Hardware implementation of a scale and rotation invariant object detection algorithm on FPGA for real-time applications

Peker, Murat; Altun, Halis; Karakaya, Fuat

doi:10.3906/elk-1408-187

Cited by 4 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…HOG tanımlayıcılarının arkasındaki temel fikir, bir görüntü içindeki yerel nesne görünümünün ve şeklinin yoğunluk eğilimlerinin veya kenar yönlerinin dağılımı ile tanımlanabilmesidir. HOG bilgisayarla görü ve görüntü işlemede sıkça kullanılmaktadır [31,32].…”

Section: Hog öZelliklerinin çıKarılması (Hog Feature Extraction)unclassified

GMACA ile Hareket Tespiti Yapılan Video Görüntülerde İnsan Hareketlerinin Tanınması

Peldek¹,

Becerikli²

2018

Gazi Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi

View full text Add to dashboard Cite

Highlights:Graphical/Tabular Abstract  The ability of the generalized multiple attractor cellular automata (GMACA) to attract the patterns was used in image processing.  Motion detection has been performed on video images with GMACA.  Motion detection images detected with the GMACA have been used to recognize human actions is a complex computer vision problem. Figure A. Recognition of human action in motion detected image with GMACAPurpose: Generalized Multiple Attractor Cellular Automata (GMACA) is the type of Cellular Automata applied to more than one cell using the rule vector. Obtaining the rule vector in the application of the GMACA has a critical precaution. The purpose of the present study is to recognize human action in motion detected images obtained with GMACA. Theory and Methods:In this study, the GMACA rule vector, which is required to produce a single-length cycle attractor, was obtained using reachability tree-based methods. Detection of human motion in video images has been accomplished using attractors generated by the rule vector. In the developed action recognition method, the video images are first converted to the gray color space. Then, the GMACA rule vector to be used for motion detection is created.Motion detection is performed using GMACA. The HOG feature vector is extracted from the motion detection images and the resulting HOG feature vectors are labeled according to their motion. The dataset is created in this way. The generated dataset is decomposed into training and test data sets by cross-validation. Recognition of human action is performed by the SVM method. Experimental results are shown by the confusion matrix Results:The overall accuracy rate on the KTH dataset is 72.7%. The accuracy rates of boxing and hand clapping actions are 83% and 82%, respectively. The overall accuracy rate on the WEIZMANN data set is 77.3%. Accuracy rates for on-the-spot actions in the WEIZMANN dataset up to 98%. These ratios indicate that the motion detection method using GMACA can be used in action recognition applications. Conclusion:In this study, action recognition using motion detection images obtained with GMACA shows that the developed method can be used in complex computer vision applications. The operating logic, which allows for discrete and simultaneous calculation, makes the use of cellular automata in image processing and computer vision applications advantageous.

show abstract

Section: Hog öZelliklerinin çıKarılması (Hog Feature Extraction)unclassified