I am performing body tracking using two ZED X cameras connected to a ZED box, and I am fusing the results from both cameras on my main Ubuntu PC via Ethernet. When I check the operation using the Python sample for camera fusion, I observe multiple body tracking results even though there is only one person in the field. This is particularly pronounced in the overlapping areas of the camera views, where afterimages appear when the tracked person moves, and duplicates occur even when the person is stationary, within a distance in around 2 meters.
Could you please advise on how to avoid this phenomenon? Alternatively, is this a limitation of the SDK’s performance?
Environment:
- ZED SDK 4.2.5
- ZED box Orin NX 16GB: body_tracking.py + streaming_sender.py (add streaming communication parameters), Ethernet connection with the main PC via network hub (communication ports = 30010, 30012)
- ZED X x2 connected to the ZED box via GMSL2 cable (cameras are positioned at a height of ~2m, with an elevation angle of -20 degrees, placed in adjacent corners of a ~6m square area at a ~45degree angle, approximately 6m apart)
- Main PC: Ubuntu 22.04 (64GB, RTX 4070 Super): camera fusion (Python sample: based on fused_cameras.py)
- Camera calibration was performed using ZED360, and there were no duplicates observed during that process.