Weakly Supervised Realtime Dynamic Background Subtraction

Abstract

Background subtraction is a fundamental task in computer vision with numerous real-world applications, ranging from object tracking to video surveillance. However, dynamic backgrounds can pose a significant challenge in this problem. While various methods have been proposed for background subtraction, supervised deep learning-based techniques are currently considered state-of-the-art. However, these methods require pixelwise ground-truth labeling, which can be time-consuming and expensive. In this work, we propose a weakly supervised framework that can perform background subtraction without requiring perpixel ground-truth labels. Our framework is trained on a moving object-free sequence of images and comprises two networks. The first network is an autoencoder that generates static background images and prepares dynamic background images for training the second network. The dynamic background images are obtained by thresholding the background-subtracted images. The second network is a U-Net that uses the same moving object-free video for training and the dynamic background images as pixel-wise ground-truth labels. During the test phase, the input images are processed by the autoencoder and U-Net, which generate static and dynamic background images, respectively. The dynamic background image helps remove dynamic motion from the background subtracted image, enabling us to obtain a foreground image that is free of dynamic artifacts. To demonstrate the effectiveness of our method, we conducted experiments on selected categories of the CDnet 2014 dataset and the I2R dataset. Our method outperformed all top-ranked unsupervised methods. It also surpassed one of the two existing weakly supervised methods, while achieving comparable results to the other method but with a shorter running time. Our proposed method is online, realtime, efficient, and requires minimal frame-level annotation, making it suitable for a wide range of real-world applications.

Investigating Training Datasets of Real and Synthetic Images for Outdoor Swimmer Localisation with YOLO

“…In other words, we investigated the merging of components of two datasets involving humans as an additional augmentation technique. In the realm of background removal, an emerging trend is to incorporate deep learning-based approaches [28][29][30] to accomplish the background removal task. A comparison study in [29] revealed the practicability of YOLO models [20] in comparison to other feature engineering-based approaches [31,32] in removing distinct moving particles from their liquid background contexts.…”

“…A comparison study in [29] revealed the practicability of YOLO models [20] in comparison to other feature engineering-based approaches [31,32] in removing distinct moving particles from their liquid background contexts. The proposed method in [28] incorporates an autoencoder along with a U-Net [33] to accomplish static and dynamic background generation tasks, respectively. Here, a set of static backgrounds were used to train an autoencoder.…”