Waste auditing plays a crucial role in effectively reducing medical waste generated by resource-intensive operation rooms. Currently, the manual waste auditing method is time-consuming and hazardous. To address this issue, the i-WASTE system has been proposed, which utilizes video recordings from a camera-equipped waste container to detect and classify medical waste. Although the system is in its pilot study phase, it has shown promising potential. The dataset used for this study consists of four waste items: gloves, hairnet, mask, and shoe cover. These items share similarities in appearance, making accurate sorting a challenging task. However, achieving high accuracy in sorting these four items would indicate the potential of the proposed architecture to generalize well to a larger number of waste classes. The video dataset was collected and labelled personally in the laboratory setting. The process involved recording videos of waste items placed in the camera-equipped waste container. Subsequently, these videos were manually labelled to identify and categorize the waste items. To improve the efficiency of waste classification, a pre-processing method based on motion detection was developed. This method helps extract and trim useful frames in both spatial and temporal dimensions, reducing unnecessary computational load. To classify waste videos, a novel architecture called R3D+C2D was proposed. This architecture combines the features learned by 2D convolutional neural networks (C2D) and 3D convolutional neural networks (R3D). By leveraging both spatial and temporal information from the videos, the proposed method aims to enhance the accuracy of waste classification. The results obtained from this pilot study are promising, with the proposed method achieving a classification accuracy of 79.9% on the challenging dataset.