Pedestrian classification on transfer learning based deep convolutional neural network for partial occlusion handling

<span lang="EN-US">The investigation of a deep neural network for pedestrian classification using transfer learning methods is proposed in this study. The development convolutional networks has significantly improved the autonomous driver assistance system classification. However, presence partially occluded parts and appearance variation under complex scenes are still robust to challenge detection system. To address problem, we six models: end-to-end (CNN) model, scratch-trained residual (ResNet50) four visual geometry group 16 (VGG16), GoogLeNet (InceptionV3), ResNet50, MobileNet. performance was evaluated publicly datasets: </span><em><span lang="EN-US">Institut National de Recherche en Sciences et Technologies du Numérique</span></em><span lang="EN-US"> (INRIA), Prince Songkla University (PSU), CVC05, Walailak (WU) datasets. experimental results show that models achieve accuracy 65.2% (end-to-end CNN), 92.92% (scratch-trained ResNet50), 97.15% (pre-trained VGG16), 94.39% InceptionV3), 90.43% 98.69% MobileNet) data from Southern Thailand (PSU dataset). Further analysis reveals deeper ConvNet architecture, more specific information features provided. In addition, architecture can distinguish patterns while being trained with samples.</span>