OCCLUSIONS IN IMAGE PROCESSING
In image processing, occlusions refer to scenarios where objects in a scene are
partially or fully blocked by other objects, making them invisible or only partially
visible in the captured image.
TYPES OF OCCLUSIONS:
• SELF-OCCLUSION:
Parts of an object occlude other parts of the
same object.
• INTER-OBJECT OCCLUSION:
One object occludes another object in the
scene.
CHALLENGES
Occlusions poses significant challenges for various computer vision
tasks such as :
OBJECT DETECTION OBJECT RECOGNITION OBJECT TRACKING 3D RECONSTRUCTION
TECHNIQUES TO HANDLE OCCLUSIONS:
• DEPTH SENSING:
Using sensors like LiDAR or stereo cameras to
differentiate objects based on their distance
DEPTH SENSING
LiDAR (LIGHT DETECTION AND RANGING) STERIO CAMERA
MULTI-VIEW APPROACHES:
Using images from different angles to see and reconstruct hidden parts.
TEMPORAL INFORMATION
• Using data from earlier and later frames in a video to guess what occluded parts
look like.
SEMANTIC SEGMENTATION:
Dividing an image into different object classes to understand the scene and manage
occlusions better.
PREDICTIVE MODELING:
Using machine learning to predict what hidden parts of an object might look like.
INPAINTING:
Filling in missing parts of an image using surrounding pixels.
APPLICATIONS:
• Autonomous Driving: Detecting and tracking cars even when they're
partially hidden.
• Augmented Reality (AR): Adding virtual objects to real-world scenes
where occlusions might happen.
• Medical Imaging: Dealing with occlusions in medical scans to detect and
diagnose conditions accurately.
• Surveillance: Recognizing and tracking people or objects in crowded
places.