ViDeCanvas: Video Object Detection Visualization Canvas
Areeb Abbasi
Department of Computer Science
Faculty Supervisor: Shahrukh Humayoun
Real-time video object detection on resource-constrained devices remains a challenging task due to the high computational demands of both object detection algorithms and video decoding. In this work, we propose ViDeCanvas tool for efficient video object detection on low computational devices by analyzing only regions of interest (ROIs) within the video stream, rather than exhaustively processing every frame. ViDeCanvas used Pixel-based Spiral Radial (PixSR) visualization to provide a comprehensive view of object detection results within video streams. For this, it offers a unified grid view that effectively portrays the temporal occurrence of objects alongside their detection accuracy and frequency of appearance within frames. Additionally, the framework facilitates comparative analyses of object detection model accuracy, video sampling filters (dense, sparse), and video similarities. Furthermore, ViDeCanvas condenses the occurrences of detected objects into a concise and informative representation, enhancing the interpretability of video analytics results.