August – December 2019

• I learnt about the concepts and applications of computer vision in the problem sets for this class.

• Specifically, I learnt about:

  • application of hough transforms for traffic sign detection,
  • optic flow for small motion using Hierarchial Lucas Kanade algorithm,
  • image transformation and homography for marker tracking,
  • pedestrian detection and tracking using Kalman and Particle Filter and,
  • face detection using Viola Jones classifier.

Final Project

• Applied transfer learning on pretrained VGG16 to correctly classify digits in image and video.
• Detected digits using Non Maximal Suppression and sliding window technique.

• Trained on Street View House Numbers dataset to obtain test accuracy of 96%.

• The report includes the output images and link to video running the classifier on unseen images.
• The presentation video describes the workflow on how this project was completed.

Tools used: Python, OpenCV, Tensorflow, Keras, NumPy

Figure 1:Traffic Sign detection using Hough Transform from OpenCV.Parameter tuning was essential to be able to detect the traffic sign shapes correctly.

Figure 2:A Particle Filter was implemented from scratch in Python.This particle filter was run on a video.The video recorded students walking on the TUD university campus.The bounding box indicates the location of the student and the points indicate the particles clustered at the estimated position.