You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello everyone. I've been poking around the openvin's codebase and i've done some diagnostics at runtime through the debug mode. And i've realized that the feature tracking is by far the most time consuming part of the calculations ,which is fairly normal, especially when you consider high resolution images. I was wondering if it wouldn't be interesting to add an external tracker concept, especially for gpu and vpu enabled devices. And then instead of a full image we could provide a vector of tracked points directly as input. We could extend the CameraData class like this :
struct TrackedPoint
{
uint32_t age = 0 ; //Number of frame we tracked for
float harrisScore = 0.f ; // Score of the tracked point
uint32_t id = 0 ; // Persistent id to track the point
cv::Point2i position ; // Position in image of the tracked point
float trackingError = 0.f ; // tracking error
};
/**
* @brief Struct for a collection of camera measurements.
*
* For each image we have a camera id and timestamp that it occured at.
* If there are multiple cameras we will treat it as pair-wise stereo tracking.
*/
struct CameraData {
/// Timestamp of the reading
double timestamp;
/// Camera ids for each of the images collected
std::vector<int> sensor_ids;
/// Raw image we have collected for each camera
std::vector<cv::Mat> images;
/// Image sizes in case we decide to only work with trackers and not full images.
std::vector<cv::Size> image_sizes;
/// Eventual tracking points revieved. These are tracked point from last frames to this frame.
std::vector<std::vector<TrackedPoint>> tracked_points ;
/// Tracking masks for each camera we have
std::vector<cv::Mat> masks;
/// Sort function to allow for using of STL containers
bool operator<(const CameraData &other) const {
if (timestamp == other.timestamp) {
int id = *std::min_element(sensor_ids.begin(), sensor_ids.end());
int id_other = *std::min_element(other.sensor_ids.begin(), other.sensor_ids.end());
return id < id_other;
} else {
return timestamp < other.timestamp;
}
}
};
I was wondering how to create a TrackExternal class which obey to the TrackBase schemas to be able to do this.
Thanks a lot.
The text was updated successfully, but these errors were encountered:
Hello everyone. I've been poking around the openvin's codebase and i've done some diagnostics at runtime through the debug mode. And i've realized that the feature tracking is by far the most time consuming part of the calculations ,which is fairly normal, especially when you consider high resolution images. I was wondering if it wouldn't be interesting to add an external tracker concept, especially for gpu and vpu enabled devices. And then instead of a full image we could provide a vector of tracked points directly as input. We could extend the CameraData class like this :
I was wondering how to create a TrackExternal class which obey to the TrackBase schemas to be able to do this.
Thanks a lot.
The text was updated successfully, but these errors were encountered: