Acquiring data Kinect v2

Start kinect2-nidaq.

Important Note: Ensure that your depth cameras are parallel to the bucket floor and that the buckets are flat in the cage shells. Environmentally attributed angles can increase the potential for extraction errors/anomalies down the line.

First, make sure you can stream frames correctly, tick the PREVIEW MODE box and click START SESSION. You should see RGB and depth frames over on the PREVIEW tab.

Note: on the bottom right of the PREVIEW tab, there are 2 sliders indicating the minimum and maximum distances to capture. Ensure that your distances have been set to a range similar to that of the real-life setup. For example, if the camera is ~70cm/27inches from the bucket floor, then the Max distance (mm) should equal ~705, while the Min distance (mm) should equal ~450, depicting a rough distance from the top of the bucket to the camera.

Then, you have a couple of options for acquiring data:

1. Record frames for a pre-specified amount of time
2. Record until you click the stop button

For both methods, it is highly recommended to record the Subject Name and Session Name in the appropriate fields. Common things to use here might be Subject Name MOUSEID7 and Session Name Odor exposure. Finally, choose the directory you want to store your data in. If you leave COMPRESS DATA unticked, then a new folder with a timestamp will be created with the following files,

• depth_ts.txt --> timestamps from the camera (column 1) timestamps from the National Instruments board (column 2, zeros if the board is off or not used)
• depth.dat --> uint16, Little-endian encoded depth videos
• metadata.json --> metadata from the session

If you leave this box ticked, then these files will be stored in a gzipped tarball with a timestamp in the same directory.

To record for 45 minutes, simply untick the PREVIEW MODE box and write 45 in the RECORDING TIME (MINUTES) field. To record depth video, make sure DEPTH STREAM is ticked, to record RGB video tick COLOR STREAM (color video is not used for any downstream analysis in MoSeq2). Finally, click on START SESSION and you should see a countdown bar in the bottom left of the window.

To record until you click STOP SESSION, make sure the RECORDING TIME (MINUTES) field is empty, then tick the RECORD UNTIL USER CLICKS STOP. Finally, click START SESSION and frames will be streamed to disk until you click STOP SESSION.

If you record any data using the National Instruments board, you can use a Python function like the following to read it in:

import numpy as np

def load_nidaq_data(filename, nch=2, dtype='<f8'):

    with open(filename, "rb") as file_read:
        dat = np.fromfile(file_read, dtype)

    nidaq_dict = {}

    for i in range(nch-1):
        nidaq_dict['ch{:02d}'.format(i)] = dat[i::nch]

    nidaq_dict['tstep'] = dat[nch-1::nch]

    return nidaq_dict

Where nch is the total number of channels you recorded. The file format is 32-bit little-endian floating-point, and bytes are written from each channel in series followed by the timestamp from the National Instruments board, so, e.g. ch1--time1 | ch2--time1 | ch3--time1 | timestamp1 | ch1--time2 | ch2--time2 | ch3--time2 | timestamp2. The timestamps are timestamps from the National Instruments board clock, which are also copied to depth_ts.txt in the second column (here, the first column contains timestamps from the camera, and the second column contains the nearest NIDAQ timestamp). Use the second column to align data from the National Instruments stream to the Kinect stream.