Real-time object detection on CUDA

Abstract

The aim of the research described in this article is to accelerate object detection in images and video sequences using graphics processors. It includes algorithmic modifications and adjustments of existing detectors, constructing variants of efficient implementations and evaluation comparing with efficient implementations on the CPUs. This article focuses on detection by statistical classifiers based on boosting. The implementation and the necessary algorithmic alterations are described, followed by experimental measurements of the created object detector and discussion of the results. The final solution outperforms the reference efficient CPU/SSE implementation, by approximately 6-89 for high-resolution videos using nVidia GeForce 9800GTX and Intel Core2 Duo E8200.

Fast and accurate circle tracking using active contour models

“…Nowadays, GPU implementation is a commonly used technique to parallelize and speed up computer vision algorithms. For instance, in [37] authors propose a GPU implementation of circle Hough transform, in [15,17,[25][26][27] authors propose GPU implementation for object segmentation and detection and in [13,22,38] authors propose GPU-CUDA implementations for image denoising and restoration.…”

Macrofeature layout selection for pedestrian localization and its acceleration using GPU

“…It has been successfully applied to various problems in computer vision including object detection [27][28][29] for accelerating algorithms through parallelization. The CUDA architecture is composed of several streaming multiprocessors sharing a large amount of memory called global memory.…”

“…Recently, several efficient HOG implementations using CUDA have been presented in [30,[27][28][29] but most of them are based on the exact implementation of the standard HOG. We modify the standard HOG, and present our HOG implementation compared to the standard HOG in Fig.…”

Algorithms for Efficient Computation of Convolution

“…The naïve convolution on the graphics hardware has been also described in [35] and included in the Nvidia Performance Primitives library [36]. Specific applications, namely Canny edge detection [37,38] or real-time object detection [39] have been studied in the literature. It can be noted that the problem of computing a rank filter such as the median filter has a naïve solution similar to the one of the convolution.…”