Status Report 5(10.19-10.26)

Team

  1. The tracking effect does not have the intended effect. We are looking into other methods to track our trash.
  2. We are on schedule at the moment.

Ruohai Ge

  1. I implemented the Kalman trajectory prediction in 3D
  2. Integrate my trajectory prediction code with Zheng’s vision pipeline by reorganizing it to a class
  3. Optimized the trajectory prediction code by reading less tracked points
  5. Help Xingsheng with the Bluetooth module connection and pointed out that TX and RX have been wrongly labeled
  6. Started to research Tracking algorithm together wit Zheng
  7. All popular methods do not work, the video shows how boosting and KLT methods work
  8. The following showed the result of Boosting Tracking and KLT tracking, the result is unsatisfying.
  11. I am on Schedule and have free time to help other teammates
  12. Next week’s deliverable
    1. Integrate with Zheng’s vision pipeline
    2. Test the pipeline

Xingsheng Wang

  1. This week the controller board has arrived with STM32 integrated. The previous STM32 board is no longer needed.
  2. I have implemented the bluetooth transmission module for the board this week and started testing. The board was unable to receive data at first and it turns out that the TX and RX pins have been wrongly labeled on the board such that the connections are not correct. Now the board is able to receive data via bluetooth from the computer.
  3. While the bluetooth module works, the transmission speed is not stable. Sometimes the board can receive and print the data immediately and sometimes there are long delays with errors of no data in the queue. I am looking into adding another buffer to see if the problem is solved.
  4. Next week I will test if I can get the platform moving through commands sent from the computer via bluetooth.
  5. I am currently on schedule.

Zheng Xu

  1. Reorganized all preexisting code into class structures for better modification and documentation purposes.
  2. Integrated vision pipeline with Ruohai’s trajectory estimation module. Still need time to modify both parts for better prediction. The current code uses all observations (including noise that are not trash) for prediction, and the outcome is not comprehensible.
  3. Tested with different configurations of object tracking and object detection combination. The current most promising combination is GMG background removal paired with CSRT tracking. The combination does not work when the tracking window is too small, and this phenomenon is quite bizarre.
  4. Tuned parameters to increase performance of background removal. After doing 1 iteration of blurring and 3 iterations of dilation the result is closest to the ground truth.
  5. I am currently a little behind the schedule due to the unpredictability of the object tracking modules. I will catch up by focusing on noise removal and migrate consistent tracking to Ruohai.
  6. Next week’s promise
    1. Modify the trajectory estimation pipeline to get a preliminary result.
    2. Set up working examples for the MVP demo.