The automatic analysis of images acquired by cameras mounted on board of drones (flying cameras) is attracting many scientists working in the field of computer vision; the interest is related to the increasing need of algorithms able to understand the scenes acquired by flying cameras, by detecting the moving objects, calculating their trajectories, and finally understanding their activities. The problem is made challenging by the fact that, in the most general case, the drone flies without any awareness of the environment; thus, no initial set-up configuration based on the appearance of the area of interest can be used for simplifying the task, as it generally happens when working with fixed cameras. Moreover, the apparent movements of the objects in the images are superimposed to that generated by the camera, associated with the flight of the drone (varying in the altitude, speed, and the angles of yaw and pitch). Finally, it has to be considered that the algorithm should involve simple visual computational models as the drone can only host embedded computers having limited computing resources. This paper proposes a detection and tracking algorithm based on a novel paradigm suitably combining a forward tracking based on local data association with a backward chain, aimed at automatically tuning the operating parameters frame by frame, so as to be totally independent on the visual appearance of the flying area. This also definitively drops any timeconsuming manual configuration procedure by a human operator. Although the method is self-configured and requires low-computational resources, its accuracy on a wide data set of real videos demonstrates its applicability in real contexts, even running over embedded platforms. Experimental results are given on a set of 53 videos and more than 60 000 frames.

Multi-object tracking by flying cameras based on a Forward-Backward Interaction

Carletti, Vincenzo;Greco, Antonio;Saggese, Alessia
;
Vento, Mario
2018

Abstract

The automatic analysis of images acquired by cameras mounted on board of drones (flying cameras) is attracting many scientists working in the field of computer vision; the interest is related to the increasing need of algorithms able to understand the scenes acquired by flying cameras, by detecting the moving objects, calculating their trajectories, and finally understanding their activities. The problem is made challenging by the fact that, in the most general case, the drone flies without any awareness of the environment; thus, no initial set-up configuration based on the appearance of the area of interest can be used for simplifying the task, as it generally happens when working with fixed cameras. Moreover, the apparent movements of the objects in the images are superimposed to that generated by the camera, associated with the flight of the drone (varying in the altitude, speed, and the angles of yaw and pitch). Finally, it has to be considered that the algorithm should involve simple visual computational models as the drone can only host embedded computers having limited computing resources. This paper proposes a detection and tracking algorithm based on a novel paradigm suitably combining a forward tracking based on local data association with a backward chain, aimed at automatically tuning the operating parameters frame by frame, so as to be totally independent on the visual appearance of the flying area. This also definitively drops any timeconsuming manual configuration procedure by a human operator. Although the method is self-configured and requires low-computational resources, its accuracy on a wide data set of real videos demonstrates its applicability in real contexts, even running over embedded platforms. Experimental results are given on a set of 53 videos and more than 60 000 frames.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11386/4714882
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 13
  • ???jsp.display-item.citation.isi??? 14
social impact