Chirag "CJ" Sirigere
Personal Projects
VIN-TRAnsfer Project (VIN-TRAP)
After a hit-and-run accident with multiple cars, it becomes difficult to apprehend the suspects involved. This project was designed to be a proof-of-concept where devices can transmit information about a suspect's car to nearby cars in the crash site. After GPS coordinates of the crash have been transmitted, law enforcement and insurance agencies can get notified of the suspect's location through a continuous transmission until apprehension.
Rocket Motion Tracker
With a directional pointer equipped with a camera at the ground station and video recognition, a rocket's motion can be followed as it ascends and descends using a flight computer's readings and the angle of the directional pointer through a feedback system. Video recognition is employed using a Kalman Filter to verify the rocket's descent and take-off.
_edited.jpg)
Sunset Detector
This project entails converting 3200 different RGB images (1600 from the train, 1000 from the test, and 600 from the validation datasets) to LST color space, segmenting the images into 49 sections, and processing the mean and standard deviation through an SVM. The baseline LST-SVM has an accuracy of 89.6% on the test dataset but is overfitted based on the validation dataset. In a second part of the project, two CNNs are used to classify the images, one from squeezenet for transfer learning and one from scratch in Keras. Both CNNs performed better than the LST-SVM with accuracies of 95% and marginal FPR versus TPR ROC curves (nearly 0 to 0.9).
Level 2 HPR Certification
This rocket is being launched on a J760 White Thunder and reaching an approximate apogee of 1550 meters with a maximum velocity of 250 meters per second!
Concrete Rocket
Yes, this crayon is the full-scale, 13-foot long, 2022 NASA USLI rocket, made out of concrete. Total weight: At least 400 pounds! Version 2 will be much smaller, lighter, and will be able to fly after proper safety procedures and testing!
16-bit μProcessor
This report describes the design specifications and process taken while creating this stack-implemented processor. Everything from the instruction set architecture to RTL instructions and sample instruction-level programs.
Playing Card Identifier
Our proposed model attempts to identify cards in natural environments through image processing and classification. To accomplish this, our model uses a region-based convolutional neural network (R-CNN) to crop important card identification regions from natural images containing cards. Then, one of three different classifiers identify the rank and suit of the card found. Our results proved promising in the identification of cards, with the highest of the three classification models producing an accuracy of 77%.
Fruit Classification Using Morphological Operations on Masks
Recognizing objects in an image can be very useful from securities in public areas to inventory management in store. We developed a system that can count the number of apples, bananas and oranges in different images by applying filters and performing morphological operations on the masks of each fruit. The system will recognize the fruit most of the time, but when the same fruits are stacking together, the system would not be able to correctly count the number of each fruit.
Remote Controlled Telescope
This monster of a robotic telescope prototype is capable of locating "constellations" placed around a room via a remote control.
