GPU implementations of object detection using HOG features and deformable models

Abstract

Vision-based object detection using camera sensors is an essential piece of perception for autonomous vehicles. Various combinations of features and models can be applied to increase the quality and the speed of object detection. A wellknown approach uses histograms of oriented gradients (HOG) with deformable models to detect a car in an image [15]. A major challenge of this approach can be found in computational cost introducing a real-time constraint relevant to the real world. In this paper, we present an implementation technique using graphics processing units (GPUs) to accelerate computations of scoring similarity of the input image and the pre-defined models. Our implementation considers the entire program structure as well as the specific algorithm for practical use. We apply the presented technique to the real-world vehicle detection program and demonstrate that our implementation using commodity GPUs can achieve speedups of 3x to 5x in frame-rate over sequential and multithreaded implementations using traditional CPUs.

Face Occlusion Detection Using Skin Color Ratio and LBP Features for Intelligent Video Surveillance Systems

“…To address the problem, GPU technique is used to accelerate process by using a parallel implementation of the HOG algorithm. Comparing CPU implementation, GPU based on parallel architecture has a better computational performance [17], [18]. The result of head detection demonstrates that our implementation using GPU can achieve a speedup of over 5 times, which allows for real-time detection.…”

Connected Smartphones and High-Performance Servers for Remote Object Detection

“…However, the computing capability required to run these algorithms in real time is significant, and that of vehicular systems as embedded systems is inadequate. Even high-end graphics processing units (GPUs) do not have sufficient performance to achieve high frame rates if the range of object sizes to be detected is broad, i.e., if the system needs to detect both near and distant objects [4]. Our prototype system clarifies the achievable latency and frame rate in remote vision-based object detection and provides the roadmap for this technology to be deployed in the real world.…”

“…Figure 13 represents the execution time in the server for the two types of inputs, i.e., joint photographic experts group (JPEG) and histogram of oriented gradients (HOG) inputs, in the prototype system. A JPEG input is in the original format sent by the smartphone, while an HOG input [4] is HOG feature vectors that can express the shape of the object in the input image. In both cases, the execution time for the detection of objects is proportional to the frequency of resizing the images.…”

Reconocimiento de rostros en tiempo real sobre dispositivos móviles de bajo costo

“…El esquema de ventana deslizante es costoso en procesamiento, por lo que se acostumbra a emplear un esquema en cascada para acelerar su funcionamiento [14]. También es factible paralelizar el proceso, los trabajos [15] y [16] son ejemplos del uso de la GPU (Graphics Processing Unit) para acelerar el algoritmo. Lámpsakos | N°.…”