[Legacy Report] Adaptive image segmentationfor pedestrian detection in far infrared vision
Image segmentation is one of crucial stages of video processing used for detection of pedestrians in infrared (IR)vision. A study of both passive (typically far IR) and active (typically near IR) technologies for IR visionand analysis of the state-of-the-art in information extraction from IR images was presented. Moreover, a deep testing of threecarefully selected segmentation methods for far IR images show high effectiveness of all techniques with single pedestrians and high contrastto the background. In case of IR frames with low contrast (e.g. with many various human objects in a group of pedestrians)the methods based on local (intra-frame) adaptation (e.g. theadaptive dual-thresholding method - ADT) operate much better than non-adaptive methods. A globaladaptation (e.g. in the Otsu method) helps to compensate variations of the frame intensity, which occur frame-by-frame. An extension of the ADT method was proposed, which veryeffectively (98 %) solves a common problem related to the non-uniformity of the IR radiation intensity upon one pedestrian.
Date and Time
Location
Hosts
Registration
Add Event to Calendar
Speakers
Karol Piniarski of Poznan University of Technology
Infrared vision systems
Biography:
Email:
Address:Poznan, Wielkopolskie, Poland