Description
Cameras (RGB, RGB-D / Kinect, IR) appear in the bibliography as the vision-modality baseline that Wi-Fi sensing is repeatedly compared against. The CSI sensing pitch is exactly "achieve camera-grade activity recognition without the privacy / lighting / occlusion problems of cameras", so any survey or motivating paragraph cites cameras as the contrast. They also appear in cross-modal supervision papers (camera-derived poses used as ground truth for CSI-based pose estimation, e.g., WiFi + camera CKD systems). For the thesis, cameras are not a deployment target but the rhetorical reference point.
Specs / capabilities
- RGB: standard webcam to high-resolution surveillance camera
- RGB-D: Microsoft Kinect (v1 / v2 / Azure), Intel RealSense, ZED stereo
- Thermal / IR: FLIR variants, used for low-light privacy-aware sensing
- Computer-vision pipelines (OpenPose, MediaPipe, YOLO) provide pose / detection ground truth
Used by (papers)
- Survey / motivation sections asserting privacy advantage of CSI sensing
- Cross-modal teacher-student training (camera teacher → CSI student)
- Vision-baseline comparisons in HAR / crowd-counting evaluations