Real-Time Imaging and Machine Vision Systems


Mustafa Ünel	Real-Time Mosaicing of Aerial Images Image mosaicing aims to increase visual perception by composing visual data obtained from separate images since a composite image provides richer description than individual images. Gaining and maintaining situational awareness from image mosaics is important for both civil and military applications. Inspection of the urban areas suffering from natural disasters and examination of the large plantations are possible civil areas of utilization. For military applications, image mosaicing can provide critical information about enemy activities in a broad perspective. We have been developing novel mosaicing technqiues for creating seamless mosaics in real-time from a set of sequential images acquired from a UAV. Our aim is to reach a reasonable accuracy in the mosaic without using a computationally expensive framework such as bundle adjustment. Visualization and Tracking of Droplets and Bubbles Obtaining kinetical and structural information about small size particles, bubbles or droplets generated in jet flow serves for better understanding of the flow characteristics and enables manipulation of these entities in several microfluidic applications. Recent studies in droplet-based microfluidic systems necessitate to localize the droplets in fluidic systems and track them throughout the flow to extract their unique features. In the context of a project funded by TUBITAK, we have been developing new vision based tracking methods to track the evolution of the droplets in jet flow. A recently developed tracker fuses shape and motion features of the individually detected droplets and employs the Bhattacharyya distance to locate the tracked droplet(s) among candidate droplets in upcoming frames. Different from several existing droplet tracking algorithms, shapes of the target droplets are not assumed to be circles or ellipses; instead shape boundaries are extracted and used as they are. The evolution of single, double and triple droplets were tracked with the average accuracy of 87%, 87% and 83%, respectively. Physical parameters of tracked droplets are then utilized to explain the underlying phenomena.

Real-Time Mosaicing of Aerial Images

Image mosaicing aims to increase visual perception by composing visual data obtained from separate images since a composite image provides richer description than individual images. Gaining and maintaining situational awareness from image mosaics is important for both civil and military applications. Inspection of the urban areas suffering from natural disasters and examination of the large plantations are possible civil areas of utilization. For military applications, image mosaicing can provide critical information about enemy activities in a broad perspective.

We have been developing novel mosaicing technqiues for creating seamless mosaics in real-time from a set of sequential images acquired from a UAV. Our aim is to reach a reasonable accuracy in the mosaic without using a computationally expensive framework such as bundle adjustment.

Visualization and Tracking of Droplets and Bubbles

Obtaining kinetical and structural information about small size particles, bubbles or droplets generated in jet flow serves for better understanding of the flow characteristics and enables manipulation of these entities in several microfluidic applications. Recent studies in droplet-based microfluidic systems necessitate to localize the droplets in fluidic systems and track them throughout the flow to extract their unique features.

In the context of a project funded by TUBITAK, we have been developing new vision based tracking methods to track the evolution of the droplets in jet flow. A recently developed tracker fuses shape and motion features of the individually detected droplets and employs the Bhattacharyya distance to locate the tracked droplet(s) among candidate droplets in upcoming frames. Different from several existing droplet tracking algorithms, shapes of the target droplets are not assumed to be circles or ellipses; instead shape boundaries are extracted and used as they are. The evolution of single, double and triple droplets were tracked with the average accuracy of 87%, 87% and 83%, respectively. Physical parameters of tracked droplets are then utilized to explain the underlying phenomena.