Computer
Suha Dh. Athab; Kesra Nermend; Abdulamir Abdullah Karim
Abstract
Microsoft Common Objects in Context (COCO) is a huge image dataset that has over 300 k images belonging to more than ninety-one classes. COCO has valuable information in the field of detection, segmentation, classification, and tagging; but the COCO dataset suffers from being unorganized, and classes ...
Read More ...
Microsoft Common Objects in Context (COCO) is a huge image dataset that has over 300 k images belonging to more than ninety-one classes. COCO has valuable information in the field of detection, segmentation, classification, and tagging; but the COCO dataset suffers from being unorganized, and classes in COCO interfere with each other. Dealing with it gives very low and unsatisfying results whether when calculating accuracy or intersection over the union in classification and segmentation algorithms. A simple method is proposed to create a customized subset from the COCO dataset by determining the class or class numbers. The suggested method is very useful as preprocessing step for any detection or segmentation algorithms such as YOLO, SSPNET, RCNN, etc. The proposed method was validated using the link net architecture for semantic segmentation. The results after applying the preprocessing were presented and compared to the state of art methods. The comparison demonstrates the exceptional effectiveness of transfer learning with our preprocessing model.
BSc; Zaki Y. Abid; Thamir R. Saeed; Sameir A. Aziez
Volume 15, Issue 1 , April 2015, , Page 1-17
Abstract
Abstract – This paper presents a moving object tracker for monitoring system which can be used in a smart city. Kernel density estimation (KDE) algorithm has been used for representing a background model, while a minimum distance between the current image and the background has been used to extract ...
Read More ...
Abstract – This paper presents a moving object tracker for monitoring system which can be used in a smart city. Kernel density estimation (KDE) algorithm has been used for representing a background model, while a minimum distance between the current image and the background has been used to extract the foreground. Also, morphological operations are carried out to remove the noise regions and to filter out ambiguous areas. The performance has been evaluated by determining the true, false, and miss detections of an object area. The optimal results have been obtained by adjusting the morphological operation sequence to be (close > thicken) combination by which the true-hits are 14 out of 16 while miss-number is 2 and zero false-hits, While, the percentage hit ratio was 87.5% (14 out of 16). Also, the salt noise introduction in video reduces the hit number from 14 to 11 when it increases from zero to 0.5 percent of the total frame pixels. The accepted absolute error ratio (in morphological properties of the matched object) is kept at 0.05 for all tests. The implementation has been built by using a combination of two platforms, ISE 14.6(2013) and Matlab(2013a) platforms, to avoid the size weakness of XC3S700A-FPGA board.