- This is a visual SLAM project and not object segmentation
- 200m
- Scene with motion
- Scene with animals/humans
- Dense vegetation
- Draw a bounding box around the matched feature
- Show how the number of matched features decrease with gaussian noise
- Extract identified region from the scene
- Prepare Promotion materials for the website
- Join frames as video 'ffmpeg -framerate 25 -i %d.jpg -c:v libx264 -vf "fps=25,format=yuv420p" common_video.mp4'
- Compress and send the region at around (rate 37.5 kbps) [Video/Image transmission/Compression] i.e streaming technologies
- Image Compression tool
- Data Transfer tool
- Lora or 3G
- See Lora Airtime calculator https://docs.google.com/spreadsheets/d/1QvcKsGeTTPpr9icj4XkKXq4r2zTc2j0gsHLrnplzM3I/edit#gid=0
- Data rates https://blog.dbrgn.ch/2017/6/23/lorawan-data-rates/
- https://stackoverflow.com/questions/33027942/opencv-convert-image-to-bytes-and-back
- Check out Zlib compression http://www.zlib.net/
- Check out FFMpeg http://ffmpeg.org
- http://cpham.perso.univ-pau.fr/WSN-MODEL/tool-html/imagesensor.html
-
CV Books <-> Implementation
Image comparison (stored target image vs real-time target image)- Gather dataset
- [ x MSE and Structural Similarity https://www.pyimagesearch.com/2014/09/15/python-compare-two-images/
- Structural Similarity deep dive https://www.pyimagesearch.com/2017/06/19/image-difference-with-opencv-and-python/
- Fourier transforms
- Difference between histograms
- Keypoint (SIFT) matching
- PCA differences
- try reducing the dimensions and using SSIM on it
- Track moving objects