Object Distance Estimation from a Binocular Vision System for Robotic Applications Using Artificial Neural Networks

The distance to objects in the robots workspace is important to bring near the end effector closer without colliding. This paper presents a distance estimation method from the binocular vision system to the objects in the robot workspace, with two images captured by the cameras and then are processed by the artificial neural networks. The proposed method was tested with a 4-layer ADALINE-type artificial neural network of 61 artificial neurons and 129,920 inputs that correspond to the pixels of the captured images. The vision system can work with low-cost cameras such as webcams because the optical parameters are not required to train the artificial neural network. Finally, test results on an experimental setup with webcams are presented and they have a suitable precision for robotic arm applications.