Edit

Share via

Quickstart: Build an Azure Kinect body tracking application

  • Article

Getting started with the Body Tracking SDK? This quickstart will get you up and running with body tracking! You can find more examples in this Azure-Kinect-Sample repo.

Prerequisites

Headers

Body tracking uses a single header, k4abt.h. Include this header in addition to k4a.h. Make sure your compiler of choice is set up for both the Sensor SDK and the Body Tracking SDK lib and include folders. You also need to link to k4a.lib and k4abt.lib files. Running the application requires k4a.dll, k4abt.dll, onnxruntime.dll, and dnn_model.onnx to be in the applications execution path.

#include <k4a/k4a.h>
#include <k4abt.h>

Open device and start the camera

Your first body tracking application assumes a single Azure Kinect device connected to the PC.

Body tracking builds on the Sensor SDK. To use body tracking, you first need to open and configure the device. Use the k4a_device_open() function to open the device and then configure it with a k4a_device_configuration_t object. For best results set the depth mode to K4A_DEPTH_MODE_NFOV_UNBINNED or K4A_DEPTH_MODE_WFOV_2X2BINNED. The body tracker will not run if the depth mode is set to K4A_DEPTH_MODE_OFF or K4A_DEPTH_MODE_PASSIVE_IR.

You can find more information on finding and opening the device on this page.

You can find more information on Azure Kinect depth modes on these pages: hardware specification and k4a_depth_mode_t enumerations.

k4a_device_t device = NULL;
k4a_device_open(0, &device);

// Start camera. Make sure depth camera is enabled.
k4a_device_configuration_t deviceConfig = K4A_DEVICE_CONFIG_INIT_DISABLE_ALL;
deviceConfig.depth_mode = K4A_DEPTH_MODE_NFOV_UNBINNED;
deviceConfig.color_resolution = K4A_COLOR_RESOLUTION_OFF;
k4a_device_start_cameras(device, &deviceConfig);

Create the tracker

The first step in getting body tracking results is to create a body tracker. It needs the sensor calibration k4a_calibration_t structure. The sensor calibration can be queried using the k4a_device_get_calibration() function.

k4a_calibration_t sensor_calibration;
k4a_device_get_calibration(device, deviceConfig.depth_mode, deviceConfig.color_resolution, &sensor_calibration);

k4abt_tracker_t tracker = NULL;
k4abt_tracker_configuration_t tracker_config = K4ABT_TRACKER_CONFIG_DEFAULT;
k4abt_tracker_create(&sensor_calibration, tracker_config, &tracker);

Get captures from the Azure Kinect device

You can find more information on retrieving image data on this page.

// Capture a depth frame
k4a_capture_t sensor_capture;
k4a_device_get_capture(device, &sensor_capture, TIMEOUT_IN_MS);

Enqueue the capture and pop the results

The tracker internally maintains an input queue and an output queue to asynchronously process the Azure Kinect DK captures more efficiently. The next step is to use the k4abt_tracker_enqueue_capture() function to add a new capture to the input queue. Use the k4abt_tracker_pop_result() function to pop a result from the output queue. The timeout value is dependent on the application and controls the queueing wait time.

Your first body tracking application uses the real-time processing pattern. Refer to get body tracking results for a detailed explanation of the other patterns.

k4a_wait_result_t queue_capture_result = k4abt_tracker_enqueue_capture(tracker, sensor_capture, K4A_WAIT_INFINITE);
k4a_capture_release(sensor_capture); // Remember to release the sensor capture once you finish using it
if (queue_capture_result == K4A_WAIT_RESULT_FAILED)
{
    printf("Error! Adding capture to tracker process queue failed!\n");
    break;
}

k4abt_frame_t body_frame = NULL;
k4a_wait_result_t pop_frame_result = k4abt_tracker_pop_result(tracker, &body_frame, K4A_WAIT_INFINITE);
if (pop_frame_result == K4A_WAIT_RESULT_SUCCEEDED)
{
    // Successfully popped the body tracking result. Start your processing
    ...

    k4abt_frame_release(body_frame); // Remember to release the body frame once you finish using it
}

Access the body tracking result data

The body tracking results for each sensor capture are stored in a body frame k4abt_frame_t structure. Each body frame contains three key components: a collection of body structs, a 2D body index map, and the input capture.

Your first body tracking application only accesses the number of detected bodies. Refer to access data in body frame for detailed explanation of data in a body frame.

size_t num_bodies = k4abt_frame_get_num_bodies(body_frame);
printf("%zu bodies are detected!\n", num_bodies);

Clean up

The final step is to shut down the body tracker and release the body tracking object. You also need to stop and close the device.

k4abt_tracker_shutdown(tracker);
k4abt_tracker_destroy(tracker);
k4a_device_stop_cameras(device);
k4a_device_close(device);

Full source

#include <stdio.h>
#include <stdlib.h>

#include <k4a/k4a.h>
#include <k4abt.h>

#define VERIFY(result, error)                                                                            \
    if(result != K4A_RESULT_SUCCEEDED)                                                                   \
    {                                                                                                    \
        printf("%s \n - (File: %s, Function: %s, Line: %d)\n", error, __FILE__, __FUNCTION__, __LINE__); \
        exit(1);                                                                                         \
    }                                                                                                    \

int main()
{
    k4a_device_t device = NULL;
    VERIFY(k4a_device_open(0, &device), "Open K4A Device failed!");

    // Start camera. Make sure depth camera is enabled.
    k4a_device_configuration_t deviceConfig = K4A_DEVICE_CONFIG_INIT_DISABLE_ALL;
    deviceConfig.depth_mode = K4A_DEPTH_MODE_NFOV_UNBINNED;
    deviceConfig.color_resolution = K4A_COLOR_RESOLUTION_OFF;
    VERIFY(k4a_device_start_cameras(device, &deviceConfig), "Start K4A cameras failed!");

    k4a_calibration_t sensor_calibration;
    VERIFY(k4a_device_get_calibration(device, deviceConfig.depth_mode, deviceConfig.color_resolution, &sensor_calibration),
        "Get depth camera calibration failed!");

    k4abt_tracker_t tracker = NULL;
    k4abt_tracker_configuration_t tracker_config = K4ABT_TRACKER_CONFIG_DEFAULT;
    VERIFY(k4abt_tracker_create(&sensor_calibration, tracker_config, &tracker), "Body tracker initialization failed!");

    int frame_count = 0;
    do
    {
        k4a_capture_t sensor_capture;
        k4a_wait_result_t get_capture_result = k4a_device_get_capture(device, &sensor_capture, K4A_WAIT_INFINITE);
        if (get_capture_result == K4A_WAIT_RESULT_SUCCEEDED)
        {
            frame_count++;
            k4a_wait_result_t queue_capture_result = k4abt_tracker_enqueue_capture(tracker, sensor_capture, K4A_WAIT_INFINITE);
            k4a_capture_release(sensor_capture); // Remember to release the sensor capture once you finish using it
            if (queue_capture_result == K4A_WAIT_RESULT_TIMEOUT)
            {
                // It should never hit timeout when K4A_WAIT_INFINITE is set.
                printf("Error! Add capture to tracker process queue timeout!\n");
                break;
            }
            else if (queue_capture_result == K4A_WAIT_RESULT_FAILED)
            {
                printf("Error! Add capture to tracker process queue failed!\n");
                break;
            }

            k4abt_frame_t body_frame = NULL;
            k4a_wait_result_t pop_frame_result = k4abt_tracker_pop_result(tracker, &body_frame, K4A_WAIT_INFINITE);
            if (pop_frame_result == K4A_WAIT_RESULT_SUCCEEDED)
            {
                // Successfully popped the body tracking result. Start your processing

                size_t num_bodies = k4abt_frame_get_num_bodies(body_frame);
                printf("%zu bodies are detected!\n", num_bodies);

                k4abt_frame_release(body_frame); // Remember to release the body frame once you finish using it
            }
            else if (pop_frame_result == K4A_WAIT_RESULT_TIMEOUT)
            {
                //  It should never hit timeout when K4A_WAIT_INFINITE is set.
                printf("Error! Pop body frame result timeout!\n");
                break;
            }
            else
            {
                printf("Pop body frame result failed!\n");
                break;
            }
        }
        else if (get_capture_result == K4A_WAIT_RESULT_TIMEOUT)
        {
            // It should never hit time out when K4A_WAIT_INFINITE is set.
            printf("Error! Get depth frame time out!\n");
            break;
        }
        else
        {
            printf("Get depth capture returned error: %d\n", get_capture_result);
            break;
        }

    } while (frame_count < 100);

    printf("Finished body tracking processing!\n");

    k4abt_tracker_shutdown(tracker);
    k4abt_tracker_destroy(tracker);
    k4a_device_stop_cameras(device);
    k4a_device_close(device);

    return 0;
}

Next steps

Get body tracking results