Intelligent Visual Recognition for Real-Time Object and Color Detection

Abstract

Object and color detection is an important part of visual recognition and scene understanding, because recognizing an object together with its color provides the contextual cues needed to interpret real-world situations. Accurate detection is difficult, however, because of changes in lighting, complex backgrounds, and differences in object shape, size, and orientation. The proposed system is designed to address these problems by using deep learning to identify objects and their dominant colors in still images, video streams, and live camera feeds. It uses a pretrained You Only Look Once (YOLO) detector, YOLOv8, trained on a standard benchmark dataset such as Common Objects in Context (COCO). For color analysis, the system extracts the pixel values of the detected object regions and applies preprocessing (resizing, pixel value normalization, noise reduction, and conversion from the RGB to the hue, saturation, and value (HSV) color space) to reduce sensitivity to illumination changes. The system outputs object labels with confidence scores and displays the detections with text annotations and bounding boxes. An easy-to-use graphical user interface (GUI) supports single-image, batch-image, and real-time detection. The system was demonstrated qualitatively on 15 test images; quantitative evaluation metrics were not computed.