DE102012205000B3 - Method for counting objects on conveying belt, involves varying displacement vector, and counting objects in one of object lists, where positions of objects are not overlapped according to shifting around optimum vector - Google Patents

Abstract

The method involves creating object lists with positions of objects (30) detected in an image. An optimum displacement vector between two detected images by comparison of the object lists is determined, where the displacement vector is provided in the object lists that contain object size and/or object color. The displacement vector is varied until an optimum displacement vector with tolerated deviation measure is found. The objects in one of the object lists are counted, where positions of the objects are not overlapped according to shifting around the optimum vector. An independent claim is also included for a camera comprising an evaluation unit to perform a method for counting objects on a conveying belt.

Description

The invention relates to a method and a device for counting objects with a camera according to the preamble of the independent claims.

Such methods are known in many variations. For example, the EP 1 856 971 B1 an egg counting sensor, which has arranged for detecting and counting the objects a plurality of detectors in two different rows. The detectors detect the contours of an object and determine the direction of movement of each individual object relative to the detectors. A counter is set high when a previously unidentified object has moved in a predetermined direction of entry into the detection area of the detectors and has been identified.

From the DE 28 26 959 A method of image processing moving objects on an assembly line is known in which a detection area within which objects can be detected is advancingly controlled during a part of the measurement period of time at a speed corresponding to the object speed. After recognizing the object and in particular an object identification and position measurement, the image frame is switched back to its initial position. According to the detected objects and position positions, a following robot is controlled.

Furthermore, from the US Pat. No. 7,932,485 B2 a device for detecting geometric data of objects known in which objects are first detected on a conveyor belt with a high-resolution camera to capture all relevant geometry data. For passing on to subsequent systems, the captured objects are reduced to easily representable geometric data. The geometry data can be output for different sections of the conveyor belt.

The object of the invention is to improve the detection and counting of objects on a conveyor belt.

The object is advantageously achieved by the method according to the invention and a suitably equipped camera. The method provides to monitor a conveyor belt with a camera, wherein in a first step with the camera images of a conveyor belt section, preferably at a constant frame rate, are recorded. For each image, an object list is created with the positions of the objects recognized or captured in the image.

By comparing the object lists or the positions of the objects in these lists, an optimal displacement vector is determined. In a final step, the objects in the object lists are counted whose positions do not overlap with the object positions in the subsequent object list after shifting around the optimal displacement vector.

This procedure has the advantage that, without speed information from the conveyor belt, the material transport on the conveyor belt can be monitored and counted on the basis of the images captured by the camera alone.

In an advantageous embodiment, first a first displacement vector is defined within defined selection possibilities for the determination of the displacement vector. By adding the displacement vector to the object positions of the object list, an object position to be expected in the following image is calculated. These expected object positions are compared with the object positions detected in the following image, a deviation measure being determined as a function of this position comparison. The displacement vector is varied until a displacement vector with a smallest deviation measure is found. The further determination of the objects to be counted is based on this displacement vector.

This approach has the advantage that in a simple and fast manner, a displacement vector can be found with an accuracy sufficient for the counting of objects.

For better recognition of the objects, it may be provided to add the object size and / or the object color as additional object parameters in the object list. Other object parameters conceivable in a non-exhaustive list are: roundness, compactness, rotation, width and / or height.

In a further advantageous embodiment, it is provided that the displacement vector is determined from a vector difference between object positions of consecutive object lists. By doing so, the number of possibilities to be varied is limited to a well-defined set of vectors.

By adding the displacement vector with at least one further object position of the object list, we calculate at least one expected object position. This object position is compared with object positions of the subsequent object list and a deviation measure is calculated as a function of this comparison.

If the calculated deviation measure exceeds a tolerated deviation measure, the Shift vector varies within the scope of the choices until a displacement vector is found with a deviation measure within the tolerance limits. The displacement vector whose deviation measure is smaller than the tolerated deviation measure forms an optimal displacement vector and is used for the determination of the objects to be counted below.

Advantageously, a camera is provided which has an evaluation unit, which is designed to carry out the aforementioned method.

Preferably, the evaluation unit of the camera is designed in such a way that, based on the images captured by the camera, an object list is created for each image and stored in a memory, wherein an optimal displacement vector is determined by comparing successive object lists. In each object list, the objects are counted whose object positions do not coincide or overlap with object positions in the following object list after a shift around the optimal displacement vector.

The invention will be explained in more detail by means of embodiments with reference to the drawings.

They show schematically:

1 a camera for monitoring a conveyor belt,

2 a look at the conveyor belt at different times,

3 two consecutive pictures of the conveyor belt situation,

4 an image overlap with an optimal displacement vector,

5 an overlap with too large a displacement vector,

6 a calculated displacement of a single object from the list of objects,

7 a flow diagram of the method according to the invention.

In the following description of the preferred embodiments, like reference numerals designate like or similar components.

1 shows a conveyor belt 40 which transports a plurality of objects at a conveyor speed v. Above the conveyor belt 40 is a camera 10 arranged, which is a subarea 20 of the conveyor belt 40 supervised.

2 shows a plan view of the conveyor belt according to 1 at different times. The left picture shows eight objects in the surveillance area 20 and are roughly in three lines 30.1 . 30.2 . 30.3 arranged. Outside the surveillance area 20 there is another line 30.4 not yet detected by the camera. In the right image, taken at a later time, is the first row of objects 30.1 has already been transported out of the detection area, while the fourth row not visible in the last image 30.4 now from the surveillance area 20 is detected.

3 shows two consecutive pictures n, n + 1 of in 2 illustrated situation. The two images differ by the displacement of the objects in relation to the tape speed and time difference between the two shots. An example is the object displacement based on the object 30.3 shown. The position of the object 30.3 is shifted according to the tape speed by a shift vector x in the subsequent picture. With knowledge of this displacement vector x, all objects of the image n can be displaced by this vector x, so that the probable position of these objects in the subsequent image n + 1 can be predetermined.

Such a displacement of these objects is in 4 shown. The first image n is shifted exactly by the displacement vector x with respect to the subsequent image n + 1. The first object series 30.1 of the first image n is due to the shift outside the image window of the second image n + 1 in a common monitoring area U are the objects of the second and third row 30.2 . 30.3 to cover. According to the invention, the objects which are located outside of the subsequent image n + 1 are counted. The shifted area spanned by the displacement vector x can thus also be understood as a counting range Z.

Basically equally effective, it may also be provided according to the invention, the newly added in the second image n + 1 objects 30.4 to count. Also this area is spanned in its height by the displacement vector x.

5 shows a situation in which the displacement vector x was chosen too large, so that the objects in the overlapping area U do not coincide.

In order to avoid such failures, it is further provided according to the invention to select only displacement vectors x, which result from the position difference of the two consecutive object lists OL _n , OL _{n + 1} . x → ε {(P _{i, n} -P _{j, n + 1} )}

This procedure is exemplified in 6 based on two objects 30.2 . 30.3 shown. In the example shown, an object was created 30.Xi _{n + 1} in the subsequent image n + 1 as a reference object for determining a displacement vector x selected. x → = P (30.Xi _{n + 1} ) -P (30.2 _n )

If the object P (selected in the following list OL _{n + 1)} 30.Xi _{n + 1} ) actually the moved object 30.2 from the first list OL _n , all other object positions also differ by exactly this displacement vector x.

In order to check whether the determined displacement vector x corresponds to the actual displacement vector x _opt , it is examined whether the remaining objects can also be imaged on themselves in the following image n + 1. For the actual or optimal displacement vector x, x _opt : ΔP = ΣΔP _i = Σ (| P '/ 1 - P _{i, n + 1} |) = Σ | (P _{i, n} + x →) - P _{i, n + 1} | = 0 or ΔP <ΔP _tol

The determination of the deviation can of course also be determined in a different way.

In the case shown, the object becomes 30.3n from the position P _{3, n} by the displacement vector x shifted to the position P _{'3, n.} Like in the 6 shown, comes the object shifted by x 30.3'n not with an object 30.Xj _{n + 1} from the following object list OL _{n + 1} to cover. The calculated displacement vector x is therefore not a solution for all other objects, and the following applies: ΔP> ΔP _tol

The displacement vector x is now varied within the solution set until the difference ΔP of the predicted and the detected position is as small as possible or equal to zero.

7 shows a flowchart of the method according to the invention. As already described, the camera monitors an area of the conveyor belt 40 and preferably records images of the object situation on the conveyor belt at a constant time interval. In step 101 images are captured continuously and in a further step 102 created for each image object lists OL _n , OL _{n + 1 of} the detected objects. Based on these object lists, in a step b all object positions P _{i, n} in the first object list OL _n are shifted by a displacement vector x _j provided in step a. In the following steps c, a difference of the positions from the first and the subsequent object list is determined. From the difference, a deviation measure ΔP or a probability measure w can be determined. The smaller the deviation .DELTA.P or the greater the match probability w, the greater the probability that the shift vector _{x.sub.j has been} optimally selected. In a step d, it is checked whether an optimal displacement vector x _{opt has} already been achieved. If the deviation ΔP has not yet reached its optimum value over one step 105 the object displacement of the first list OLn is carried out with a new displacement vector x until an optimal displacement vector x _opt is present. If such a vector is present, the object count is performed with this optimal displacement vector x _opt in a step e. These process steps are accomplished by incrementing the images in the step 103 for all subsequent pictures.

The optimum displacement vector x _opt found for the first image sequence n, n + 1 is preferably used as start vector x _j for the subsequent image sequences.

Claims (7)

Method for counting objects on a conveyor belt with a camera, comprising the steps of: - taking pictures of a conveyor belt section ( 20 ) with a camera ( 10 ), - creating an object list (OL) with positions (P _i ) of the objects detected in an image, - determining an optimal displacement vector (x _opt ) between two successive acquired images (n, n + 1) by comparing the object lists (OL _n , OL _{n + 1} ) by specifying a first displacement vector and varying it until an optimal displacement vector with a tolerated deviation measure is found, - counting the objects in one of the object lists (OL _n , OL _{n + 1} ) whose positions ( Pi) after shifting around the optimal displacement vector (x _opt ) do not overlap. Method according to Claim 2, in which a first displacement vector (x) is initially specified within defined selection possibilities, - in a further step, the object positions are added to the displacement vector (x) in order to calculate an expected object position (P ' _{i, n} ), The expected object positions (P) are compared with the object positions (P _{i, n + 1} ) of the subsequent object list (OL _{n + 1} ), and a deviation measure (ΔP) is determined as a function of the deviations (ΔP _i ), - the displacement vector ( x) is varied as long as possible within the selection options optimal displacement vector (x _opt ) with tolerated deviation measure (ΔP <ΔP _tol ) is found. Method according to one of the preceding claims, in which a) the displacement vector is determined from a vector difference between object positions (P _{i, n} , P _{i, n + 1} ) of consecutive object lists (OL _n , OL _{n + 1} ), b) calculation at least an expected object position (P ' _i ) by adding the displacement vector (x _j ) with at least one further object position (P _{i, n} ) of the object list (OL _n ), c) comparing the at least one expected object position (P) with object positions (P _{i , n + 1} ) of the subsequent object list (OL _{n + 1} ) and calculation of a deviation measure (ΔP) as a function of this comparison, d) comparison of the deviation measure (ΔP) with a tolerated deviation measure (ΔP _tol ), wherein when the tolerated deviation _{measure is} exceeded (ΔP _tol ), the steps a) to d), while selecting a further displacement vector (x _{j + 1} ), within the possible vector differences, are performed until the deviation measure (ΔP) the tolerated A bweichungsmaß (.DELTA.P _tol) below, wherein the displacement vector (x), the deviation (.DELTA.P) is smaller than the tolerated deviation (.DELTA.P _tol) an optimal displacement vector (x _opt) forms, e) using this displacement vector (x _opt) for the determination of the objects to be counted. Method according to one of the preceding claims, in which the objects of the previously created object list (OL _n ) are counted, whose position (P ' _i ) displaced by the optimum displacement vector (x _opt ) no longer lies in the subsequently acquired image (n + 1) , Method according to one of the preceding claims, in which the object lists (OL _n , OL _{n + 1} ) additionally contain the object size and / or object color as object parameters Camera with an evaluation unit, which is designed to carry out the method according to one of the preceding claims. Camera according to claim 6, wherein the evaluation unit is configured such that based on the captured by the camera images for each image an object list (OL _n ) is created and stored in a memory, wherein by comparing successive lists of objects (OL _n , OL _{n +1),} an optimum motion vector (x _opt) is determined, and (in each object list OL _n) the objects to be counted, the object positions (P _{i, n)} according to a displacement of the optimum displacement vector (x _opt) not (with object positions P _{i, n + 1} ) in the subsequent object list (OL _{n + 1} ) overlap.

Images