Object detection and navigation with Visual Camera?

asked 2016-05-22 22:56:52 -0600

astronaut gravatar image

updated 2016-05-26 20:40:42 -0600


I have just ventured into computer vision and trying to demystify various intricacies of it. I am trying to detect object in outdoor environment with Visual Camera and navigate a mobile robot to the center of the object.

I would like to use OpenCV (with C++) for object detection and ROS with Ubuntu for navigation (path planing). So as far as understand the first step is extract features (feature detection) method using OpenCV. I think in my case I need just a contour detection . What methods can be use inside OpenCV for this first step? Is it HOG detection method appropriate for this problem or do I need to use SURF or SIFT feature detection methods? After feature detection I understand need to apply classification methods in order to recognize the door , true?

When object detection successfully applied need to navigate the mobile robot to the center of the Aircraft door. This can be done with Kalman filter inside OpenCV or can use the ROS that supports OpenCV and features like SIFT or SURF for navigation and path planning to the detected object?

Also any code or out of the box methods would be great of help. Thanks

edit retag flag offensive close merge delete


If you can, I think that you should also consider using a depth (RGB-D) camera like the Kinect instead of regular camera.

Eduardo gravatar imageEduardo ( 2016-05-23 14:45:29 -0600 )edit

I was thinking about but my consider is if the Kinect will work under any outdoors conditions like rain, sun, dark... . Do you think that would be a problem? How about any out of the box algorithms for feature and contour detection? Any help on that?

astronaut gravatar imageastronaut ( 2016-05-23 19:48:03 -0600 )edit

I don't think Kinect will work outdoor as it uses infrared projector but you can use stereo camera instead. In my opinion, feature-based methods will not work. If you are new to the computer vision field, you should start with basic methods like contour detection (Canny edge detector), line extraction, rectangle detection or detect the corners in the image and try to get big rectangle from corners. Also you can look at the litterature, for example: Visual Door Detection Integrating Appearance and Shape Cues.

Eduardo gravatar imageEduardo ( 2016-05-24 16:54:57 -0600 )edit

I done contour detection using Opencv tutorials. Works ok. How can I a expand now more this algorithm to line extraction and rectangle detection or corner detection and try to get big rectangle from the corners? Im using C++ and opencv

astronaut gravatar imageastronaut ( 2016-05-24 23:57:00 -0600 )edit