RU2606370C1 - Method for segmentation of laser scans and system for its implementation - Google Patents

Abstract

FIELD: data processing.

SUBSTANCE: invention relates to processing digital data basing on using scanning devices and can be used for laser scans (LS) in systems for positioning and navigation of self-contained loaders. For this purpose the process of segmentation of a two-dimensional LS includes the stages of: the LS deployment into a one-dimensional signal (array of values); processing the obtained array using numerical differentiation of the array values; obtaining an array of values of a differential and the second differential, which is divided into segments in the form of an array of arrays of values, wherein every said segment is a set of points belonging to a separate object on the LS; determining for each segment the extreme points describing the object; determining angles from the LS origin point to each of the said extreme points of the object (extreme angles); determining values of the object central angles using the obtained values of extreme angles by calculating the average value; creating a list, which includes angle values from the LS origin point to each of the extreme points, the central angle and a corresponding to the object set of points describing the object.

EFFECT: faster operation.

4 cl, 4 dwg

Description

Technical field

The claimed group of inventions relates to the field of digital data processing, in particular, laser scans (LS), which are used in navigation systems of autonomous loader machines when moving indoors.

A known method of segmentation and selection of geometric primitives from 2D laser scans for use in applications designed for robotic machines (Cristiano Premebida, Urbano Nunes "SEGMENTATION AND GEOMETRIC PRIMITIVES EXTRACTION FROM 2D LASER RANGE DATA FOR MOBILE ROBOT APPLICATIONS").

In the known method, the boundaries of the objects are determined by a slow sequential comparison of the distances between two adjacent points using a modified Kalman filter. The main disadvantage of the known solution is the slow speed when determining the boundaries of the object, which consists in comparing segments between frames.

Disclosure of invention

The problem to which the claimed invention is directed, is to create a method for segmenting laser scans, allowing at a higher speed to determine the boundaries of the objects presented on them.

The technical result is to increase the speed of segmentation of drugs and the speed of searching for correspondences between objects in successive frames.

The technical result is achieved due to the method of segmentation of a two-dimensional laser scan (LS), containing stages in which:

- deploy the drugs into a one-dimensional signal (array of values);

- process the resulting array using numerical differentiation of the array values;

- get an array of differential values and the second differential, which is divided into segments in the form of an array of arrays of values, and each said segment is a set of points belonging to a separate object on the LAN;

- determine for each segment the extreme points characterizing the object;

- determine the angles from the reference point of the drug to each of the mentioned extreme points of the object (extreme angles);

- carry out the determination of the values of the central angles of objects using the obtained values of the extreme angles by calculating the average value;

- create a list that includes the values of the angles from the reference point of the drug to each of the extreme points, the central angle and the corresponding set of points that characterize the object.

The technical result is also achieved due to the method of tracking recognized objects on the drug in time, containing stages in which:

- carry out the segmentation of drugs using the method according to p. 1 and get a new list of objects;

- check if the current list of objects is empty;

- if the current list of objects is empty, then all objects from the new list are given unique UIDs and all objects together with the UID are added to the current list, otherwise they compare objects from the current and new lists using the nearest neighbor algorithm, in which the current and new lists are compared containing the values of the central angles of the objects from the starting point of the drug, and get a list of correspondences between the objects in the current and new lists;

- check if the new list of objects is empty, and if it is empty, then delete the remaining objects in the current list that were not mapped to any object in the new list, otherwise

- select an object from the new list and check whether this object corresponds to any object from the current list and if such an object exists in the current list, then update the parameters (set of points, extreme angles, central angle) of this object in the current list based on the parameters of the object in the new list, otherwise the object is considered to appear in the field of view, it is assigned a new UID and the object is added to the current list;

- remove the object from the new list and iteratively repeat the steps of the method from the step of checking whether the current list of objects is empty until the new list is empty and all unmatched objects are removed from the current list.

The claimed result is also achieved by the system for segmenting a two-dimensional LAN containing at least one processor, at least one memory means containing machine-readable instructions that, when executed, prompt at least one processor to execute the aforementioned LAN segmentation method.

Description of graphic images

FIG. 1 illustrates the major steps of the steps of a method for recognizing objects on consecutive drugs.

FIG. 2 illustrates a general flowchart when performing a segmentation method of a drug.

FIG. 3 illustrates a general flowchart for tracking objects.

FIG. 4 illustrates a system for performing the claimed method.

The implementation of the invention

According to FIG. 1, the claimed invention includes three main steps that are performed sequentially. At step 100, the drug is obtained for its subsequent segmentation at step 200, the definition of objects on the drug, and their subsequent analysis and maintenance at step 300.

Each scan point is represented as a single value - the distance of the point from the scanner. Points are sorted by increasing scan angle. According to FIG. 2, when the segmentation method 200 is implemented in step 201, the drug is expanded (converted) into a one-dimensional signal (array of values). The segmentation method is intended for quick analysis of drugs and the allocation of the boundaries of objects in it in order to use them for identification and subsequent tracking.

After deploying the LAN (step 201), at step 202, the signal (array of values) obtained at step 100 is processed, for which numerical differentiation of the array values is performed, as well as repeated numerical differentiation of the differential array. Two new arrays of values are created, shorter, respectively, by one and two elements than the original LAN. In the obtained first array 2021, the value of the difference in the remoteness of a point with the same index and a point with an index is written one more. In the second array 2022, the value of the difference of the value from the first array 2021 with the same index and the values from the first array 2021 with the index are greater by one. In the two arrays of differential and second differential received, the values are compared with the given thresholds for each array. Points from the array of drugs corresponding to values that exceed the threshold in two other arrays 2021 and 2022 are considered the boundaries of the object, the points lying between them are collected in separate arrays corresponding to different objects.

At step 203, extreme points characterizing the object are determined for each segment, which are, in particular, the extreme left and rightmost points from the set of points characterizing the object. Since the array of drugs is sorted, such points will be the first and last points from the segment as the beginning and end of the object.

At step 204, for each object detected on the drug, the angles from the reference point of the drug to the extreme points are calculated.

At step 205, the values of the central angles of the objects are determined using the extreme angles obtained at step 204 by calculating the average value.

At step 206, a new list of objects is created, which includes information obtained in the previous steps, in particular, the angles from the reference point of the drug to each of the extreme points, the central angle, and the set of points that characterize the object corresponding to the object.

In FIG. 3 shows a sequence of steps of a method 300 used to track objects on a drug. The method 300 uses the data obtained by the method of segmentation 200.

At step 301, it is checked whether the current list of objects is empty.

If the current list of objects is empty, then proceed to step 302 where unique objects are issued to all objects from the new list, and all objects together with the MIA are added to the current list. Otherwise, proceeds to step 303.

At step 303, objects from the current and the new list are compared using the nearest neighbor algorithm (see below), in which the current and new lists containing the values of the central angles of the objects from the starting point of the drug are compared. A new list was obtained by implementing the method of segmentation 200. After comparison, a list of correspondences between the objects in the current and new lists is obtained.

Next, at step 304, it is checked whether the new list of objects is empty, and if it is empty, then at step 305 the objects remaining in the current list that were not mapped to any object in the new list are deleted. These are the objects that were not found on the new drug. Those. it is believed that these objects were lost from sight and should no longer be accompanied.

If the new list is not empty, then at step 306 an object is selected from the new list, and at step 307, the existence of this object is checked for any object from the current list. If a similar object exists in the current list, then at step 308 the parameters (set of points, extreme angles, center angle) of this object in the current list are updated based on the parameters of the object in the new list. If such an object cannot be found in the current list, then the object is considered to have appeared in the field of view. Next, at step 309, a new UID is assigned to it and the object is added to the current list.

Next, at step 310, the object is removed from the new list and the condition is checked again at step 304.

Thus, objects from the new list are sequentially selected, compared with objects from the current list, and removed from the new list. After a certain number of iterations, the new list will be empty and will go to step 305.

The nearest neighbor algorithm, applied to the new and current lists, builds comparisons between the objects in them.

наименьший среди всех j=[1…m]. Поиск делается перебором всех значений j. Отсюда определяется, что a(i*) и b(j*) - ближайшие соседи. Поскольку количество элементов в последовательностях может быть разным, некоторые элементы могут не получить ближайшего соседа.In the general case, the nearest neighbor algorithm is as follows, there are two lists, a and b, with the number of elements, respectively, n and m. For each element of the sequence a (a (i *); i = [1 ... n]) there is an element b (j *); j = [1 ... m] such that

smallest among all j = [1 ... m]. The search is done by iterating over all j values. This determines that a (i *) and b (j *) are the nearest neighbors. Since the number of elements in the sequences may be different, some elements may not get the nearest neighbor.

среди всех j=[1…m] делается с помощью предварительного создания k-d дерева по списку b и дальнейшего использования этого дерева. Т.е. делается несколько уровней кластеризации по углу для списка b, которые отличаются размерами кластеров. Получается дерево кластеров, в котором каждый кластер на определенном уровне содержит 2 или более кластеров уровнем ниже. Поиск по такому дереву кластеров будет происходить быстрее при большом количестве объектов в последовательности b.In another particular case, the search for the smallest

among all j = [1 ... m] is done by first creating a kd tree from list b and then using this tree. Those. several angular clustering levels are made for list b, which differ in cluster sizes. The result is a cluster tree in which each cluster at a certain level contains 2 or more clusters at a lower level. Searching for such a tree of clusters will occur faster with a large number of objects in sequence b.

Consistently applying the approach described above to each new drug, it is possible to accompany objects (to know points belonging to the same object) throughout the entire time the object is in the field of view of the scanner. In FIG. 4 is a perspective view of a system 300 for performing drug segmentation procedures.

System 400 includes one or more computing processors 410, one or more memory means 420 (RAM, ROM, NAND, EEPROM, etc.), one or more means for storing data 430 (HDD, SSD, flash drives, and etc.), interfaces 440 and input / output devices 450, communication devices 460 (LAN, WLAN, irDa, etc.) connected via a common data bus 470.

System 400 may be implemented on the basis of widely known IBM PC compatible devices, such as a PC, laptop, mainframe, supercomputer, etc., or may be based on mobile computing devices, such as a tablet or mobile phone.

System 400 may also be part of a robotic machine to enable autonomous navigation of indoor machines.

Claims (17)

1. A method for segmenting a two-dimensional laser scan (LS), comprising stages in which: - deploy the drugs into a one-dimensional signal (array of values); - process the resulting array using numerical differentiation of the array values; - get an array of differential values and the second differential, which is divided into segments in the form of an array of arrays of values, and each said segment is a set of points belonging to a separate object on the LAN; - determine for each segment the extreme points characterizing the object; - determine the angles from the reference point of the drug to each of the mentioned extreme points of the object (extreme angles); - carry out the determination of the values of the central angles of objects using the obtained values of the extreme angles by calculating the average value; - create a list that includes the values of the angles from the reference point of the drug to each of the extreme points, the central angle and the corresponding set of points that characterize the object. 2. A system for segmenting a two-dimensional LAN, containing at least one processor, at least one memory tool containing machine-readable instructions that, when executed, prompt at least one processor to execute the method of claim 1. 3. A method for tracking recognized objects on a LAN in time, comprising stages in which: - carry out the segmentation of drugs using the method according to p. 1 and get a new list of objects; - check if the current list of objects is empty; - if the current list of objects is empty, then all objects from the new list are given unique UIDs and all objects together with the UID are added to the current list, otherwise they compare objects from the current and new lists using the nearest neighbor algorithm, in which the current and new lists are compared containing the values of the central angles of the objects from the starting point of the drug, and get a list of correspondences between the objects in the current and new lists; - check if the new list of objects is empty, and if it is empty, then delete the remaining objects in the current list that were not mapped to any object in the new list, otherwise - select an object from the new list and check whether this object corresponds to any object from the current list and if such an object exists in the current list, then update the parameters (set of points, extreme angles, central angle) of this object in the current list based on the parameters of the object in the new list, otherwise the object is considered to appear in the field of view, it is assigned a new UID and the object is added to the current list; - remove the object from the new list and iteratively repeat the steps of the method from the step of checking whether the current list of objects is empty until the new list is empty and all unmatched objects are removed from the current list. 4. The method according to p. 3, characterized in that in the nearest neighbor algorithm, a k-d tree is created, followed by a search in it.

Images