WO2016098044A1

WO2016098044A1 - Method for counting aquatic species

Info

Publication number: WO2016098044A1
Application number: PCT/IB2015/059728
Authority: WO
Inventors: Germain BOUSSARIE; Nils TEICHERT; Dominique PONTON; Raphaël LAGARDE
Original assignee: Institut De Recherche Pour Le Developpement; Ecole Normale Superieure De Lyon; Hydro Reunion
Priority date: 2014-12-18
Filing date: 2015-12-17
Publication date: 2016-06-23
Also published as: FR3030837A1; FR3030837B1

Abstract

The invention concerns a method for counting aquatic species passing through a container (1, 2, 3) provided with a camera (7) having a field of view, said method comprising a step of acquiring images originating from said camera, and a measurement step during which the objects appearing on an image are measured. The method is remarkable in that it further comprises a marking step in order to associate each object with the length of same, a monitoring step for monitoring the progress of an object in said container by means of successive images, a counting step for incrementing a counter by one unit when an object has left the field of the camera (7), or in other words when it no longer appears on the current image.

Description

METHOD FOR COUNTING AQUATIC SPECIES

The present invention relates to a method for counting aquatic species,

The field of the invention is that of the study of aquatic species, in particular that of benthic species in water courses.

Conventional methods for monitoring diadromous fish migrations from seacoast to freshwater are mainly based on spot sampling in time and space using techniques that are not optimal for study of fish movements in rivers. Underwater video has long been recognized as a good alternative but this approach usually requires a lot of work to extract the information from the video footage.

The study of migration flows first requires the counting of aquatic species.

The present invention thus relates to a video method for counting aquatic species.

The subject of the present invention is therefore a method of counting aquatic species passing through a box provided with a camera provided with a field of view, comprising a step of acquiring images from the camera, a step of the extent to which objects in an image are measured; a remarkable process in that it further comprises:

a marking step for associating each object with its length,

a follow-up step to follow the progression of an object in the caisson by means of successive images,

a counting step for incrementing a counter by one unit when an object has left the field of the camera, ie when it no longer appears on the current image. According to an advantageous embodiment of said method, it comprises a distortion correction step for each acquired image.

According to another advantageous embodiment of said method, it comprises a conversion step for transforming each acquired image into a binary image.

According to an advantageous arrangement of this embodiment, said method comprises a selection step for removing from said binary image objects having a size smaller than a first predetermined threshold or greater than a second predetermined threshold.

It may also include at least one of the following:

said method may comprise a mapping step for establishing an Euclidean distance map of an object

said method may comprise a thinning step for reducing the width of the object to one pixel in order to determine the length N.

said method may comprise an erosion step, to remove the two pixels at the ends of the object, this object having an eroded length N-2.

said method may comprise a step for calculating the integrated density of the object before erosion ID1:

NOT

ID1 = EDM (i)

said method may comprise a step for calculating the integrated density of the object after erosion ID2:

N-l

ID2 = EDM (i)

= 2

According to another advantageous embodiment of said method, the measurement step provides the estimated length L of said object:

L = N-2 + ID1-ID2. According to yet another embodiment of said method, the tracking step consists in identifying the object j in an image such as that which is closest to it in the preceding image at a proximity distance between two predetermined thresholds.

According to an advantageous arrangement of this embodiment, said counting step is implemented when the object j leaves the field of vision of the camera.

According to another advantageous arrangement of this embodiment, said method comprises a step for retaining as the length of said object the average of the estimated lengths in several successive images.

According to another advantageous embodiment of said method, when said box comprises two accesses, said counting step takes into account the direction of circulation of said object.

According to another advantageous embodiment of said method, when said box is provided with a lighting means, it further comprises a lighting step.

The present invention will now appear in more detail in the context of the following description of an exemplary embodiment with reference to the appended figures which represent:

FIG. 1, an example of a material structure necessary for the implementation of the invention, and

- Figure 2, a diagram reflecting the different steps of the method according to the invention.

With reference to FIG. 1, the material structure of the invention is presented. This is a U-shaped box with a bottom 1 honzontai and two vertical panels facing each other on opposite edges of the bottom 1, a left side panel 2 and a right side panel 3. The studied aquatic species are free to circulate between the two side panels 2, 3,

The background is light because it is the background.

For example, its dimensions are 50 cm x 50 cm. At the top of the two side panels 2, 3 is fixed a frame 4 which holds a guide tube 5 facing the center of the bottom 1. A support rod 6 is held in this guide tube, rod at the end of which is fixed a camera 7.

The frame must withstand strong currents. For example, it is made of metal. It can be disassembled easily to facilitate the transport of the box.

The caisson is fixed on the bed of the stream by means of four piles placed at the four corners of the bottom.

An umbrella (not shown) can be provided on the box to reduce the parasitic reflections of solar rays.

We can provide "arms" mesh fine mesh and tissues to channel the individuals to the entrance of the box.

The camera 7 is, for example, fixed at 30 cm from the bottom 1. It has for example a resolution of 1280 × 960 pixels and it operates, for example, at a frame rate of 30 frames per second. Its lens has a very wide angle of view.

We can also provide a lighting module (not shown).

In the first place, let us specify that a video sequence is converted into a sequence of images, a very common operation for the person skilled in the art.

With reference to FIG. 2, the method that is the subject of the present invention is now detailed.

The first step 20 is to correct the distortion introduced by the lens which is, remember, a wide angle, to obtain sufficient accuracy on length measurements. This is a common handling for the skilled person and it will not be more detailed. This correction is applied to each image extracted from the video sequence.

The second step 21 consists in detecting the aquatic species. For this purpose, several operations are necessary.

The first step is to convert each image to an 8-bit encoded image.

The second operation is to smooth each image by applying, for example, a Gaussian function with a radius of 3 pixels,

The third operation is to create a binary image, that is to say that its pixels are either 0 or 0. For this purpose, the gray level of the local maxima is detected and a noise level is set. The pixels showing a value greater than that of the local maxima decreased by the noise level are retained. The other pixels are eliminated.

Then, according to a so-called third selection step 22, a first size threshold is applied to reduce the noise and eliminate much smaller particles than the target organisms and a second size threshold is applied to remove much larger objects than the targeted organizations.

The fourth step 23 called mapping, consists in establishing a Euclidean distance map of the objects in the image. In this regard, reference is made to the article by Leymarie F. and Levine MD "Fast to Stay Scan Distance Spread on the Discrete Rectangular Lattic", CVGIP (Computer Vision Graphics and Image Processing): Image Understanding, 55 (1), January 1992, pp. 84-94. This article can be obtained on the internet at the following address:

http: //pages.cs.w sc.edu/~dyer/cs766/readings/ieymarie-cvgip92.pdf!

This step generates a matrix, or a distance map, which corresponds to the distance between each pixel and the pixel closest to the white background. The fifth step 24 called thinning step is intended to reduce the width of each object to a pixel in order to determine the length N. Reference will be made to the article by Zhang T, Y. and Suen CY of 1984 " A fast parallei algorithm for thinning digital patterns, "Image Processing and Computer Vision, 27 (3), pages 236-239. The method consists in sequentially removing pixels from the edges of an object unless the deletion of one pixel leads to dividing the object into two distinct entities.

However, there is a problem: this thinning step creates a loss of information at the ends of target organisms if they are rounded (for example the head of a fish). Some pixels are removed at this level. This lost information is recovered thanks to the Euclidean distance map.

The sixth so-called erosion step consists of removing the first and the last pixel of an object, this object having a length of N pixels now having a length of N-2 pixels.

The seventh step 26 makes it possible to calculate the integrated density 1D1 of the object before erosion:

NOT

IDl = ΣEDM (i), where

= 1

i is the current index of the pixel varying from 1 to N,

- EDM (i) is the value of the Euclidean distance map of the ith pixel.

The eighth step 27 makes it possible to calculate the integrated density ID2 of the object after erosion:

ID2 = ΣEDM (i), where

= 2

i is the current index of the pixel now varying from 2 to N-1. The ninth step 28 said measuring step provides the estimated length L of the object:

L = (N-2) + ID1 - ID2.

The tenth step 29 is the step of tracking the objects. This tracking is done by comparing the position of an object in an image to the position it occupied in the previous image. Noting (x _{t + 1} , y _{t + 1} ), the coordinates of the center of the object in the current image and (x _t , y _t ) the coordinates of the center of the object in the previous image, calculates the square of the distance separating these two positions is (x _t + i - x _t ) ² +

This operation is performed for all objects in order to establish a matrix.

The lowest value of the matrix, i.e. the shortest distance, is selected and the corresponding object is identified in each image.

Once this object is identified, we delete the row and the corresponding column of the matrix and we start again until the last object is identified.

However, it occurs periodically that an object leaves the field of the camera, so that it disappears from the image. This natural phenomenon is highlighted if an object disappears purely and simply but this disappearance can coincide with the appearance of a new object.

This situation is overcome by providing a threshold on the distance of an object between its position in the current image and its position in the previous image.

In case this threshold is crossed, it is declared that the object in question has left the field of the camera.

Thus, according to an eleventh step 30, the counting step, a counter is incremented by one unit, which counter specifies the number of objects having left the field of the camera.

Incidentally, at this stage, we can specify that the length of a given object can be averaged over the value it takes in several successive images.

Moreover, it is understood that the box has two accesses. The person skilled in the art realizes immediately that it is possible to determine ie direction of movement of an object, from front to back or the opposite. Thus, the counting step can be subdivided into two, one for a direction of circulation, the other for the other direction.

The invention has been presented according to a preferred embodiment. It applies to any affordable variant by the skilled person. In particular, any step may be replaced by an equivalent step without departing from the scope of the present invention.

Claims

1) Method for counting aquatic species passing through a box (1, 2, 3) provided with a camera (7) provided with a field of view, comprising a step of acquiring images from said camera, a measuring step (28) during which objects in an image are measured, which method is characterized by further comprising:

a marking step (21) for associating each object with its length; a tracking step (29) for following the progression of an object in said box by means of successive images;

a counting step (30) for incrementing a counter by one unit when an object has left the field of the camera (7), ie when it no longer appears on the current image.

2) Method according to claim 1, characterized in that it comprises a distortion correction step (20) for each acquired image.

3) Method according to any one of the preceding claims, characterized in that it comprises a conversion step (21) for transforming each acquired image into a binary image.

4) Method according to claim 3, characterized in that it comprises a selection step (22) for removing from said binary image objects having a size less than a first predetermined threshold or greater than a second predetermined threshold.

5) Method according to claim 4, characterized in that it comprises a mapping step (23) for establishing an Euclidean distance map (EDM) of an object,

6) Method according to claim 5, characterized in that it comprises a thinning step (24) for reducing the width of said object to a pixel in order to determine the length N. 7) Method according to claim 6, characterized in that it comprises an erosion step (25), to remove the two pixels at the ends of said object, this object having an eroded length N-2.

8) Method according to claim 7, characterized in that it comprises a step (26) for calculating the integrated density of said object before erosion (ID1):

NOT

^{ID1 =} Σ i = ^EDM C)

9) Method according to claim 8, characterized in that it comprises a step (27) for calculating the integrated density of said object after erosion (ID2):

10) Method according to claim 9, characterized in that said measuring step (28) provides the estimated length L of said object:

L = N-2 +! D1-D2.

11) Method according to any one of the preceding claims, characterized in that said tracking step (29) consists of identifying the object j in an image such as that which is closest to it in the previous image at a distance proximity between two predetermined thresholds.

12) Method according to claim 11, characterized in that said counting step is implemented when the object j leaves the field of view of the camera.

13) Method according to claim 11, characterized in that it comprises a step for retaining as length of said object the average of the lengths estimated in several successive images.

14) Method according to any one of the preceding claims, characterized in that said box having two accesses, said counting step takes into account the direction of circulation of said object. 15) Method according to any one of the preceding claims, characterized in that said box being provided with a lighting means, it further comprises a lighting step.