The motivation behind our work is to improve commuters overall train ride experience by providing crowd density information of incoming trains. Given the incoming train information, commuters can make an informed decision about which train cabin to board preferring those cabins with lower crowd density. Our proposed solution is to process extracted images from train cabin security footages using Convolutional Denoising Autoencoder-Convolutional Neural Network (CDAE-CNN) to predict crowd density in image frames. Experiment results show that adding the CDAE processing to the framework improves the performance in reconstructing noisy images before feeding the images as input data to CNN, thus overall improving the classification accuracy. Experimental results show that CDAE-CNN consistently outperforms CNN in labeling the train cabin image frames in various image datasets.