Implementation of HOG algorithm for real time object recognition applications on FPGA based embedded system

Abstract

Özetçe HOG algoritması son yıllarda kompleks içerikli görüntülerde nesne tanıma işlevini yüksek başarım ile gerçekleştirebilen bir algoritma olarak kullanılmaktadır. Bu basit fakat etkin tanımlayıcının donanımsal gerçeklenmesi bir çok temel örüntü tanıma işlevinde kullanılabilirliği söz konusu olduğundan önemlidir. Bu çalışmada HOG algoritmasının, farklı geometrik şekilleri yüksek bir başarım ile tanıyabilen donanımsal uyarlaması FPGA üzerinde gerçekleştirilmiştir. İncelenen görüntü içerisinde geometrik şeklin yatay ve dikey kenar bulma algoritmaları kullanılarak gradyan büyüklük ve açıları hesaplanmıştır. Elde edilen sonuçlar histogram üretmek için kullanılmış ve bu histogramların her bir şekil için ayırt edici karakteristiğe sahip olduğu gözlemlenmiştir. Kural tabanlı bir sınıflandırıcı uyarlanarak gerçek zamanlı olarak nesne tanıma uygulamalarında kullanılabilecek bir gömülü sistem yaklaşımı başarı ile sunulmuştur. Abstract Recent years HOG algorithm has been used to recognize objects in images, with complex content, with a very high success rate. Hardware implementation of this algorithm is very important because of the fact that it can be used in many object recognition applications. In this work HOG algorithm is implemented on FPGA to recognize different geometrical figures with a very high success rate. Objects vertical and horizontal edges have been sharpened using edge detection algorithms to calculate magnitude and angle of the local gradients. Obtained result are used to construct the histograms of gradient orientation. It is observed that each constructed histogram have distinctive features for every object. Rule based classifiers has been used to implement a successful real time object recognition approach on embedded system.

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

“…In this study, a computationally light but powerful robust method, based on histogram of oriented gradients (HOG) and average magnitude difference function (AMDF), is proposed. Although a similar line of studies is reported in the literature [5,9,13], which employ HOG for shape detection, we are proposing a new hardware-friendly algorithm, which employs AMDF as an effective and easy-to-implement decision module on FPGA. The method uses a set of features extracted from the HOG module, which is later classified by the AMDF module as explained in [14][15][16].…”

“…(4). In order to avoid the nonlinear arctangent operation for the angle calculation, several approximation methods are proposed such as 'piece-wise' [5], 'look up tables' (LUT) [10], and 'CORDIC' [20]. As the accurate calculation of the HOG values mainly depends on the angle of the gradient, θ xy , the angle calculation is highly important for the performance of the proposed method.…”

“…Implementation of image processing algorithms on reprogrammable hardware devices with an ability of parallel processing became a 'hot' research topic, and plenty of work has been reported in the literature [1][2][3][4][5]. As the requirement of computationally intensive image-processing algorithms for real-time industrial applications, such as robotic vision, imposes high-speed and highly parallel implementations, in recent years a trend has been witnessed towards utilizing field programmable gate arrays (FPGAs) or graphics processing units (GPUs) instead of using CPUs, which are inherently sequential processing devices.…”

GPU implementations of object detection using HOG features and deformable models

“…Previous work on the implementation of HOG-based object detection are limited to either hardware implementations [8], [9], [13] or specific parts of HOG algorithms [1], [20]. There is even no quantitative investigation of what implementation issues could prevent HOG-based object detection from being deployed in real-world applications.…”