72 Add YOLO object detection #167
Conversation
jackguo380
requested review from
divyabudihal,
nancyhong123,
hannahvsawiuk and
zhaoshengEE
|
Still WIP, not fully done yet. Still TODO:
|
jackguo380
force-pushed
the
72_yolo
branch
3 times, most recently
from
7765b26 to
aecd39b
Compare
| #include <functional> | ||
| #include <deque> | ||
|
|
||
| class BoundingBoxTransformer |
| , tf_timeout_(tf_timeout) | ||
| {} | ||
|
|
||
| void handleCameraImage(const sensor_msgs::ImageConstPtr& image_msg, |
There was a problem hiding this comment.
we use capital letters for all words in the function names for the rest of the codebase (except for mapviz plugins)
| tfScalar radius; | ||
| tfScalar height; | ||
|
|
||
| BoundingCylinder(std::string objectClass, int id, ros::Time creationTime, |
There was a problem hiding this comment.
Do you ever use the default constructor? If not can you delete it?
| @@ -0,0 +1,46 @@ | |||
| cmake_minimum_required(VERSION 2.8.3) | |||
| project(propbot_object_detection) | |||
There was a problem hiding this comment.
rename to propbot_perception, or put this in a propbot_perception directory and name it propbot_object_detection
| @@ -0,0 +1,128 @@ | |||
| # Propbot Object Detection | |||
There was a problem hiding this comment.
Great README! I won't comment on this for now, once you put it into the design doc, we can fix things up and open another pr to change the readme if necessary
| //ROS_INFO_STREAM("Find Angles: " << angle_left << ", " << angle_right << | ||
| // " from " << laser->angle_min << ", " << laser->angle_max); | ||
|
|
||
| float left_from_min = angle_left - laser->angle_min; |
There was a problem hiding this comment.
would be also helpful to provide a brief description above each function
| } | ||
|
|
||
| if (groups.size() >= 2) { | ||
| //if (groups.size() >= 3) { |
| last_image_time_ = image_msg->header.stamp; | ||
|
|
||
| // Update our camera model from this image | ||
| pinhole_camera_.fromCameraInfo(cam_info_msg); |
There was a problem hiding this comment.
the CameraInfo message shouldn't change during operation. It would be sufficient for you to just load it in once. The CameraInfo message depends on some calibrations that are done before operation.
| const darknet_ros_msgs::BoundingBoxesConstPtr& bb_msg) | ||
| { | ||
| if (!pinhole_camera_.initialized()) { | ||
| ROS_WARN_STREAM_THROTTLE(1, "Waiting for Camera Initialization"); |
There was a problem hiding this comment.
use ROS_WARN instead to keep things consistent with the rest of our code
|
|
||
| // Find a 3D point (in the laser frame) corresponding to the pixel | ||
| // this is done by extending the ray from camera to pixel out to ray length | ||
| auto projectPt = [&](int x, int y) -> tf::Point { |
There was a problem hiding this comment.
nice, why a lambda instead of a private function tho? bc nothing else uses it?
|
|
||
| <!--- Launch Object Detection --> | ||
| <include file="$(find propbot_object_detection)/launch/detect_objects.launch" /> |
There was a problem hiding this comment.
add a setting "track_dynamic_objects"?
Adds YOLO object detection into simulation.
The integration of YOLO into ROS is done by darknet_ros. Unfortunately I had to fix a bunch of bugs with OpenCV in my own fork of it as well as fork darknet to fix some stuff as well.