Salient object detection is a very active area of interest, based on human selective attention which shifts our focus across prominent areas in a scene. Many models have been created and achieved impressive accuracies, however, this often comes at a high computational efficiency cost making them less attractive for real-time robotic systems. We develop a novel salient object detection algorithm and investigate the trade-off between accuracy and computation time by comparing pixel-based and region-based processing. Salient object detection has been used in a variety of applications, however, this work is focused on developing an algorithm considering obstacle avoidance for a robot operating in a cluttered and dynamic environment. The strategy is evaluated on the publicly available MSRA10K salient object dataset, against three current state-of-the-art methods.
|Number of pages||64|
|Publication status||Published - 29 Jul 2018|
|Event||Irish Machine Vision and Image Processing Conference - Belfast, United Kingdom|
Duration: 29 Aug 2018 → 31 Aug 2018
|Conference||Irish Machine Vision and Image Processing Conference|
|Period||29/08/18 → 31/08/18|
- Obstacle Avoidance