Surface Profile Based on Sensor Fusion Cristina Santos, Jaime Fonseca, Paulo Garrido, Carlos Couto Department of Industrial Electronics School of Engineering - University of Minho Braga-Portugal This paper reports a sensor system which it has been designed and constructed to acquire the profile of surfaces. This system is based in a CCD camera for object boundary-determination mounted over a robot manipulator shoulder and ultrasonic sensors for depth measurement mounted on a fixture at the wrist of the manipulator. It has been used an average weighted by degrees of confidence for raw sensor data fusion, based on a heuristic set of rules. 1. INTRODUCTION From an economical point of view may be interesting to replace a single highly accurate but expensive sensor by several less precise low-cost sensors requiring with additional post-processing electronics. Using several low- cost sensors combined with intelligent post-processing can compensate the its low accuracy. These sensors can be either of the same type or give complement information. With the same type of sensors the goal is to increase the quality of the resulting sensor information. Of course, the improvement must be reasonable compared with the increasing complexity of the measurement system in order to keep the overall cost still attractive. As the computing power cost is decreasing everyday and the low cost sensors is bound to proliferate in the near future, multisensor systems and sensor fusion techniques bound to became more and more popular. Several sensor fusion methods have been reported that deal with this problem. Durrant-Whyte has developed a Bayesian estimation technique for combining touch and stereo sensing[8]. Tang and Lee proposed a generic framework that employs a sensor-independent, feature-based relational model to represent information acquired by various sensors[19]. In[9], a Kalman filter update equation was developed to obtain the correspondence of a line segment to a model, and this correspondence was then used to correct position estimation. In[18], a extended Kalman filter was used to manipulate image and spatial uncertainties. The work described in this paper is concerned with the development of a robot based system to acquire the profile of a surface. A PUMA 560 manipulator was equipped with a CCD video camera and one ultrasonic sensor in the shoulder and a set of four ultrasonic sensors in the wrist, to acquire data for internally representing the geometry of the part’s surface, exploiting the mobility of the robot. The camera defines the work area while the sensors enable to get the surface profile. We used an average weighted by degrees of confidence for raw sensor data fusion to obtain more reliable representation in profile surface acquisition. This paper is organised as follows: Section 2 describes our hardware configuration, Section 3 describes the software procedures which implement image acquisition and depth map filling of the part to be acquired, Section 4 presents tests and results and, Section 5 presents the conclusions and future work suggestions. 2. HARDWARE CONFIGURATION 2.1. Sensor System Design The sensor system consists of a National Electronics monochrome CCD video camera and five ultrasonic sensors. The CCD camera and one ultrasonic sensor are mounted over the shoulder of the Puma, both horizontally aligned, to enabling a circular scanning around the robot stand. The other four ultrasonic sensors are mounted on a fixture at the wrist of the manipulator, hence benefiting from all degrees of freedom of the robot.