Image processing for video recognition is like below.
1. The process of image processing for landing inducement
1) Assume that Drone was moved above the Heliport
2) Shooting the land with below camera
3) Checking exist of Heliport using Feature Matching from recorded video
4) Recongnition of Heliport Corner and center point of Helipot from the video
5) Deliver the information of Vector about center of Heliport to the Drone FC.
6) Repetitive moving Drone to the center of Heliport in the sky
7) Vertial landing when the Drone position is same with ceter point of Heliport
8) Landing sucess and charging start
2. Recongnition of Heliport using Feature Matching
1) The result of recognition using Feature Matching
[Recognizable 'eraser' image]
[Recognition is started when eraser on the table.]
[The result of recognition - All fifty two matching points were recognized and we can know there is eraser]
There is a delay of 10 seconds is consumed so improving the speed is absolutely necessary.
3. Heliport recognition
1) Heliport Prototype Image
[Heliport original pattern image]
2) Heliport recognition test
[Real Heliport Target image to be recognized]
[The result of recognition - The recognition is well done. All fifty matching points were recognized]
4. Calculation of the center point of Heliport
[The result of finding Helipot Centroid - Green points is the result of calculation of the Moments,
파란색 점은 Contour & Bounding Rect로 계산한 중심점]
Moment로 계산한 결과 보다는 Contour & Bounding Rect 방식으로 계산한 결과가 더 정확한 것으로 판단되어 Contour & Bounding Rect 방식으로 결정하였다.
이제 영상처리 모듈을 드론 비행 모듈과 Integration 잘 하고, 영상처리를 통해 얻은 정보로 드론을 잘 컨트롤 하기만 하면 무사 착륙에 성공할 수 있을 것 같다.
5. Testing in the real environment
We tested it on the real environment.
1) Test1 - Recognition in the sky
There are a few wrong recognition points but finding Heliport with Matching more than 30.
Finding Centroids - Checking center of Heliport
2) Test2 - Recognition in the sky
Maching ratio increased when the closely 조금 더 가까이에서 Heliport 찾기 시도 성공~ Matching율이 높아졌다.
Finding center of Heliport has no problem.
It's a Prototype but Heliport size is a little small so we can recognize more easily if there is more large Heliport. Then recognition will be done more easily.