Development of an algorithm for detecting moving objects based on deep learning in real-time use on a quadcopter flying robot

Quadcopter flying robots are powerful and technological devices for flight operations, and many studies have been carried out to control and make them intelligent. But most of these studies have focused on their motion control and less attention has been paid to their application for moving object detection in a dynamic environment such as the street. In this article, we develop a hybrid method based on Viola-Jones algorithm and deep neural networks so that can be intelligently used by quadcopter to identify moving objects in the real environment. In this method, deep neural network training is done using Viola-Jones classifier cascades, by adding objects as video sequences, with a real-world background. By experimentally implementing this Viola-Jones cascade optimized algorithm on a quadcopter robot, we show that in detecting frames containing moving objects received from the camera installed on the robot, it has high performance rate, low false alarm (negative-error, positive-error) and low error rate. Algorithm performance in this test was 89% positive correct diagnosis and 13% error in diagnosis. Also, the accuracy increases when the Gaussian blur filter is used.