Our purpose is to devise an algorithm for an Israeli Navy USV that will enable it to autonomously detect potential obstacles along its path.
Abstract
Our purpose is to devise an algorithm for an Israeli Navy USV that will enable it to autonomously detect potential obstacles along its path. 4 Cameras are installed on the USV therefore the obtained video is panoramic. This part main goal is to detect potential obstacles using active contours and cost functions.
Flowchart
Alignment
- Gives a reliable frame of reference in naval videos
- Narrows down the area of search for objects
- Simplifies object tracking
- Divide the video into segments
Object Detection
- Active contours method
- Cost functions
- Initial conditions: object search by scanning
- Initial conditions: basic object tracking
Active contours
- We define “object” as a region in a given picture that has different properties than its surroundings
- Our goal is to find a contour that separates this region from the background
Cost functions
- Balloon – minimize the weighted area inside the contour.
- Chan-Vese/Max-Lloyd – minimal variance of gray levels inside and outside of the contour
- Align – aligns the contour so it is perpendicular to the gradient of the picture
- Bhattacharyya – minimizes the correlation between the histograms inside and outside of the contour
Object search by scanning
- Divide the area underneath the horizon into a number of blocks
- The boundary of each block defines a contour
- Apply the Active Contour method for each contour
Results
Conclusion
- We have successfully realized a solution to the problem of object detection in naval environment
- Active contours is a viable method for purposes of object detection in both pictures and videos
- There is still much to do in the subjects of object tracking and preliminary processing of naval environment videos
Future Work
- More sophisticated tracking algorithms
- Multi-object solutions; combining object scanning and tracking
- Preliminary image processing; pyramid, noise filtering
- Introduction of distance and perspective considerations
- Research and trial of new cost functions, modification and improvement of current cost functions
- Research of parameters
- Further improvement of the horizon detection algorithm