Plant health detection using NDVI and ORB-SLAM
This project focused on determination of plant health by calculating a parameter called ‘Normalized Difference Vegetation Index’ using the captured IR images from a sensing & navigation unit. The project was implemented using a NoIR (No-Infrared) camera for determining plant health and a Stereo Camera to perform ORB-SLAM (Oriented Fast and Brief), all connected to a Raspberry Pi. The aim of the project was to create a system for tracking health conditions of plants in indoor environments such as a greenhouse, lab, etc. The initial trial was performed in Northeastern University’s campus where both live and dead plants were scanned to test our NDVI results.
Panoramic Image Stitching
During this lab, we employed the CALTECH camera toolbox to calibrate the
camera on our phone to obtain undistorted images to be used for Panoramic Photo Stitching. To create a Panorama, a set of images are required with overlapping field of view and then stitched together using these three steps: Image calibration, Image registration and Image blending. We implemented the Harris corner detection algorithm for feature detection and used the Image mosaicing algorithm(MATLAB Photomosaicing) to get a panoramic image.
Sensor Fusion for Dead Reckoning
This lab provided us the insight of sensor performance and how the accuracy can be increased using sensor fusion. We collected IMU and GPS data through sensors installed in Northeastern's Autonomous car while driving in Boston area. We performed sensor fusion utilizing Magnetometer, Gyroscope and Accelerometer data for Dead Reckoning and developed Complimentary filter utilizing MATLAB to remove bias and drift from magnetometer and Gyroscope respectively. We analyzed accelerometer data to remove bias and noises for estimating velocity and displacement and formulated the true trajectory using GPS and compared it with estimated trajectory obtained using Dead Reckoning.
Imu sensor & noise characteristics
Our objective was to gain a better understanding of the various noise characteristics present in the accelerometer and gyroscope data by utilizing the Allan Variance analysis method. For this, we gathered and evaluated data from the IMU-sensor, specifically the VectorNAV-100. This analysis was conducted on a 5-hour stationary dataset.
Additionally, we created a model in MATLAB to replicate an IMU with these noise characteristics to gain insight into how they could impact the sensor selection process for different applications. Lastly, we evaluated the performance of a stationary inertial sensor by employing various statistical tools to visualize the error distribution. We successfully found different noise present in IMU which are White noise, Pink noise and Red noise and characterized by Angle Random Walk, Bias Instability and Rate Random Walk respectively.
RTk-gnss
RTK-GNSS provides better accuracy over GPS and same was studied through collecting data under different environment conditions (Open space or Occluded space) with stationary and moving conditions. To achieve this, we used two two GPS receivers, one receiver was stationary at a known location called the
base station, and the other was mobile or stationary at an unknown
location called the rover. The base station sends correction data to the
rover through a radio link, which is used to calculate its precise
position by combining the signals it receives from the satellites and
the data from the base station. we analyzed the RTK-GNSS data using least square error to confirm achieved accuracy up to centimeter level.
gps
In this project, We have programmed device driver in ROS using Python for parsing GPS data and converting to UTM coordinates. In order to study the GPS navigation, two types of data were collected, one at a stationary location and other while walking. The effect of several factors on the accuracy were studied using various statistical methods.