My Projects
Vehicle Propagation Model.
Consider a wide open sea where multiple sensors on different platforms are recieving spuradic detections of an unknown vessel on the water. The goal of this method was to utilize a discrete approach, which compared to a particle based approach would: 1. Be repeatable, so the results are guaranteed to be the same for the same set of inputs. 2. A discrete approach would allow for all data to be transfered from one platform to another. This allows multiple searching vehicles to compare maps and gather data from each other.
A few assumptions from this project are:
1. The target exists.
2. The target continued in a linear trajectory at the last known velocity/heading without turning.
The underlying method was to use the estimated heading/speed and its variance to generate a kernel to perform convolusions over the probability grid. This approach was inspired by a linearized hidden markov model.
However, certain assumptions and constraints to the problem required discretization. The discretization resulted in numerical artifacts, yeilding poor results. However, it was an interesting learning experience.
A video demo may be seen here.