Zhurnal Radioelektroniki - Journal of Radio Electronics. eISSN 1684-1719. 2021. №11
Contents

Full text in Russian (pdf)

Russian page

 

DOI: https://doi.org/10.30898/1684-1719.2021.11.15

UDC: 681.3, 004.8

 

APPLICATION OF CONVOLUTIONAL DEEP NEURAL NETWORKS FOR SOLVING SOME TRAJECTOR DATA ANALYSIS PROBLEMS

 

M. I. Kostiuchek, A. V. Makarenko

 

V. A. Trapeznikov Institute of Control Sciences of Russian Academy of Sciences, 117997, Moscow, 65 Profsoyuznaya street

 

The paper was received October 30, 2021

 

Abstract. The application of convolutional deep neural networks for solving the problem of classification and untangling of trajectories is considered. Objects are classified by their cinematic characteristics. Res-Net type model with the cross-entropy loss function has been developed for the classification problem. We have achieved 0.883 classification quality according to the F_1-metric. The model classifies objects by speed better than by other cinematic characteristics. The model distinguishes objects from noise with high accuracy. The model for classification is tested for resistance to noise. Noise is added to the data in the form of random points, or some points of the trajectory are removed. The model is trained on data with ten noise points and tested on data with a large number of noise points. Classification quality decreases "smoothly" with an increase in the number of noise points, without sharp jumps. Object trajectories for classification can have break points. The trajectories of two objects are untangled. For the untangling problem, a UNet-type model has been developed. Loss function in the form of a combination of cross-entropy and Tversky loss has been used. Also, the developed model returns an estimate of the speed of an object based on hidden features. It was used to account for the dynamics. We have achieved 0.911 classification quality according to the F_1-metric for the untangling problem. The models are trained and tested on a synthetic dataset.

Key words: object classification, trajectory untangling, deep learning, synthetic dataset, convolutional neural network.

References

1. Kazakov I.E., Mal'chikov S.V. Analiz stokhasticheskikh sistem v prostranstve sostoyanii [Analysis of stochastic systems in state space]. Moscow, Nauka Publ. 1983. 384 p. (In Russian)

2. Yarlykov M.S., Mironov M.A. Markovskaya teoriya otsenivaniya sluchainykh protsessov [Markov theory of estimation of stochastic processes]. Moscow, Radio i svyaz' Publ. 1993. 464 p. (In Russian)

3. Markovskaya teoriya otsenivaniya v radiotekhnike, pod red. M.S. Yarlykova [Markov estimation theory in radio engineering, edited by M.S. Yarlykov]. Moscow, Radiotekhnika Publ. 2004. 504 p. (In Russian)

4. Tatuzov A.L. Neironnye seti v zadachakh radiolokatsii [Neural networks in Radiolocation]. Moscow, Radiotekhnika Publ. 2009. 432 p. (In Russian)

5. Goodfellow I., Bengio Y., Courville A. Deep Learning. Cambridge, Massachusetts, USA, MIT Press. 2016. 800 p.

6. Portsev R.J., Makarenko A.V. Convolutional Neural Network for Noise Signal Recognition. 2018 IEEE 28th International Workshop on Machine Learning for Signal Processing (MLSP). Aalborg, Denmark. 2018. P.1-6. https://doi.org/10.1109/MLSP.2018.8516920

7. Makarenko A.V. Deep Convolutional Neural Networks for Chaos Identification in Signal Processing. 2018 26th European Signal Processing Conference (EUSIPCO). Rome, Italy. 2018. P.1481-1485. https://doi.org/10.23919/EUSIPCO.2018.8553098

8. Bobin A.V., Azarov V.A., Bulgakov S.A., Savin D.A. Metodika raspoznavaniya letatel'nykh apparatov i radiolokatsionnykh lovushek v konture upravleniya sistemy kontrolya vozdushnogo prostranstva na osnove neirosetevoi tekhnologii [A technique for recognizing aircraft and radar traps in the control loop of an airspace control system based on neural network technology]. Izvestiya MGTU «MAMI» [Izvestiya MGTU «MAMI»]. 2013. T.4 №1. С.123-130. (In Russian)

9. Park S.H., Kim B., Kang C.M., Chung C.C., Choi J.W. Sequence-to-Sequence Prediction of Vehicle Trajectory via LSTM Encoder-Decoder Architecture, 2018 IEEE Intelligent Vehicles Symposium. Changshu, China. 2018. P.1672-1678. https://doi.org/10.1109/IVS.2018.8500658

10. Khosroshahi A., Ohn-Bar E., Trivedi M.M. Surround Vehicles Trajectory Analysis with Recurrent Neural Networks. 2016 IEEE 19th International Conference on Intelligent Transportation Systems (ITSC). Rio de Janeiro, Brazil. 2016. P.2267-2272. https://doi.org/ 10.1109/ITSC.2016.7795922

11. Wang L., Zhang L, Yi Z. Trajectory Predictor by Using Recurrent Neural Networks in Visual Tracking. IEEE Transactions on Cybernetics. 2017. Vol.47. P.3172-3183. https://doi.org/10.1109/TCYB.2017.2705345

12. Liu Y., Hansen M. Predicting Aircraft Trajectories: A Deep Generative Convolutional Recurrent Neural Networks Approach. CoRR. 2018. https://arxiv.org/abs/1812.11670

13. Kostiuchek M.I., Makarenko A.V. Classification of observable 3D moving object by their kinematic characteristics by deep convolution neural network. Proceedings of the 5th International Conference on Stochastic Methods (ICSM-5, 2020). Moscow. 2020. P.326-330. (In Russian)

14. Kostiuchek M.I., Makarenko A.V. A Technique for Generation of Synthetic 3D Trajectories of Moving Objects in Kinematic Approximation. Materialy 15-i Mezhdunarodnoi konferentsii «Ustoichivost' i kolebaniya nelineinykh sistem upravleniYA» (konferentsiya Pyatnitskogo) (Moskva, 2020) [Proceedings of the 15th International Conference "Stability and Oscillations of Nonlinear Control Systems" (Pyatnitsky Conference) (Moscow, 2020)]. Moscow. 2020. P.219-221. (In Russian)

15. He K., Zhang X., Ren S., Sun J., Deep residual learning for image recognition. CoRR. 2015. https://arxiv.org/abs/1512.03385

16. Lin M., Chen Q. Yan S. Network in network. 2nd International Conference on Learning Representations (ICLR) 2014. Banff, AB, Canada. 2014. https://arxiv.org/abs/1312.4400

17. Ioffe S., Szegedy C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. Proceedings of the 32nd International Conference on Machine Learning, PMLR. Lille, France. 2015. Vol.37. P.448-456.

18. Hinton G.E., Srivastava N., Krizhevsky A., Sutskever I., Salakhutdinov R., Improving neural networks by preventing co-adaptation of feature detectors. CoRR. 2012. http://arxiv.org/abs/1207.0580

19. Polyak, B. T. Some methods of speeding up the convergence of iteration methods. USSR Computational Mathematics and Mathematical Physics. 1964. Vol.4. P.1–17. https://doi.org/10.1016/0041-5553(64)90137-5

20. Ronneberger O., Fischer P. Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. CoRR. 2015. http://arxiv.org/abs/1505.04597

21. Salehi S.S.M., Erdogmus D. Tversky loss function for image segmentation using 3D fully convolutional deep networks. CoRR. 2017. http://arxiv.org/abs/1706.05721

For citation:

Kostiuchek M.I., Makarenko A.V. Application of Convolutional Deep Neural Networks for Solving Some Trajectory Data Analysis Problems. Zhurnal radioelektroniki [Journal of Radio Electronics] [online]. 2021. №11. https://doi.org/10.30898/1684-1719.2021.11.15 (In Russian)