Information Technology Reference
In-Depth Information
initial location; a robot can update its new location with limited information from
odometer and fine-tune the location with the help of a laser rangefinder. Second type
is the robot does not know the initial location. This is a complicate problem since the
robot might be in any possible location. This problem is sometimes called the global
localization [6], [9], [13,14,15].
The global localization is matching the information gathered by the sensors on a
robot with the information stored in a given map. Finding the best matched features
leads to finding the most probable current location of the robot. Thus, it is necessary
to have a pre-build map for global localization. The more information a map
possesses, the higher accuracy a robot can achieve. For example, laser rangefinders
are often used to scan the distance of each angel from a robot, and these distance
values will be used to match with the stored values of each candidate point [9,10],
[17], [21]. The information associated with all the candidate points is collect
previously, which is a laborious task. Similarly, camera can be used as the sensor; it
can be used to detect objects [11] or building topological maps with images [5], [12],
[20]. Sensors are also used for map building, such as sonar and infrared sensor [7],
inertia sensor [16], camera [19], and laser rangefinder [19], [23].
In previous works, the feature matching method on global localization works well for
static environment [6], [9]. However, it does not work well in the real world dynamic
environment, because the position of objects might be changed, or people might be
walking around. To cope with the complicated dynamic environment, the feature
matching method has to be refined to be able to recognize the moving objects. Recognize
moving object can be done by camera and image processing method [3], [11]. However,
photographing is not always agreeable in house because it may involve personal privacy.
In this paper, we use only the information provided by laser rangefinder. We
improve the signature-rs method proposed in previous work [9] and propose a new
multi-group feature matching scheme. We collect information from all possible
360
°
directions, and group them into several sets of information which only use
180
°
just
as in previous work. Therefore, our method is robust since the moving object will
affect only some of the information sets, not all of them. To reduce the cost of
calculation, we also adopt the method in previous to group similar features. [8]
The paper is organized as the following. Section 2 introduces how the features are
generated and our feature grouping method. Section 3 shows that feature matching
with real-time localization can be improved with robot moving. The 4th section
reports the experimental results and the comparison between static and dynamic
environment. Section 5 is discussion and conclusion.
2
Feature Generation
2.1
Global Hypothetical Feature
In this paper, our assumption is that there is no additional tag for the robot in the
environment. The robot use only the information gathered from the laser rangefinder to
identify its own location. Since the compass is not very accurate in indoor environment,
we also assume that the robot is without compass, i.e., the orientation is unknown.
Search WWH ::
Custom Search