Embedded Vision Sensor Research Articles

Inverse Kinematics and Model Calibration Optimization of a Five-D.O.F. Robot for Repairing the Surface Profiles of Hydraulic Turbine Blades

This paper presents and discusses the results of an ongoing R&D project aiming to design and build a fully automated prototype of a specialized spherical robotic welding system for repairing hydraulic turbine surfaces eroded by cavitation pitting and/or cracks produced by cyclic loading. The system has an embedded vision sensor built to acquire range images by laser scanning over the blade's surface and produce 3D models to locate the damaged spots to be registered in a 3D coordinate system into the robot controller, enabling the robot to repair the flaws automatically by welding in layers. The paper is focused on the robot kinematic model and describes an iterative algorithm to process the inverse kinematics with only five degrees-of-freedom. The algorithm makes use of data collected from a vision sensor to ensure that the welding gun axis is perpendicular to the blade's surface. Besides this, it proposes a modelling and optimization mathematical routine for more efficient robot calibration, which can be used with any type of robot. This robot calibration optimization scheme finds the optimal error parameter vector based on the condition number of the manipulator transformation composed from the partial derivatives of the error parameters. Experimental results proved both the iterative algorithm to perform the inverse kinematics and the technique to optimize robot calibration to be very efficient.

Novel Feature Descriptor for Low-Resource Embedded Vision Sensors for Micro Unmanned-Aerial-Vehicle Applications

Feature point matching is an important step for many vision-based unmanned-aerial-vehicle applications. This paper presents the development of a new feature descriptor for feature point matching that is well suited for micro unmanned aerial vehicles equipped with a low-resource, compact, lightweight, low-power embedded vision sensor. The Basis Sparse-Coding Inspired Similarity descriptor uses theory taken from sparse coding to provide an efficient image feature description method for frame-to-frame feature point matching. This descriptor requires simple mathematical operations, uses comparatively small memory storage, and can support color and grayscale feature descriptions. It is an excellent candidate for implementation on low-resource systems that require real-time performance, where complex mathematical operations are prohibitively expensive. To demonstrate its performance, the feature matching result was used to calculate a frame-to-frame homography that is essential to unmanned-aerial-vehicle applications such as pose estimation and obstacle detection for navigation. The proposed descriptor was tested on two video sequences and one dataset of real aerial images. Experimental results show that, along with performing in situations where existing complex descriptors cannot be used, the Basis Sparse-Coding Inspired Similarity descriptor also performs slightly better than these other methods on the task of homography calculation. Our experimental results and analysis show that the Basis Sparse-Coding Inspired Similarity descriptor is an excellent candidate for a resource-limited vision sensor for micro unmanned aerial vehicles.

AUTONOMOUS 3D OBJECT MODELING BY A HUMANOID USING AN OPTIMIZATION-DRIVEN NEXT-BEST-VIEW FORMULATION

An original method to build a visual model for unknown objects by a humanoid robot is proposed. The algorithm ensures successful autonomous realization of this goal by addressing the problem as an active coupling between computer vision and whole-body posture generation. The visual model is built through the repeated execution of two processes. The first one considers the current knowledge about the visual aspects and the shape of the object to deduce a preferred viewpoint with the aim of reducing the uncertainty of the shape and appearance of the object. This is done while considering the constraints related to the embodiment of the vision sensors in the humanoid head. The second process generates a whole robot posture using the desired head pose while solving additional constraints such as collision avoidance and joint limitations. The main contribution of our approach relies on the use of different optimization algorithms to find an optimal viewpoint by including the humanoid specificities in terms of constraints, an embedded vision sensor, and redundant motion capabilities. This approach differs significantly from those of traditional works addressing the problem of autonomously building an object model.