Computer Vision Exercises

Solutions for the computer vision tests

0) Dependencies

OpenCV -v=4.0.0 tkinter -v=8.6

1) Optical flow and tracking

The script Optical_flow.py runs the Optical Fluck computation through the Lucas Kanade method. Steps:

1. In the terminal, on the root folder, run python Optical flow.py. The webcam will take a picture.
2. Use the mouse to mark a rectangle with the object to be traced. Click once to start, then once to finish
3. Press 'c'. The object will then bem traced.
4. Press Ctrl+c in the terminal to stop.

2) Image Stitching

The script Image_stitching.py does o stitching. Steps:

1. In the terminal, on the root folder, run python Image_stitching.py.
2. Select in the window the images that you want to stitch and press Open.
3. The stitched image should appear in the screen and then get saved as output.png

3) Object Detection

The script Object_detection.py runs YOLOv3-tiny through the Deep Neural Network module from OpenCV. YOLO has a set of architectures available, all trained in the COCO Dataset trainval, with a performance up to 57.9 mean average precision for the YOLOv3-608, and of 60.6 for the YOLOv3-spp, in COCO Dataset test-dev, both being able to run at 20 FPS in a Pascal Titan X GPU. Due to the limited resources of the computer where these examples were developed, YOLOv3-tiny was chosen.

A YOLOv3-tiny reaches a performance of 33.1 mAP and may run up to 220 FPS. The computer where this scipt was developed managed to run at 4 FPS, and has the following settings:

• Intel Core i3-5005 @ 2GHz
• 12Gb RAM memory
• On-board GPU (Not used in this script script)

1. On terminal, in the root folder, run python Object_detection.py. A webcam window will appear. Try showing objects such as a phone or a mug in front of the camera
2. To finish, press 'c'. The terminal will then print a simple report with the FPS rate during the script runtime.