Note

This C++ sample is available only with our Professional plan. The corresponding Python sample is available with all Metavision Intelligence Plans.

Vibration Estimation using C++

The Analytics API can be used to estimate the frequency of vibrating objects.

The sample metavision_vibration_estimation.cpp shows how to compute:

• a frequency map, that is a per-pixel estimation of the frequency of the received events

• the dominant frequency among all pixels, that is the most common frequency.

This sample also allows computing the dominant frequency in ROIs and checking the frequency at specific locations (i.e. pixel by pixel) in the frequency map.

Expected Output

Metavision Vibration Estimation sample visualizes the frequency map (i.e. frequency estimated per pixel), the dominant frequency in the camera’s FOV, and selected ROI(s) with their frequency(ies):

The sample allows the following interactions:

• selecting an ROI (with mouse click and drag) for which the dominant frequency will be shown next to the ROI

• moving a mouse pointer over vibrating pixels to see frequency pixel by pixel

How to start

First, compile the sample as described in this tutorial.

To start the sample based on the live stream from your camera, run:

Linux

./metavision_vibration_estimation

Windows

metavision_vibration_estimation.exe

To start the sample based on recorded data, provide the full path to a RAW file (here, we use a file from our Sample Recordings):

Linux

./metavision_vibration_estimation -i monitoring_40_50hz.raw

Windows

metavision_vibration_estimation.exe -i monitoring_40_50hz.raw

To check for additional options:

Linux

./metavision_vibration_estimation -h

Windows

metavision_vibration_estimation.exe -h

Code Overview

Pipeline

Metavision Vibration Estimation sample implements the following pipeline:

Frequency Map Async Algorithm Stage

This stage uses the Metavision::FrequencyMapAsyncAlgorithm to generate a frequency map of the vibrating objects from the CD events. The Metavision::FrequencyMapAsyncAlgorithm is asynchronous in the sense that a frequency map is not produced for each input buffer of events but rather at a fixed refresh rate in the camera’s clock. So for each input buffer of events, 0, 1 or N frequency map(s) might be produced.

To feed the Metavision::FrequencyMapAsyncAlgorithm with the input CD events, we specify the consuming callback of the stage using the Metavision::BaseStage::set_consuming_callback() method:

set_consuming_callback([this](const boost::any &data) {
    try {
        auto buffer = boost::any_cast<EventBufferPtr>(data);
        if (buffer->empty())
            return;

        freq_algo_->process_events(buffer->cbegin(), buffer->cend());

    } catch (boost::bad_any_cast &c) { MV_LOG_ERROR() << c.what(); }
});

Then, to retrieve the frequency map, we subscribe to the output callback of the Metavision::FrequencyMapAsyncAlgorithm and send it to the next stages using the Metavision::BaseStage::produce() method:

frequency_map_pool_ = FrequencyMapPool::make_bounded();
freq_algo_->set_output_callback([this](Metavision::timestamp t, FrequencyMap &frequency_map) {
    TimedFrequencyMap timed_frequency_map;
    timed_frequency_map.first  = t;
    timed_frequency_map.second = frequency_map_pool_.acquire();

    // The frequency map is passed via a non constant reference meaning that we are free to swap it to avoid
    // useless copies.
    cv::swap(*timed_frequency_map.second, frequency_map);

    produce(timed_frequency_map);
});

Note

The fact that the frequency map is passed to the callback via a non constant reference by the Metavision::FrequencyMapAsyncAlgorithm allows us cv::swap it and avoid useless copies. This way the Metavision::FrequencyMapAsyncAlgorithm can continue to update the next frequency map while the current one is sent without any copy to the next stages. Apart from the swap, the frequency map is not modified in the callback.

Vibration GUI Stage

This stage is in charge of displaying the frequency map in a way that makes it easy for you to visualize the vibrating objects in the camera’s FOV.

Each input frequency map is converted to an RGB frame using the Metavision::HeatMapFrameGeneratorAlgorithm and a given color map. In addition, the dominant frequency of the scene is computed using the Metavision::DominantValueMapAlgorithm and printed in the RGB frame.

Furthermore, the GUI stage allows you to define ROIs in the frequency map by clicking and dragging. The dominant frequency inside that ROI is then automatically computed.

Finally, you can check the frequency pixel by pixel by pointing the mouse cursor at a specific location in the frequency map.

All the previously described processing is done synchronously, meaning that for each input frequency map, an RGB frame is displayed and all the user interactions are handled. For that reason, everything is done the consuming callback of the stage:

set_consuming_callback([this](const boost::any &data) { display_callback(data); });

And the full display callback:

window_->poll_events();

auto ts_frequency_map = boost::any_cast<TimedFrequencyMap>(data);

const auto &input_ts     = ts_frequency_map.first;
auto &input_freq_map_ptr = ts_frequency_map.second;
if (!input_freq_map_ptr)
    return;

// Generate the heat map
freq_map_display_algo_->generate_bgr_heat_map(*input_freq_map_ptr, display_frame_);

print_dominant_frequency(*input_freq_map_ptr);
print_pixel_frequency(*input_freq_map_ptr);
print_rois(*input_freq_map_ptr);
print_help_message();

window_->show_async(display_frame_, false);
if (window_->should_close())
    this->pipeline().cancel();

The following image shows an example of a frequency map displayed in the GUI stage and in which an ROI has been defined: