Status Report 3(10.6-10.12)

Team

1. Prepared presentation notes that include more details in the method comparison and testing metric.
2. Wrote design review reports.

Most Significant risks and contingency plans

1. The mobile base does not contain a controller, we need to buy another one. The one on the amazon prime ships too late, so we need to think about some ways to find other suitable controllers or contact directly to the Chinese supplier for faster shipments.

Schedule Update

A little behind schedule this week, need to make up and speed up during the following weeks.

Ruohai Ge

1. Build the mobile base together with Zheng
2. Finished parts of the design review report:
   1. Design requirements
   2. Design trade studies – Kalman filter, dead reckoning, PID
   3. System description – Kalman filter, dead reckoning, PID
3. Tried to estimate trajectory using the Kalman filter
   1. A lot of noises happen in the first 3-4 frames, develop a way to get rid of them for the first few frames.
   2. Write the pipeline to predict the trajectory based on the predicted point and updated it along with the video
   3. The result is not very satisfying. Only a small range can be predicted. When the range is bigger, the trajectory behaves very wierd
   4. Some trajectory even goes in the wrong direction. The last picture shows a complete wrong trajectoy
   5. The model need to be updated or some new method need to be deployed.
   6. 
4. Schedule status
   1. A little behind the schedule. Spend too much time on the design report and encounter too many problems for trajectory estimation
5. Deliverable for next weeks
   1. Estimate the trajectory using the Kalman filter on the same video. Make sure a complete and reasonable trajectory can be seen.
   2. Predict landing points based on altitude

Xingsheng Wang

1.  Purchased the battery power supply for the STM32 board and the motors. The power supply has both 5V USB and 12V output and a capacity of 3000mAh. Timing experiments will be conducted later.
2. Reviewed example code provided by the mobile base manufacturer.
3. Finished parts of the design review report:
   1. Abstract
   2. Index terms
   3. Introduction
   4. Design trade studies
   5. System description
   6. Project management
4. Schedule status: This week I am behind schedule because of heavy course work and interviews. Next week I will be back on schedule as other course works lighten and no other interviews pending. There is a mid-semester break next week which I can use to do a lot of 18500 work.

Zheng Xu

1. Accomplishments
   1. Capture moving object in the frame.
      • Continuing from the background subtraction work from last week. I first added connected component analysis to get the continuous areas and their stats like position and area.
      • Then the algorithm choose a frame where the object in question is present in the background removal frame, and use the bounding box generated from connected component to initiate a KCF tracker.
      • The KCF tracker will keep tracking the object until the the object is no longer seen.
      • As is shown on the video, the ball can be tracked, however, the tracking frame is not always present, and it can have false positive tracking. To deal with these problems, I have a couple of solutions listed in the next week’s promises.
   2. Implemented Fast Moving Object Detection(FMO)
      • https://arxiv.org/pdf/1611.07889.pdf
      • Because there are specific parameters that I haven’t been able to tune, the algorithm does not generate any significant result yet.
   3. Tested overall speed and frame rate.
      • The frame rate is not stable, and it could dip to around 10 fps. This is due to redundant operations still existing in the code. The next version of the program will not run tracking and detection at the same time. (Tracking should happen after detection, not at the same time).
2. Schedule situation: on schedule
3. Promises for next week:
   1. Tune parameters for FMO detection and yield a satisfactory result.
   2. Improve both runtime and performance of object tracking and detection by removing redundant operations, add general shape detection (if the bounding box of the shape is long or square) and experiment with different frame sizes.