santosh's Projects
This repository contains the source code, dataset folder, CMake file and the results of this project. The project is to estimate Time to Collision to objects on the road by fusing the LiDAR and Camera data. Yolo algorithm is used to detect objects, key-point detectors and descriptors are used to track the objects, LiDAR points are used to estimate the distance. For a deep understanding of the project, please refer to the readme file.
A ball is balanced on an inclined plane through PID control with a servo motor as an actuator.
This project code recognizes and visualizes 2D features in Camera images. The images are captured in real-time traffic scenario. The project gives options for key-point detectors and descriptors. The repository can be cloned and used to detect 2d features in any kind of camera image.
This repository contains the .py solution file for the challange given by Kopernikus Automotive for the Perception Engineer interivew process. This repository also contain the pdf file for theoretical answers for questions relating to the dataset and the challenge.
This repo contains the source code, make file and the build file of the project. The project is to detect 3D objects in a continuous stream of LiDAR point clouds. The code is functional to any type of OpenCV point cloud point types. Feel free to clone and execute the repo.
This repository contains practice problems of various C++ data structures and algorithms.
MSP430FR5969 is an ultra low power launchpad. The launch pad consits of Timers and UART serial Communication. This repositiory consists of tasks focused on manipulating the I/O operations, LEDs, Interrupts, Timers, Memory and Energy dissipation..
Open Source Computer Vision Library
A pick and place mechanism is developed and deployed on to a Lego-ev3 manipulator robot
This is a LiDAR point cloud processing software. It has features to add synthetic noise to a given point cloud, remove noise and detect objects.
In this project, an FMCW RADAR is configured. The transmite and recieved are processed uisng FFT and CFAR to determine the range and velocity of the obstacle"
Config files for my GitHub profile.
Attention transformer architecture is the start of the art in natural language processing. It is proven to have the best attention distance over the data of a sample. Recent literature on first ever attempt to implement pure transformer on image classification claims that, when pre-trained on sufficiently large datasets like ImageNet-21k and JFT-300M, transformer out performs the state of the art CNNs. However, google has the exclusive ownership to JFT-300M dataset and even if it were open sourced, these datasets require either very expensive GPUs or large amount of training hours. Hence, in this project we have studied transformerโs relevance without pre-training. Its performance on the simple MNIST dataset is similar to CNNs (99.39%), while, on much complex dataset CIFAR-10 it is above average (71.2%) but with significantly less parameters.