Abstract
Video segmentation is the foundation of many senior computer vision applications. For robotic vision applications, video object segmentation is facing more difficulties. Since the videos are usually collected from those complicated scenes and the camera perspective may be static for some time, object saliency and motion clues which are widely used in traditional video segmentation approaches could be unreliable. Besides, its online processing ability is of crucial importance. In this paper, we propose an online video object segmentation method, which is organized as a segmentation-by-detection framework. It first effectively segments out meaningful objects in short video frame batches with the help of object detectors, and then the segmentation of different batches can be associated with each other via bi-directional notebook based connecting strategy. Experiments conducted on public available datasets verify the above-mentioned property and show good performance of the proposed method.
Highlights
With the rapid development of smart devices, efficient multimedia processing modules are more required than any times before
Papazoglou et al [30] defined the objects according to the motion cues and designed a fast strategy to estimate whether a pixel is inside the object, the segmentation will be refined in whole video scope with appearance model and spatial constraint
We propose to solve the pixel-wise object segmentation task via object detection in this paper, which can reduces the dependence on elaborate training data, unreliable motion and saliency clues in static and complex environment
Summary
With the rapid development of smart devices, efficient multimedia processing modules are more required than any times before. Traditional video segmentation methods will be greatly limited By contrast, for such complex scenes which usually appears in robot applications, we design segmentation-bydetection strategy in a streaming way, which is more effective and robust for such less constrained robot collecting videos and satisfies the demand of stream processing in robot application. 1) We propose a new streaming video object segmentation framework for complex video sequences, which usually appear in vision robot applications In these videos, multiple targets may appear in a single coarse scene and may not exist in each frame. 4) Bi-directional notebook connection method, which utilizes bi-directional feature association of segmentation records, is proposed to connect the segmentation results in different frame batches
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have