Recognition of Dynamic Targets using a Deep Convolutional Neural Network

Recognition and tracking of dynamic objects play an important role in the development of service robots behaviors allowing them to co-exist with humans and other autonomous machines in shared environments. They can simplify the design of autonomous navigation and obstacle avoidance algorithms as well as the ability to operate within multi-agent formations. In this investigation an approach is proposed to use a deep convolutional neural network for recognition and tracking of pre-specified dynamic objects on a sequence of images. It is regarded as a substantial part on the way of achieving our goal to design a dynamic target following behavior for a service robot, based on data received from its onboard camera. During the conducted experiments the implemented deep learning neural network is able to recognize and localize on a sequence of images a moving in the lab mobile robot iRobot Create. The developed algorithm localizes the recognized object and begins considering it as a potential dynamic target.