Hierarchical Object based Representation

Table 3.1. An attempt to compare the different representation methods. Xindicates that the method is elegant and appropriate. 4 indicates that extra work is needed or the method is inapplicable.

Representations Feature-based Grid-based Direct

Uncertainty management X X 4

Loop closing mechanism X 4 4

Sensor characteristics 4 X X

Environment representability 4 X X

Data Compression X 4 4

In outdoor or urban environments, features are extremely difficult to define and ex-tract because both stationary and moving objects do not have specific sizes and shapes.

Therefore, instead of using an ad hoc approach to define features in specific environments or for specific objects, free-form objects are used.

At the preprocessing stage, scans (perception measurements) are grouped into seg-ments using a simple distance criterion. The segseg-ments over different time frames are in-tegrated into objects after localization, mapping and tracking processes. Instead of using track in tracking terminology, segment is used because of the perception sensor used in this work. Because not only moving targets but also stationary landmarks are tracked in the whole process, the more general term, object, is used instead of the term, target.

Registration of scan segments over different time frames is done by using the direct method, namely the ICP algorithm. Because range images are sparser and more uncertain in outdoor applications than indoor applications, the pose estimation and the correspond-ing distribution from the ICP algorithm are not reliable. For dealcorrespond-ing with the sparse data issues, a sampling-based approach is used to estimate the uncertainty from correspon-dence errors. For dealing with the uncertain data issues, a correlation-based approach is used with the grid-based method for estimating the uncertainty from measurement noise.

For loop closing in large environments, the origins of the object coordinate system are used as features with the mechanism of the feature-based approaches.

Our approach is hierarchical because these three main representation paradigms are used on different levels. The direct method is used on the lowest level and the feature-based approach is used on the highest level. Objects are described by a state vector, or object-feature and a grid map, or object-grids. Object-features are used with the mechanism of the feature-based approaches for moving object tracking and for loop closing. Object-grids are used to to take measurement noise into account for estimating object registration uncertainty and to integrate measurements over different time frames. Figure 3.5 shows an example of the hierarchical object based representation.

3.2 HIERARCHICAL OBJECT BASED REPRESENTATION

Figure 3.5. Hierarchical object based representation. The black solid box denotes the robot (2mx5m).

In this section, scan segmentation, sensor noise modelling and sparse data issues will be described. The sampling and correlation based approach for estimating the uncertainty of object registration will be addressed in Section 3.3. The hierarchical object based repre-sentation for moving object tracking and for SLAM will be addressed in Section 3.4 and Section 3.5 respectively.

Scan Segmentation

Scan segmentation is the first stage of the hierarchical object-based approach. (Hoover et al., 1996) proposed a methodology for evaluating range image segmentation algorithms, which are mainly for segmenting a range image into planar or quadric patches. Because objects in outdoor environments do not have specific sizes and shapes, these algorithms are not suitable.

Here we use a simple distance criterion to segment measurement points into objects.

Although this simple criterion can not produce perfect segmentation results, more precise segmentation will be accomplished by the localization, mapping and moving object track-ing processes ustrack-ing spatial and temporal information over several time frames. Figure 3.6 shows an example of scan segmentation.

Figure 3.6. An example of scan segmentation.

Perception Sensor Modelling

It is well known that several important physical phenomena such as the material properties of an object, the sensor incidence angle, and environmental conditions affect the accuracy of laser scanner measurements. Although laser rangefinders such as SICK laser scanners provide more accurate measurements than sonar, radar and stereo cameras, neglecting measurement noise in the localization, mapping, and moving object tracking processes may be over optimistic in situations using data collected from a platform at high speeds in outdoor environments.

According to the manual of SICK laser scanners (Sick Optics, 2003), the spot spacing of SICK LMS 211/221/291 is smaller than the spot diameter for an angular resolution of 0.5 degree. This means that footprints of consecutive measurements overlap each other.

The photo in Figure 3.7 taken from an infrared camera shows this phenomenon. A red rectangle indicates a footprint of one measurement point.

With regard to range measurement error, we conservatively assume the error as 1% of the range measurement because of outdoor physical phenomena. The uncertainty of each measurement point z_kⁱ in the polar coordinate system is described as:

3.2 HIERARCHICAL OBJECT BASED REPRESENTATION

Figure 3.7. Footprints of the measurement from SICK LMS 291

Σ_zi

k =

· σ²_ri 0 0 σ²_θi

¸

(3.1) The uncertainty can be described in the Cartesian coordinate system by the head-to-tail operation described in Section 2.1. Figure 3.8 shows the SICK LMS 211/221/291 noise model.

−20 −15 −10 −5 0 5 10 15 20

−5 0 5 10 15 20

meter

meter

10 11 12 13 14 15

1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

meter

meter

Figure 3.8. SICK LMS 211/221/291 noise model. Left: the whole scan. Right: the enlargement of the blocked region on the left. The distributions of the measure-ment points are shown by 2σ ellipses (95% confidence).

In most indoor applications, it is assumed that a horizontal range scan is a collection of range measurements taken from a single robot position. When the robot is moving at high speeds, this assumption is invalid. We use the rotating rate of the scanning device and the velocity of the robot to correct the errors from this assumption.

Sparse Data

Compared to indoor applications, the distances between objects and sensors in out-door environments are usually much longer, which make measurements more uncertain

and not as dense. Sparse data causes problems of data association in the small, or correspon-dence finding, which directly affect the accuracy of direct methods. In the computer vision and indoor SLAM literature, the assumption that corresponding points present the same physical point is valid because data is dense. If a point-point metric is used in the ICP algorithm, one-to-one correspondence will not be guaranteed with sparse data, which will result in decreasing the accuracy of transformation estimation and slower convergence.

Research on the ICP algorithms suggests that minimizing distances between points and tangent planes can converge faster. But because of sparse data and irregular surfaces in outdoor environments, the secondary information derived from raw data such as surface normal can be unreliable and too sensitive. A sampling-based approach for dealing with this issue will be addressed in the next section.