- https://doi.org/10.1109/silcon67893.2025.11327427
Multi-Class Indoor Object Detection Using Optimized Convolutional Architecture on Home Objects-3K Dataset
- Nov 6, 2025
- D Janardhan Reddy +5 more
House object detection plays a pivotal role in enabling intelligent home automation, surveillance, and assistive technologies. The proposed model architecture integrates advanced modules—including convolutional layers, feature fusion blocks, and optimized detection heads—facilitating accurate multi-class localization. A robust data augmentation pipeline was employed using techniques like blur, grayscale conversion, and CLAHE to improve generalization. Experimental evaluation on the custom HomeObjects-3K dataset comprising 404 validation images with 3,466 object instances demonstrated strong detection performance. The model achieved an overall precision of 0.734, recall of 0.683, mAP@0.5 of 0.729.Notably photo frame (mAP@0.5: 0.876), sofa (0.878), and table (0.841) were detected with high accuracy.
