Description of Invention
Object detection is crucial in computer vision for identifying and locating specific objects within images. This study uses the YOLOv11n model to detect objects in indoor environments, utilizing a dataset of 2,213 images across seven object classes. The dataset includes variations in background, lighting, and occlusion, which pose challenges for detection. To improve model performance, the images were resized and augmented. ResNet18 was used for feature extraction, and Grad-CAM++ was applied to visualize the areas influencing the model's decisions. Additionally, feature engineering techniques like Region of Interest (ROI), Histogram of Oriented Gradients (HOG), and Local Binary Patterns (LBP) were implemented to enhance detection capabilities. Principal Component Analysis (PCA) was applied for feature selection. The YOLOv11n model was trained and achieved impressive performance with precision values of 0.920 for training, 0.900 for validation, and 0.885 for testing. Recall values were 0.901 for training, 0.877 for validation, and 0.802 for testing. The mean Average Precision (mAP) at 50% Intersection over Union (mAP50) was 0.948 for training, 0.921 for validation, and 0.896 for testing, demonstrating improved detection accuracy across varied conditions.