Object Orientation Detection Based on Machine Vision and Artificial Neural Network

This paper focuses on the problem of finding object orientation around Yaw & Pitch & Roll angels. The object orientation is computed in a real time manner using a mono-camera and three points on a solid object in a machine vision software. Three points should be selected from environment at the beginning. In order to reduce wreckful effects of environmental lights on detecting colorful objects and also to reduce the number of used software filters, IR LEDs with 850nm invisible wavelength are used. Artificial Neural Network (ANN) is used for solving this problem since orientation's equations are nonlinear and real-time solving for them is impossible. For solving the problem a feed forward artificial neural network with one hidden layer and 21 nodes in that is used, which has 3 nodes for output layer and 6 nodes for input layer. For having high accuracy in ANN, output data is also obtained from a MPU-9150 installed on a 2-DOF orientional parallel robot and compared to ANN outputs. 7243 data from Roll and Yaw angles and 751 data from Pitch angle is obtained from MPU-9150 sensor and the later 2-DOF orientional parallel robot and 467 data remains nonuse for learning ANN. After learning the neural network, results compared to nonuse data for ANN learning and desire results obtained with 0.038 maximum error.