This paper deals with the quality of digitised points obtained with noncontact probes. The digitising system is analysed so that each source of inaccuracy can be isolated. In particular, for systems such as triangulation laser sensors, the use of the CCD camera is not influence free, and generates nonhomogeneous errors. All sources of inaccuracy of the digitising system lead to a point cloud, the quality of which is described through indicators. These indicators correspond to the digitising noise and the density of the data. In addition to those usual indicators, we suggest qualifying the point cloud through completeness and accuracy. The completeness identifies the dimension of the digitising gaps, while the accuracy is associated with the measurement uncertainty of a 3D point. For each indicator, an evaluation method is presented and then applied. However, the use of those quality indicators only makes sense if they are related to the point exploitation.
Nowadays, 3D digitising systems deliver a numerical representation of part shapes through large sets of 3D data. Most works have concentrated on the topic of automated digitising of part shapes; few of them propose an evaluation of the digitising quality. Indeed, as digitising points are generally exploited by applications such as reverse engineering, free form copying or part inspection, the quality of the delivered point cloud, as regards its exploitation becomes an important issue. Therefore the question is: is the point cloud quality suitable for its exploitation? The quality is defined by the standard as a set of characteristics that confers to an entity the ability to satisfy implicit and expressed requirements [1]. This definition implies that quality cannot be dissociated from user’s requirements that are most often linked to point exploitation. Moreover, the expected quality levels are not equivalent for each point exploitation. However, there is no explicit definition available for characterising the quality of optical digitising data.
Most authors are interested in factors that influence the point cloud quality. A brief study of the digitising system brings out that the most significant causes of inaccuracy can obviously be linked to optical phenomena: speckle, use of a CCD camera, distortion,… [2], [3], [4] and [5]. Few of them propose to clearly identify digitising errors. Some authors suggest procedures to evaluate measurement accuracy of various digitising systems based on the measurement of specific shapes [6]. The digitising quality is evaluated in terms of accuracy, and special attention is given to how the object is recovered. Noncontact 3D devices can also be globally evaluated by the measurement of an artefact, including effects of both the scan depth and the view angle [7]. The influence of the scan planning can thus be highlighted.
However, integrating accuracy to the point exploitation is seldom carried out. Massios and Fisher integrate a quality criterion for selecting the best next view to scan an object, and show that the overall range data quality is thus improved [8]. Prieto suggests integrating digitising noise to define the most adapted scan planning for an application of part inspection. The digitising noise is previously correlated to scan depth and incidence angle [9]. In early works, Hoppe et al. show the difficulty to reconstructing surfaces from unorganised points. In particular, they suggest to qualify the data by indicators: δ—noisy and ρ—dense. The first indicator is more linked to the data sampling errors, and the second is directly linked to the sampling density. Both indicators are representative of the digitising point quality, and obviously influence the accuracy of surface reconstruction [10].
In this paper, we propose to evaluate the quality of digitising data in relation to the further point cloud exploitation. As suggested by Hoppe et al., the point quality is evaluated through quality indicators. However, considering that points can be exploited for surface reconstruction, but also for remanufacturing free form copying or inspection, specific indicators have to be defined. So, in addition to δ-noisy and ρ-dense, unorganised data points are also qualified through their completeness and accuracy. The next section is dedicated to the analysis of the various sources of digitising errors. Characteristics of the point cloud quality thus obtained are emphasised. Then, we suggest a definition for quality indicators and procedures in order to evaluate them. As they are defined, the user can link point cloud exploitation to its need.
3D digitising systems are now widely used for they deliver much more data than mechanical probes, in a shorter time. The sampled data are used in many applications such as inspection, reverse engineering, and free form copying. Evaluation procedures must be developed for such systems in order to guarantee the concordance between raw data and user’s intent. In this paper, we propose to qualify the point cloud obtained by quality indicators, in relation with the point exploitation. Indeed, applications using 3D digitised data are different, and the notion of quality of a point cloud cannot be dissociated from the application. Throughout the definition of several indicators, we try to bring keys to the user’s evaluation of point cloud quality. These indicators point out problems which arrive in digitising process such as: inhomogeneous and noisy data, noncomplete point cloud (with digitising gaps), and inaccurate. We define and implement methods to evaluate all these indicators. We illustrate our purpose with the example of a real scan of a part using a plane laser sensor, for which indicators are calculated and linked to point exploitations.
