Authors:
A. I. Glushkov
1
;
V. G. Shepelev
1
;
S. D. Shepelev
2
;
K. A. Magdin
3
;
I. Slobodin
1
;
A. I. Burzev
4
and
V. D. Mavrin
3
Affiliations:
1
South Ural State University, 76 Lenin prospekt, Chelyabinsk, Russia
;
2
South Ural State Agrarian University, 13 st. Gagarina, Troitsk, Russia
;
3
Kazan Federal University, 18 Kremlyovskaya str, Kazan, Russia
;
4
T. F. Gorbachev Kuzbass State Technical University, 32A Il'icha st., Kemerovo, Russia
Keyword(s):
Carbon Oxide CO, Environmental Monitoring, Neural Networks, Polluting Emissions, Statistical Analysis of Differences, Sulphur Dioxide SO2.
Abstract:
This paper studies the use of machine vision in the environmental monitoring of harmful traffic flow emissions. The purpose of the study is to develop a methodology for the high-quality and complete collection of data on the atmospheric emissions of harmful substances from traffic flows. The data is collected within the entire functional area of intersections and adjacent road sections falling within the video surveillance camera angle. Our solution is based on the use of the YOLOv3 (You Only Look Once) convolutional neural network architecture and SORT (Simple Online and Real-time Tracking) tracker. The system is based on the real-time collection and interpretation of the data obtained from street video surveillance cameras using convolutional neural networks. In this study, we focused on collecting the data on two substances: carbon oxide CO and sulphur dioxide SO2. We chose these substances taking into account their stable properties, which allow them not to react with other subst
ances. To assess the quality of the obtained data on harmful emissions, we verified their identity based on laboratory measurements of the Environmental Monitoring Centre Public Institution. An analysis of the data sample confirmed the absence of statistically significant differences in the calculations of the emissions using neural networks versus the laboratory measurements.
(More)