də edilir. Cyph
ur.
II INTERNATIONAL SCIENTIFIC CONFERENCE OF YOUNG RESEARCHERS
589
Qafqaz University
18-19 April 2014, Baku, Azerbaijan
4) Lucy ilə filmdə çəkilən aktyorların siyahısı
Q: START lucy=node:Person(name="Lucy Liu") MATCH lucy-[:ACTS_IN]->movie, co_actor-[:ACTS_IN]->movie
RETURN DISTINCT co_actor.name
R: SELECT DISTINCT co_actor.name FROM Person lucy JOIN Actor a1 ON lucy.person_id = a1.person_id JOIN Actor
a2 ON a1.movie_id = a2.movie_id JOIN Person co_actor ON co_actor.person.id = a2.person_id WHERE lucy.name =
"Lucy Liu”
Lucy-nin çəkildiyi filmlərin rejissorlarının siyahısı Q: START lucy=node:Person(name="Lucy Liu") MATCH lucy-
[:ACTS_IN]->movie, director-[:DIRECTED]->movie RETURN director.name, count(*) ORDER BY count(*) desc
R: SELECT dir.name, count(*) FROM Person lucy JOIN Actor on Person.person_id = Actor.person_id JOIN Director
on Actor.movie_id = Director.movie_id JOIN Person dir on Director.person.id = dir.person_id WHERE lucy.name = "Lucy
Liu" GROUP BY dir.name ORDER BY count(*) DESC
ENABLE MOBILE ROBOT TO RECOGNIZE AND CLASSIFY OBJECTS
Huseyn PASHAYEV, Asad MAMMADOV
Qafqaz University
h.pashayev@ictsrcqu.org , a.mammadov@ictsrcqu.org
AZƏRBAYCAN
The thesis is a survey about works of enabling mobile robot to recognize and classify objects. For recognition robot can
use haptic, statistical or visual
information like camera, which returns a cloud of points, according to which these points
segmented into clusters, where each cluster is primitive geometric structure and the geometric relationship between these
clusters defines the object.
If we are talking about recognition and classification of objects the term affordance have to be initially proposed. This
term was initially proposed by J. Gibson [1] and now widely used in robotics. So what is the affordance and how it’s used in
robotics? For each object there are a number of actions which is possible to perform with it depending on object which is
interacting with it. Now let suppose the chair, human can push the chair can sit on it, so there chair affords pushing and
sitting on it for human, however mobile robot (like Parallax Eddie – Reference Platform) can only push the chair but not sit,
so there the affordance of chair for mobile robot is only pushing. So the robot can recognize such objects then, classify the
object in classes and according to affordance of class behave with this object. Why affordance is so important? Suppose a
robot which helps us at home. The robot need to be able to recognize object then classify whether this is a chair class, or
table class or even door class, then robot have to know how to behave with object, whether to pull it( with chairs class), go
across(with tables class) or just pass through it(door class). In such problems main aim is not to detect object preciously, the
aim here is to detect the correct class to which this object belong to, such that general properties of affordance for objects in
same class are the same. Nowadays the field object recognition and classification is widely investigating and there a lot of
works done in this field. General object recognition rely on visual appearance of object, where invariant descriptors used to
recognize and classify object. In other words the snapshots of object are taken from different positions, then local and global
descriptors are used to summarize data and finally distance between vectors of these descriptors is calculated to model the
object. In article [2] the robot has 2D camera and uses local descriptors and SIFT (Scale-invariant feature transform)
features. In SIFT the algorithm is to identify local features of training object and then detect this object on image containing
this and many other objects. Another main property in SIFT is the solid proportion of selected points as feature. Suppose we
take 4 corners of door as one feature of this object, so in case if the proportion between these points will change then this
will cause failure. Like if door is opened or closed for each position of the door proportion between these points will
change, what will cause failure, that means on flexible and articulated objects this won’t work because geometry changes
what causes the proportion change on points which are selected as feature. However on practice SIFT method uses large
numbers of features which reduce the error caused by such local variations. In article [3] the robot controlled via internet
where user give voice commands to robot like “Grasp the ball” and etc. In this work neural network proposed which help to
classify object in low resolution and noise images using invariant descriptors. Another article is [4]. In this article proposed
to use multiple perception of object and motor modality. The robot using statistical technique, and using motor function
grasps the object then rotating it take snapshots from different points and also while observation listen sounds from object,
based on audio-video and haptic information robot recognize and classify object. Another work about classification of
objects is “Grounding semantic categories in behavioral interactions: experiments with 100 objects” [5] In this article 100
object taken and divided in 20 categories in order to test the humanoid robot, where for each category different behavior
assigned. Robot uses supervised recognition method where he process not only visual information but also taking in account