Regresar

Robust System for Crowd Counting and Estimation Using Multi-column Filters on Unmanned Platforms

Abstract:

The automatic estimation combined with the use of sensors is increasingly useful for various applications that improve the quality of life of human beings and their environment. These developments become more relevant if they are used in complex scenarios, such as those generated from images and videos obtained from aerial images, from optical sensors installed in unmanned systems. In this research, an algorithm with a Multicolumn Gaussian Filter was implemented, which, from images and videos obtained in real-time, with flights at altitudes above 100 m, can estimate the number of people that make up conglomerate crowds in certain sectors. A particularity of the contribution of this research is the ability to perform estimation in real-time, with a consequent multicast transmission, whose low-cost architecture allows the information to be transmitted to geographically distant locations with minimal latency. The results obtained in the research, generate high reliability, to estimation, detection, and counting of people in complex scenarios. The calculation of the error in the training phase shows 6.37% MAE error in estimation accuracy and 15.33% MSE error, these measures verify robustness in the estimate. Also, the solution implemented shows an accuracy rate of 92.39% in the classification, in addition, the field tests using UAVs show high performance in real application situations. Finally, a compensation factor has been introduced, which has made it possible to avoid dispersion in the estimation of the automatic crowd count in real employment situations.