| Resumo: | One of the main problems in the development and debugging of robotic systems is the amount of data stored in files containing sensor data (ex. ROS proprietary log files - BAGS). If we consider a robot with several cameras and other sensors that collect information from the environment several times per second, we quickly obtain very large files. Besides the concerns regarding storage and, in some cases, transmission, it becomes extremely hard to find important information in these files. In this thesis, we tried to solve both problems studying and implementing data compression solutions to reduce the referred files. The main focus was image and video compression, by far the most storage consuming data. Moreover, we conducted a detailed study about the effect of lossy compression methods in the performance of some state of the art image analysis algorithms. Another contribution was the development of an intelligent video player to help roboticists in their work while they evaluate the recorded data after experiments. Parts of the video that do not contain relevant information are skipped during the play. Based on the results, we concluded that ROS native compression is not sufficient. Furthermore, solutions based on ROS, or virtually any robotic system that has to deal with image/video data, would benefit with the use of a H.265 codec, as it provides the smallest number of bits per pixel without a significant penalty on the performance of image analysis algorithms.
|
|---|