Deep Learning requires huge amount of data with related labels, that are necessary for proper training. Thanks to modern videogames, which aim at photorealism, it is possible to easily obtain synthetic dataset by extracting information directly from the game engine. The intent is to use data extracted from a videogame to obtain a representation of various scenarios and train a deep neural network to infer the information required for a specific task. In this work we focus on computer vision aids for automotive applications and we target to estimate the distance and speed of the surrounding vehicles by using a single dashboard camera. We propose two network models for distance and speed estimation, respectively. We show that training them by using synthetic images generated by a game engine is a viable solution that turns out to be very effective in real settings.
Estimation of Speed and Distance of Surrounding Vehicles from a Single Camera / Zaffaroni, Mirko; Grangetto, Marco; Farasin, Alessandro. - ELETTRONICO. - LCNS 11751:(2019), pp. 388-398. (Intervento presentato al convegno ICIAP 2019 - International Conference on Image Analysis and Processing tenutosi a Trento (IT) nel 9 - 13 september 2019) [10.1007/978-3-030-30642-7_35].
Estimation of Speed and Distance of Surrounding Vehicles from a Single Camera
Marco Grangetto;Alessandro Farasin
2019
Abstract
Deep Learning requires huge amount of data with related labels, that are necessary for proper training. Thanks to modern videogames, which aim at photorealism, it is possible to easily obtain synthetic dataset by extracting information directly from the game engine. The intent is to use data extracted from a videogame to obtain a representation of various scenarios and train a deep neural network to infer the information required for a specific task. In this work we focus on computer vision aids for automotive applications and we target to estimate the distance and speed of the surrounding vehicles by using a single dashboard camera. We propose two network models for distance and speed estimation, respectively. We show that training them by using synthetic images generated by a game engine is a viable solution that turns out to be very effective in real settings.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/2750857
Attenzione
Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo