Object Tracking with Dynamic Detection in Sliding Camera

Team Members: Youjia Li, Eddie Wu

 

Structure

objectTracking

objectTracking.py contains all the essential parts of optical flow operation, including getFeatures,estimateAllTranslation and applyGeometricTransformation.py.

mrcnn_detect

mrcnn_detect.py contains all the essential parts of object detection and instance segmentation, making use of published Mask R-CNN architecture, and it is used in objectTracking.py

create_output_video

create_output_video.py runs everything together through calling of objectTracking function.

input_videos

This directory contains all the input videos we are testing.

output_videos

This directory contains all the resulting output videos we are generating for this project.

 

Development Environment & Usage

1. Unzip the zipped file
2. Change directory to the Final_Project folder using cd in command line
3. (Optional) Create a virtual environment using virtualenv venv --python=python3.6
4. Run pip install -r requirements.txt. If using CPU only, install CPU-Optimized Tensorflow, and make sure Python version is 3.6 or less (if virtual environment was not created per last step, as 3.7 doesn't fully support Tensorflow yet)
5. Modify the rawVideo variable in create_output_video.py to be the name of input video file
6. Run create_output_video.py. If mask_rcnn_coco.h5 (pretrained Mask R-CNN weights) is not downloaded yet, it will first be downloaded to the current directory and then used for the application.