Projects‎ > ‎


I applied and was selected for the Internship Program  at Industrial Technology Research Institute, Taiwan. This was my first trip abroad and turned out to be a wonderful experience. Umm, few minor problems - I didn't know Mandarin, my mentor didn't know English, took me 2 months two teach him how to pronounce my name (still he spells it Karmal in all the mails, God knows why). I was kinda vegetarian before going to Taiwan, now I am not. Before been at Intern, I didn't have much background in Computer Vision or Machine Learning. Keeping that in mind, I think I did fairly well during my two month stay.

  • Project: Automatic Cruise Control System (I was contributed to a small part of it)
  • Company: Industrial Technology Research Institute
  • Duration: May 2010 to July 2010
  • Location: Hsinchu, Taiwan
  • Supervisor: Dr. K. H. Cheng, Researcher, Intelligent Mobility Technology Division, Mechanical and System Research Laboratories, ITRI
(founded in 1973, ITRI revolutionized the semiconductors and electronics industry in Taiwan. Here is a link to the Wiki page.)

It turns out that I had signed a Non-Disclosure Agreement at ITRI (again, in Mandarin) and I can't actually reveal what I did in my project. So, I am removing all the details and file links, feel free to contact me for anything. Here is a presentation I gave during my colloq on my internship.

  • To study and understand the camera calibration parameters, functioning of camera calibration toolbox for MATLAB and use it to find the internal and external parameters for a given camera, remove distortions and create Inverse Perspective Transformation for an image
  • To develop an algorithm to correctly identify a pair of tail-lights as candidate vehicle obstacle during the night time or extreme weather conditions
Detecting obstacles is one of the challenges faced in perception for an unmanned vehicle for autonomous navigation. While algorithms for detection of Obstacles and Lane during day time have been demonstrated to work with robustness with respect to varying light conditions and in real-time, detection of vehicles in real time for aiding an autonomous car by giving vehicle collision warning in extreme weather conditions is an emerging research area.

I proposed an algorithm for night time vehicle detection and tracking with inclusion of Support Vector Machine spanning all extreme weather conditions (Autonomous Cruise Control System) (~MATLAB Run Time – 1image/sec) The method can be tested and improved for other climatic conditions (current testing done on the present database of ITRI). The candidate taillights and headlights of vehicles are isolated and Support Vector Classifier is trained with over 400 samples. Results, Conclusion and future scope - discussed (and removed from here).

The first part of my internship (~15 days) were kind of an intro to computer vision and an understanding of what was going on over there. What I did was
  • Camera Parameters Calibration for correction of image distortion and calculation of intrinsic/extrinsic params
  • Bird’s Eye View Transform of an image (preceded by corner detection and image rectification) applied further for LDW and FCW in Vision Based Driving Assistance
In the second part (first part onwards), I was assigned the problem as described in the abstract.

In IIT, we all have to do a 3 credit course EEC410, in which a presentation (and report and training diary) has to be given on the basis of your internship. I, (of course) got an  an A :) and you can download the final presentation here.