Received 22 February 2018
The paper reviewes the structure and operational principles of the SSD (Single Shot MultiBox Detector) deep convolutional neural network and describes the results of experimental research on the possibility of using this network for people and car detection by computer vision system of the mobile robot. The research showed that the SSD network can be used for this task, but it does not have very high accuracy and speed. The accuracy of the network was improved by modifying (histogram equalization) of input images and network learning. The research concluded that it is necessary to build own database for learning and to refine the network architecture for further efficiency and effectiveness improvement of the method.
Neural network, convolutional neural network, deep learning, object detection, computer vision system.
Orlova, S. and Komarov, A. (2018). Exploring of SSD deep convolutional neural network for people and car detection by computer vision system of mobile robot. Robotics and Technical Cybernetics, 2(19), pp.21-25.
- Fomin, I., Gromoshinsky, D. and Stepanov, D. (2016). Visual features detection based on a deep learning network in autonomous driving problems. In: Proceedings of the 26th International Conference GraphiCon2016. pp. 430-434. (in Russian).
- Shaoqing Ren, Kaiming He, Ross Girshick and Jian Sun (2017). Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), pp. 1137-1149.
- Joseph Redmon, Santosh Divvala, Ross Girshick and Ali Farhadi (2016). You Only Look Once: Unified, Real-Time Object Detection. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
- Wei Liu and et al. (2016). SSD: Single Shot MultiBox Detector. In: The 14th European Conference on Computer Vision (ECCV).
- Fomin I., Gromoshinskii D. and Bakhshiev A. (2018). Object Detection on Images in Docking Tasks Using Deep Neural Networks. In: Kryzhanovsky B., Dunin-Barkowski W., Redko V., ed., Advances in Neural Computation, Machine Learning, and Cognitive Research. Springer, pp 79-84. https://doi.org/10.1007/978-3-319-66604-4_12.
- Fomin, I., Bakhshiev, A. and Gromoshinskii, D. (2017). Study of Using Deep Learning Nets for Mark Detection in Space Docking Control Images. Procedia Computer Science, 103C, pp. 59-66.
- Viola, P., Jones, M. (2001). Rapid object detection using a boosted cascade of simple fea-tures. In: Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR).
- Simonyan, K. and Zisserman, A. (2017). Very Deep Convolutional Networks for Large-Scale Visual Recognition. [online] Visual Geometry Group, Department of Engineering Science, University of Oxford. Available at: http://www.robots.ox.ac.uk/~vgg/research/very_deep [Accessed 12 Sept. 2017].
- GitHub. (2017). A port of SSD: Single Shot MultiBox Detector to Keras framework. [online] Available at: https://github.com/rykov8/ssd_keras [Accessed 12 Sept. 2017].
- TensorFlow. (n.d.). TensorFlow: An open-source software library for Machine Intelligence. [online] Available at: https://www.tensorflow.org [Accessed 12 Sept. 2017].
- Keras documentation. (n.d.). Keras: The Python Deep Learning library. [online] Available at: https://keras.io [Accessed 12 Sept. 2017].
- The PASCAL Visual Object Classes Homepage. (n.d.). PASCAL VOC Database viewed. [online] Available at: http://host.robots.ox.ac.uk/pascal/VOC [Accessed 12 Sept. 2017].
- OpenCV-Python Tutorials. (n.d.) Histograms - 2: Histogram Equalization. [online] Available at: http://docs.opencv.org/3.1.0/d5/daf/tutorial_py_histogram_equalization.html [Accessed 12 Sept. 2017]