Accelerated Moving Humans Detection Algorithm using Combined Global Descriptors on GPU Based on CUDA

Abstract

Day by day, the ability to detect and to identify automatically the objects among images and videos without constraint has becoming more and more important. The security systems, robots, smartphones and smart devices need to know the semantic meaning of image. The increase in object detection and identification algorithms is essentially related to the increase in complex object specification and authentication techniques. This could be resolved only when using the parallel architectures that can support heavy parallel processing such as GPU. In this chapter, we propose to present an implementation of moving humans detection algorithm on GPU based on the programming language CUDA. We proposed an implementation of an algorithm to extract the image features using the Fourier descriptor on GPU. We have proposed a second implementation to extract the image features based on the HOG descriptor on GPU. To detect the moving objects, we have implemented a background subtraction algorithm based on the GMM: Gaussian Mixture Model on GPU. In order to integrate these implementations in the main moving humans detection algorithm, the use of preprocessing and filtering techniques is necessary at this level as well as the CCL: Connected Component Labeling method which allows extracting the Moving objects from the rest of the image. The implementation of such kind of algorithm on GPU allows a great performance in terms of execution time.KeywordsHuman detectionFourier descriptorCompute Unified Device ArchitectureHistogram of oriented gradientsSupport Vector MachineGraphics Processing UnitGaussian Mixture Model