Computer Vision

I wrote Matlab code to identify and track cars in this video clip provided by MathWorks. The operations I used were filtering, background subtraction, thresholding, morphological dilation and erosion, and region property extraction. The bounding boxes I’ve added to the video indicate where in each frame the cars have been located by my algorithm and the estimated size of each car.

I’ve also used machine learning algorithms, such as ACF and Yolo to detect and track cars in this video. While using these specialized tools can give you a leg up in algorithm development, they only recognize certain types of common objects and the images (or frames) usually still require some pre-processing. These packaged algorithms, particularly those based on deep learning, also tend to be more computational expensive.

Knowing how to segment out objects and extract region properties through image processing is an essential skill for edge deployment applications, as well as for identifying objects not already familiar to available machine learning algorithms.