

Reference numeral 1-2 denotes an extraction processing unit, which is disclosed in Japanese Patent Application No. Hei 2-240.
Based on the principle of moving object extraction in No. 000, a two-dimensional spatio-temporal image in which images passing through a slit of an object moving in a scene are stacked is extracted by difference processing, and the extraction area is shaped to obtain the area of the moving object image .

Reference numeral 1-3 denotes a background image correction unit which obtains an initial background image and updates the background image based on the background setting principle in Japanese Patent Application No. 2-2400000, and is locally continuous due to environmental fluctuations and the like. If it is determined that the state is incorrectly extracted, the background image is locally updated.

Reference numeral 1-4 denotes a parameter setting unit which estimates a reference value for counting from the characteristics of the frequency distribution of the characteristic amounts obtained by the counting process within the designated time. Reference numeral 1-5 denotes a labeling processing unit, which performs labeling of the shaped area based on the principle of area labeling in Japanese Patent Application No. 2-2400000, and calculates a feature amount for each label area.

Reference numeral 1-6 denotes a count processing unit which performs a division process with the count reference value set by the parameter setting unit 1-4 by using the obtained characteristic amount for each moving object region, and a multi-stage threshold process. Estimate the equivalent number of people. Further, the estimation of the moving direction is performed based on the counting principle in Japanese Patent Application No. 2-2400000.

Reference numeral 1-7 denotes an output section for displaying the counting result, which outputs the counting result obtained by the counting section 1-6.

[0013]

According to the present invention, the counting error of each area is absorbed by the non-integral processing of the count output and the determination of the reference value and the threshold value reflecting the moving image, and the sudden reconstruction of the background image by local reconstruction of the background image. The most important feature is that stable extraction processing is performed even for continuous continual changes, and the placement of the background with a clear difference from the moving object enables accurate counting of moving objects. This is a difference from the prior art.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS As one embodiment of the present invention, a method for counting the number of persons walking on a sidewalk will be described with reference to the drawings. It is easy to apply the object to another moving object instead of a person.

FIG. 2 shows an embodiment of the counting process of the walking person. First, in the image input unit 2-1, image information corresponding to the measurement line is sampled by the slit image forming unit 2-2 from an image obtained by photographing a person passing through the measurement line on the sidewalk as a background at a fixed angle. A one-dimensional slit image is obtained and stored in the slit image memory 2-3 together with the extraction time. However, the acquisition of the one-dimensional slit image is not limited to the TV camera, but may be a one-dimensional sensor. The sample signal may be a signal of distance data such as an ultrasonic wave or an infrared ray, and the brightness may be replaced with a distance value. Hereinafter, a description will be given as an example of TV camera input under visible light.

Here, the background of claim 6 is that, as shown in the example of FIG. 5, a sheet of a material having a uniform and uniform color, having a uniform brightness and being almost non-reflective, or a similar road surface is used. This is achieved by painting. In addition, as shown in the example of FIG. 6, the background of claim 7 is uniform and uniform in color along the measurement line from the side opposite to the imaging direction of the TV camera so that a silhouette image of a passing person can be obtained. This is achieved by performing illumination that emits bright light.

Initial background image configuration In FIG. 2-4, the user issues a command in anticipation of the initial frame or when there is no pedestrian, takes out the image, and stores it in the background image memory 2-6. Alternatively, a fixed background image is prepared in advance.

Next, in the moving object region extraction unit 2-5,
Input image f read from slit image memory 2-3
The absolute value of the difference between (x, t) and the background image b (x) stored in the background image memory 2-6 is calculated, and the changed part is binarized and extracted. An image G (x, t) is created and stored in the binarized image memory 2-7 together with the binarized image at the immediately preceding time. That is,

[0019]

(Equation 1)

Is calculated. Here, Th (x) is an extraction threshold, and x is the width of a slit image taking x = 0, 1,..., Xsize. The area shaping unit 2-8 converts the binarized image G (x, t) for one scanning line and the binarized image G (x, t-1) one time before read out from the binarized image memory. Use,
Enlargement, often used as a binary image technique for image processing
The noise component is removed by performing a reduction process, and the image data is returned to the binarized image memory 2-7.

In the area labeling section 2-9, the shaped binary image is labeled based on the labeling method of the area in Japanese Patent Application No. 2-22400000. , 1st and 2nd moments, and feature quantities such as the minimum and maximum values of the area coordinates are calculated and stored in the label memory 2-10 together with the label numbers. At this time, the feature quantity is integrated with the label when it is determined that the two areas are connected.

The count processing unit 2-11 analyzes the characteristic amount obtained by the area labeling unit 2-9, calculates the total number of pedestrians and, if necessary, the speed of the pedestrian, and stores the count values in the count value buffer 2 Send to -12.

The calculation of the total number of pedestrians is described in Japanese Patent Application No. Hei 2-240.
Based on the counting method in No. 000, the area value equal to or larger than the counting threshold is divided by the counting reference value while discriminating between the noise area and the area corresponding to single passage using the area of the extracted feature quantity. The total number is obtained by calculating the number of persons and sequentially accumulating the number of persons obtained for each area.

The counting method according to the third aspect is realized by changing the reference value used for conversion into integers according to the estimated number of passers as in the following equation. For example, if the area of the extracted feature amount is M ₀ , the noise region and one, two,.
The boundary values for discriminating the passage of persons and K persons are _Tn and T, respectively.
_1, T ₂ ..., as T _K-1, T _K, as follows, except noise region outputs the number of people area M ₀ corresponds if in the range of K of persons, K people or more counts Calculates the number of persons by converting the quotient of the division by the count reference value S into an integer.

[0025]

(Equation 2)

In the counting method according to the fourth aspect, the division result is used without using an integer conversion function as in the following equation.

[0027]

(Equation 3)

[0028] However, T _'s is a threshold of noise area. The calculation of walking speed and direction is described in Japanese Patent Application No. 2-2400.
This is performed based on the operation of calculating the speed of the moving object and determining the moving direction of the moving object in No. 00.

Here, the counting reference used in the counting section
Value SS and extraction threshold T_n, T _{0,1,2, ...}Is para
In the meter setting section 2-13, the area within the specified time
The frequency distribution of the feature values obtained in the labeling part is obtained, and the
Estimate by finding extreme values.

For example, when the histogram 3-1 of the area value of the extraction area within a certain time is obtained, as shown in FIG. 3, the histogram becomes multimodal in accordance with the number of persons having the average walking speed. An extreme value p ₁ (3-2) corresponding to the single passage is estimated, and the value at that time is set as a count reference value SS. The counting threshold value may be obtained by performing extreme value estimation on the boundary q ₀ (3-3) between the noise region and the boundary q ₁ (3-4) between the passing of two persons and the single passage. Alternatively, it may be determined by a relational expression considering overlap of a plurality of persons based on a threshold value of single passage. This process is
Each value is determined by eye measurement from the histogram, or automatically determined by a program for estimating an extreme value, and used for the counting process. Further, the counting is performed not only at the start of the counting operation but also as necessary, and the obtained parameters are stored in the counting parameter memory 2-14.

In the background updating section 2-15, when a pedestrian is not present, an image is taken out at a timing set by the user or set by the user, and the background image is stored in the background image memory 2-15.
Accumulate in 6. Alternatively, the current time t
If the sum of the differences between the image and the immediately preceding time t-1 is small, the background image is also updated if only the background image exists or if the labeling processing unit determines that there is no area being tracked. Further, the background image may be adaptively updated for each pixel by using a background updating method based on the construction of a background model in Japanese Patent Application No. Hei 3-187981.

[0032]

(Equation 4)

However, when the background suddenly changes due to environmental change, or when the background image is updated while a pedestrian is staying on the measurement line, a place where the erroneous extraction continues continuously. May exist. Therefore, the background image is locally updated. That is, in a spatio-temporal image in which continuous erroneous extraction continues as shown in FIG. 4, the boundary lines 4-1 at both ends of the location become long straight lines. The width W (4-2) and the elapsed time (length) H (4) of the region from the current time t to the immediately preceding T frame of the extraction region with a certain label L
If −3) is equal to or more than each threshold value and the values α ₁ and α ₂ (4-4, 4-5) representing the linearity at both ends obtained by the following formulas are equal to or less than certain threshold values, It is determined that a target error has occurred, and the background is updated only at that point.

[0034]

(Equation 5)

Here, x ₁ and x ₂ are the coordinate values of the left end and the right end at time t of each extraction region with the label L, respectively, and max () and min () are the objects from time t to t−T, respectively. This function calculates the maximum and minimum values in time. In addition, a method of calculating based on the curvature of the boundary line or the like can be used for the determination of the linearity.

The counting output section 2-16 outputs the counting result obtained by the counting section.

[0037]

According to the present invention, the number of objects passing through a slit position can be counted stably and accurately from a slit moving image obtained from an image of a TV camera or the like, and automatic measurement can be performed. There is.

In the prior art, if the counting error for one extraction area is large, the counting accuracy over a long period of time becomes unstable, and if the slit image is suddenly changed, accurate extraction cannot be performed thereafter. Was. Further, if the moving object and the background image information are similar, separation becomes difficult at the time of extraction processing.

[Brief description of the drawings]

FIG. 1 is a configuration diagram illustrating a moving object counting method.

FIG. 2 is a diagram showing a configuration of an embodiment of a walking person counting process.

FIG. 3 is a diagram illustrating a method of setting a count reference value and an extraction threshold used in a count processing unit.

FIG. 4 is a diagram illustrating a method of updating a local background image.

FIG. 5 is a diagram for explaining a method of setting a background described in claim 6;

FIG. 6 is a diagram for explaining a method of installing a background described in claim 7;

[Description of Signs] 1-1 Image input unit 1-2 Extraction processing unit 1-3 Background image correction unit 1-4 Parameter setting unit 1-5 Labeling processing unit 1-6 Count processing unit 1-7 Count output unit 1 8 Background 1-9 Moving Object Moving on Background 1-10 Measurement Line 2-1 Image Input Unit 2-2 Slit Image Configuration Unit 2-3 Slit Image Memory 2-4 Initial Background Image Configuration Unit 2-5 Moving Object Area Extraction unit 2-6 Background image memory 2-7 Binary image memory 2-8 Area shaping unit 2-9 Area labeling unit 2-10 Label memory 2-11 Count processing unit 2-12 Count value buffer 2-13 Parameter setting Section 2-14 Count Parameter Memory 2-15 Background Update Section 2-16 Count Output Section 3-1 Histogram of Area Value of Extracted Area within a Time 3-2,3-3,3-4 Extreme Value on Histogram 4 -1 Outline of area of continuous erroneous extraction 4-2 Width of area 4-3 Length of area 4-4, 4-5 Value indicating linearity of area

Continuation of front page (56) References JP-A-2-230475 (JP, A) JP-A-62-46388 (JP, A) JP-A-3-118677 (JP, A) JP-A-2-113374 (JP) , A) Kenji Mase, "Pedestrian Counting Using xt Spatio-Temporal Images," IEICE Technical Report, Image Processing (IE), September 1990, Vol. IE-90, No. 202, pp. 29-36 Kenji Mase, "Estimation of Object Velocity from Non-Epipolar Surface Images", Information Processing Society of Japan, March 1991, Vol. 91, No. 30, 91-CV-71-2, pp1-8 (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06M 7/00 G06T 1/00 JICST File (JOIS)

Claims (7)

(57) [Claims] 1. A processes the two-dimensional images of the one-dimensional digital signal can be time stacked direction including the fixed and moving object obtained by the imaging unit and the background, and extracts the moving objects area in the image extracting Processing unit, parameter setting unit that sets parameters used for counting processing, calculates the area of the area together with the number of areas, and moves by the sum of the integer part of the quotient of division by the count reference value determined by the parameter setting unit A counting unit for estimating the number of objects, and a geometry of the moving object region extracted by the extraction unit
Determining erroneously extracted continuous regions by examining the characteristics
A moving object counting method, comprising: a background image correcting unit that reconstructs a background image when erroneous extraction is determined , and counts moving objects. 2. A parameter setting section for estimating a count reference value in a counting section by examining a frequency distribution of an area of a region extracted by a plurality of extraction sections and obtaining an extreme value of the distribution. 2. The moving object counting method according to claim 1, wherein: 3. The count processing unit calculates the area of the region together with the number of regions, and estimates the number of moving objects based on a numerical output corresponding to a determination result based on the counting threshold value determined by the parameter setting unit. 2. The moving object counting method according to claim 1, wherein: 4. The counting processing unit calculates the area of the region together with the number of regions, and estimates the number of moving objects based on the sum of quotients of division with the counting reference value determined by the parameter setting unit. The moving object counting method according to claim 1. 5. The background image correcting section determines the erroneous extraction due to a sudden change in the background image by examining the linearity of the outline of the region extracted by the extraction processing section, and locally updates the background image. The moving object counting method according to claim 1, wherein 6. The moving object counting method according to claim 1, wherein a sheet for forming an image of uniform, uniform color and equal brightness is installed or painted on the background. 7. The moving object counting method according to claim 1, wherein an illumination that emits uniform, uniform-color, equal-luminance light is installed on the opposite side of the imaging system.