Application of Deep Neural Network for the Vision System of Mobile Service Robot

The solution of object detection task is valuable in many fields of robotics. However, application of neural networks for mobile robots requires the use of high – performance architectures with low power consumption. In search of suitable model, a comparative analysis of the YOLO and SqueezeDet architectures was conducted. The task of detecting wooden cubes by mobile robot with the camera with the aim of collecting them was solved. A specific dataset was constructed for the training purposes. Applied SqueezeDet neural network has reached precision 89% and recall 82% for IOU ≥ 0.5.

[1]  Ronald Davis,et al.  Neural networks and deep learning , 2017 .

[2]  Forrest N. Iandola,et al.  SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).

[3]  Andreas Geiger,et al.  Vision meets robotics: The KITTI dataset , 2013, Int. J. Robotics Res..

[4]  Ali Farhadi,et al.  You Only Look Once: Unified, Real-Time Object Detection , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[5]  Xindong Wu,et al.  Object Detection With Deep Learning: A Review , 2018, IEEE Transactions on Neural Networks and Learning Systems.

[6]  Forrest N. Iandola,et al.  SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size , 2016, ArXiv.