Baseline of this code is strongly this official repository. Original README is in the followed link. However, this code has been a problem of dynamic allocation in scheduling and threading during parsing real-time images. Therefore, edited author fixes below issues: from v1 to v7. In addition, the edited author adds tracking function to YOLO by Linear Kalman Filter.
Original Author: Marko Bjelonic, marko.bjelonic@mavt.ethz.ch
Edited Author : Byung-KwanLee (leebk@kaist.ac.kr)
- Divide two cpp function for yolo and lane
- Completely solve memory issue considering sharing resource using boost lock
- Subscribe lane detection information lane_msgs/lane_array.h, lane_msgs/lane.h
- Classify objects which lane they are in, on the top view, without vanishing point
- Remove boundingbox (made by developer in original ros yolo), then simplify publish procedure
- Chage protocol of darknet_ros_msgs, more specific
- Eigen data type change, MatrixXd -> MatrixXf, VectorXd -> VectorXf (double -> float)
- Manage Dynamic Memory Allocation and Remove
- Critical Issue for real-time
- Check CPU Resource
- Should be analyzing resource of tracking_thread function made by LBK
- Extra Memory Issue (V5*)
- Synchronization with darknet free code and tracking thread
- Keeping prediction without detection
- Add boundingbox message to distacne from the object
- Adding Kalman Filter on the Detection thread
- Transition and Updating Error Covariance
- Dynamic Allocation and Remove Completely
- Transfer variables on each member function with no Sync Error
- Publishing Object ID with Bounding Box info for sensor fusion
- Matching tech by class
- Completely generate multi thread for tracking 30HZ
- CompressedImage subscriber (Image topic inavailable, camera callback function)
- solution for exit code -11 (gdb : Segmentation error)
- no visualization imageview
- publish Image and CompressedImage message