Keywords
low complexity tracking; underwater computer vision; data compression
Location
Session C1: VI Data Mining for Environmental Sciences Session
Start Date
12-7-2016 10:50 AM
End Date
12-7-2016 11:10 AM
Abstract
A biological experiment that requires a continuous tracking of a group of 10 to 20 fish specimens was designed to obtain empirical data to estimate biological energy models from specimen’s activity. The surface in which specimens can move is enormous (2,5m x 5m) in comparison to the sizes of the individuals and it is mandatory to work with high resolution images in order to cover all the area. Furthermore, as each experiment must work uninterruptedly during fifteen days the cost of storing and post-processing all the generated information in its rough form is prohibitive, incurring into a big data problem. The proposed approach deals with the need of implementing a real time processing and in this paper we present the computer vision algorithms used to locate and identify specimens and track them along the time series of images. Additionally, in order to verify all experiments, we have designed a compression strategy to save the relevant information of each frame that makes possible to store it in a sustainable way, but with the property that could be partially reversible, enabling to display back a reconstructed reproduction of the fishes’ movements.
Included in
Civil Engineering Commons, Data Storage Systems Commons, Environmental Engineering Commons, Hydraulic Engineering Commons, Other Civil and Environmental Engineering Commons
Location, identification, and low complexity tracking algorithms with an image compression strategy for a long term multi-tracking application
Session C1: VI Data Mining for Environmental Sciences Session
A biological experiment that requires a continuous tracking of a group of 10 to 20 fish specimens was designed to obtain empirical data to estimate biological energy models from specimen’s activity. The surface in which specimens can move is enormous (2,5m x 5m) in comparison to the sizes of the individuals and it is mandatory to work with high resolution images in order to cover all the area. Furthermore, as each experiment must work uninterruptedly during fifteen days the cost of storing and post-processing all the generated information in its rough form is prohibitive, incurring into a big data problem. The proposed approach deals with the need of implementing a real time processing and in this paper we present the computer vision algorithms used to locate and identify specimens and track them along the time series of images. Additionally, in order to verify all experiments, we have designed a compression strategy to save the relevant information of each frame that makes possible to store it in a sustainable way, but with the property that could be partially reversible, enabling to display back a reconstructed reproduction of the fishes’ movements.
