Vision-based absolute position sensing on technical surfaces

Abstract

Measuring the position on a technical surface is a fundamental task in many industries. The manifold of technical surfaces, each designed to fulfill a specific requirement, impose several constraints for a potential sensing approach. In many cases, position measurement should be contactless and without modification of the measurement target, even if the roughness of the surface is low. Spatially extended surfaces or retrofitted machinery parts often impede measurements against the end face and demand for robustness against temporal pollution of the measurement object. The supreme discipline is to prevent time-consuming homing procedures, and as such to offer an absolute position measurement. To overcome all these challenges, this thesis investigates the use of image based measurement systems in order to measure the absolute in-plane position of technical surfaces without touching or modification of the surface itself. This can be either by direct observation or, in case of smooth surfaces, through the indirect observation of so-called laser speckle patterns. The first system presented is a compensation based displacement measurement technique using objective laser speckle patterns. These patterns are generated by lens-less observation of a small laser-spot and are prone to change even under small displacements of the surface (decorrelation). Through active compensation of the estimated speckle pattern displacement, the limiting decorrelation is prevented which allows operation independently of the diameter of the laser spot. In a prototypic implementation, an absolute position error of 12.6 μm over a measurement range of 2mm is reached. The lens-less design and a small laser spot diameter, here 900 μm, allow for a high degree of miniaturization and a small area of interaction. In a second system, the laser speckle patterns are used to generate an image based absolute scale. Based on the idea of a simple look-up table with potentially large memory consumption, a feature-of-interest based system is proposed and implemented. For position measurement, a probabilistic position estimator is proposed which uses a global map, that is generated under consideration of the specific properties of laser speckle patterns. The performance of the system is evaluated for incoherent and coherent illumination in terms of statistical independence, position accuracy, and reliability of state recovery behavior. The laboratory system is able to estimate the position of a cylindrical rod with a peak to peak error of 39.6 μm and an RMS error of 5.6 μm over a measurement range of 100mm. Thereby, the memory consumption is only 207 kB for incoherent and 628 kB for coherent illumination. The final presented system enhances the feature-based approach by thinking of each feature as an individual measurement. To deal with slowly varying surfaces, new features are embedded within an initially generated global map, similar to other simultaneous localization and mapping schemes. Three different illumination scenarios – white light illumination, observation of objective laser speckles, and observation of subjective laser speckles – are compared within 1000 identical scan cycles of a cylindrical rod over a measurement distance of 200mm. For relative position estimation, the two systems based on laser speckle patterns reach a maximum error of 77 μm with a standard deviation of 11.5 μm, theoretically independent of the measurement distance. The proposed solution increases the likelihood for a successful recovery of the absolute position estimate, such as after sudden power-loss, for all evaluated illumination scenarios. For white light images and objective laser speckle patterns the success rates are between 50% to 60% and with subjective laser speckle patterns 99.8% are reached. It is thus possible to determine the absolute position contact-less and without modification of the surface on technical, optically rough objects with high probability.