Realsense-K4A_depth_comparison

Overview

All the required codes for data capture to the analysis of all the experiments mentioned in the research paper Comparing Depth Estimation of Azure Kinect and Realsense D435i Cameras are present in the codes/ directory.

File Structure

└── code
    ├── config
    │   ├── azurekinect.json
    │   ├── default_config.json
    |   └── realsense.json
    ├── experiments
    │   ├── ambiance.py
    │   ├── camera_distance.py
    │   ├── object_color.py
    │   ├── object_motion.py
    │   ├── object_texture.py
    │   └── temperature.py
    ├── libs
    │   ├── experiments.py
    │   ├── type_convert.py
    │   └── visualization.py
    ├── recorder
    │   ├── azure_kinect_mkv_reader.py
    │   ├── azure_kinect_recorder.py
    │   ├── realsense_recorder.py
    │   └── viewer_realsense.py
    └── utils
    |    ├── file.py
    |    └── initialize_config.py
    └── requirements.txt

Data Capture

The recorder/ directory contains all the required Python codes for data capture for both cameras.

Azure Kinect

Record Data:

python3 recorder/azure_kinect_recorder.py --config config/azurekinect.json --output <output.mkv> -a --len <time_in_sec>

Export Data:

python3 recorder/azure_kinect_mkv_reader.py --input <path-to-captured-mkv-file> --output <path-to-output-dir>

RealSense

python3 recorder/realsense_recorder.py --config config/realsense.json --record_imgs --iteration <nth-iteration> --len <time_in_sec> --npy <path-to-save-data> --start <x1 y1 :starting-pixel-of-frame> --end <x2 y2 :ending-pixel-of-frame>

Experiments

libs/ and experiments/ directories contain all the required Python codes for data visualization and analysis for each experiment.

Azure Kinect outputs are .mkv video, and color and depth images, so we need to extract the depth data from each depth image in the depth/ directory. For that, we have added an export_depth flag in each experiment code which will render a color image. When one/multiple rectangular ROIs are drawn on that color image, depth data within the ROIs will be extracted for all the depth images within the depth/ directory of the corresponding color/ directory and saved as a .npy file.

Export Azure Kinect depth

python3 experiments/<experiment_name>.py --input <path-to-color-dir> <path-to-depth-dir> --npy <npy-filename> --export_depth --camera_id 1

Visualize data

python3 experiments/<experiment_name>.py --npy <path-to-single/multiple-npy-files> --camera_id 0 (realsense) or 1 (kinect) --fig_path <path-to-save-image>

The visualizations are different for different experiments, so you will need to see the code in detail for that. However, in general, we have 3 different visualizations:

1. Heatmap
2. Temporal Plot
3. Distribution Check

For each of these, we have respective flags. You will need to set them to True as per your need.

Contact

For pre-recorded data and further information, please contact the authors through the paper. Full version pre-print of the paper is available here. Paper will soon be published on LNNS lecture series of Springer.