AUTOMATIC TARGET IDENTIFICATION FOR LASER SCANNERS

Medicilon’s Chemistry department has more than 100 chemists, who are experienced in the cooperation with major domestic and international pharmaceutical and biotech companies. Our services cover a variety of research interests in novel drug research, including target validation, hits evaluation, lead optimization, candidate nomination, preclinical development and IND filing. Email:marketing@medicilon.com.cn web:www.medicilon.com

ABSTRACT: Terrestrial laser scanners are becoming increasingly important for many fields of imaging applications, providing a great amount of 3D positional information in a fast and efficient way. This information is always expressed by means of coordinates in a somewhat random 3D space defined by the scanner orientation, which changes whenever the scanner is moved. Therefore, targets are usually employed either for registration (i.e. for the referencing of the data in a common 3D space) or for referencing of the data into a local coordinate system. The use of targets for these purposes is a standardized process, which is invariably carried out by proprietary software. However, the algorithms used for the identification of targets (i.e. automated definition of the centre of the target) are not described by the software vendors. In this paper, methods for automating target identification which are based on fuzzy classification, gridding and averaging techniques are presented. Experiments are conducted using a Cyrax 2500 terrestrial laser scanner in laboratory conditions. The performance of the proposed methods is compared and assessed with reported methods from published literature. Furthermore, given the fact that due to reflectance topographic artefacts are observed on the surface of the reflective targets, experiments are also conducted for different scan angles and distances. 1. INTRODUCTION Terrestrial laser scanning allows for detailed and precise documentation of objects of interest. In practice, collection and processing procedures are adapted to the type of application (e.g. use of different resolutions, acquisition of multiple, overlapping scans from different distances, points of view). However, regardless of the application (e.g. conducting metrological experiments, registering multiple scans, referencing the position of the data in a given coordinate system etc.), automatic target identification is a matter of great significance. Therefore, the need for a reliable and precise algorithm that identifies targets automatically is important. In this paper, the capabilities of a current commercial laser scanner system (Cyrax 2500) regarding target identification are explored and several new methods for automatic target identification are presented. The second section of the paper gives a brief overview of the Cyrax 2500 system and presents two experiments conducted for evaluation of the repeatability of collected data from multiple scans. In the third section, the way that target centres are determined using the Cyclone software is described along with several methods for target identification proposed in published literature. The properties of the reflective targets are thoroughly examined and new algorithms for target identification are described. In the final section, experiments conducted to evaluate the stability, reliability and accuracy of the proposed methods are described and comparative results are presented. 2. SYSTEM OVERVIEW AND REPEATABILITY CHECK The experiments presented in this paper were all conducted using a Cyrax 2500 laser scanner. The instrument has a field of view of 40o by 40o , and operates with a green laser beam of 532nm. The spot size is less than 6mm for distances up to 50m, distances are measured with an accuracy of ± 4mm and the angles are measured with an accuracy of ± 60micro-radians. The accuracy in the position of single points is, according to the manufacturer, approximately ± 6mm for distances that range between 1.5m – 50m. The scan rate is very high, namely 1000 pts/second. The system is operated using a laptop and the processing of the data can be carried out using the Cyclone software suite (www.cyrax.com). Measurement repeatability is a very important property for a laser scanner system. In order to evaluate this property for the Cyrax 2500 system, two experiments were conducted. The former involved scanning four targets mounted on four pillars of the internal EDM calibration baseline of NTUA. The latter involved the scanning of five targets placed on a wall. For both cases, nine scans were collected for each one of the targets. The collected data were exported into an ASCII format which contains the cartesian coordinates in the scanner’s system along with the signal strength (reflectivity) for each point in the scan. The selection of the target image from each point cloud was performed through the proprietary Cyclone software. In each scan, the mean X, Y and Z values were calculated for each of the targets. Also, in order to evaluate the repeatability of the reflectivity, the mean value and standard deviation were calculated. Another part of the process was the calculation of the radiometric centre of each target i.e. the weighted mean X, Y and Z values, using the reflectivity as a weight. Using the derived mean values, the standard deviation was calculated for each one of the targets. Furthermore, in order to see how the use of reflectivity values implemented in the calculations affects the results, the mean absolute difference of the mean and the weighted mean values were calculated in each case. These calculations, though fairly simple, provide an efficient way to evaluate the repeatability. Table 1 shows the results from the target data collected at the baseline. In Table 2 the results for the case of the targets on the wall are given. The small standard deviation in both cases indicates that the repeatability of the scans is very high. Regarding the mean absolute differences, in the first case they appear to be rather small. This can be attributed to the fact that the acquired point clouds for each one of the targets were trimmed before any computations, so that the remaining points would describe only the target. However, this was not the case for the targets on the wall. The whole area that was scanned for each one of the targets was exported. This resulted in differences of a few millimetres, especially along the X and Y directions. The above results indicate that the repeatability of the measurements is very high and that the reflectivity should definitely be used in order to identify the centre of the target. 3. ALGORITHM PRESENTATION When Cyrax retroreflective targets are available, it is possible to define the position of their centres using the proprietary software. However, this is possible only during the data collection stage because of the way that this process is implemented. Specifically, the scanner acquires the data needed for defining the centre of the target after the user has selected a point near the actual centre of the target using the viewer of the software. The scanner then performs a dense scanning around the depicted position. A grid of 38x38 points is created and the centre of the target is defined using these data. The density of the scan data at this stage is found to be of approximately 1mm. However, the way that the centre of the target is defined remains unknown. Although not very well documented, the topic of automatic target identification has been previously addressed in the literature (Gordon et al., 2001; Lichti et al., 2000)]. In Lichti et al. (2000) three different methods are described. The first defines the centre of the target as the position with the maximum radiance. The second defines the centre by the mean position of the radiometric centre of the 4 strongest returns. The third algorithm defines the centre of the target as the radiometric centre of all returns. These methods will be referred to henceforth as ‘maxrad’, ‘maxrad4’ and ‘radcent’, respectively. In the following experiments, these methods will be applied and used for comparison purposes with the new developed methods. All the aforementioned methods have significant flaws that are not mentioned in the literature. The methods ‘maxrad’ and ‘maxrad4’ often fail because the position with maximum signal strength does not always correspond to the actual centre of the target. This is clearly shown in Figure 1 which shows part of a target with three different markers indicating the position of the centre as calculated using each of the three aforementioned algorithms. The red points correspond to points of the target with a relatively large value of reflectance. They also show the topographic artefacts that are observed for the highly reflective areas of a target. In Figure 1a, a front view of the target and the calculated centres is given, whereas in Figure 1b, the same target is presented from a different angle for visualisation purposes. In both figures, the ‘maxrad’, ‘maxrad4’ and ‘radcent’ positions of the centre are indicated in black, green and blue respectively. It is obvious that the ‘radcent’ algorithm has the best performance in this case. This was also confirmed by several experiments that were conducted and will be presented in the following section.