Dynamic Multiple Object Segmentation with Spatio-Temporal Filtering.

Area of Science:

• Computer Vision
• Image Processing
• Robotics

Background:

• Object detection and segmentation in dynamic environments are challenging.
• Moving cameras, like drones, introduce complex motion dynamics.
• Distinguishing multiple moving objects from background clutter is crucial for surveillance and autonomous systems.

Purpose of the Study:

• To develop and validate an algorithm for localizing and extracting multiple moving objects from image sequences captured by moving cameras.
• To differentiate moving objects from the background and from each other.

Main Methods:

• Construction of a motion feature space (distance and direction) to map feature point trajectories.
• Application of a clustering algorithm based on trajectory distinctiveness for object-background and object-object separation.
• Segmentation of complete moving objects by identifying source pixels within local regions.

Main Results:

• Significant differentiation in feature space between moving objects and background for trajectories exceeding 10 frames.
• 67% of frames correctly classified based on feature points.
• Accurate localization of moving objects with an average Intersection over Union (IOU) of 0.76 and average contour accuracy of 0.57.

Conclusions:

• The proposed algorithm effectively localizes and segments multiple moving objects in images from moving camera platforms.
• The method demonstrates robust performance in challenging dynamic scenarios.
• This approach offers a valuable tool for applications requiring precise moving object extraction from aerial imagery.