Certain investigations on pedestrian detection using multiresolution analysis and deep learning

Abstract

In computer vision, pedestrian detection is a rapidly growing researcharea due to the importance of real time applications such as video surveillancesystem, person identification, automated driver assistance and automotivesafety. Due to the vulnerability of pedestrians in heavy traffic, especially inurban areas, the detection of pedestrians becomes essential for road safety.The existing techniques of Deformable Part Model, extended deep model,RealBoost method and Deep Neural Networks have been employed indifferent scenarios for pedestrian detection. To improve the accuracy, reducethe false positive rate, memory requirement and time consumption (trainingand response time) over the existing methods, three novel techniques namely Multiresolution Morlet Decomposition based Iterative Learning DeformablePart (MMD-ILDP) model, Soft sign Gaussian Recurrent Deep Neural Network(SGRDNN) technique and Multi-Resolution Deformable Part Based Fully Recurrent Deep Neural Learning (MDP-FRDNL) have been proposed in this research work. newline