Employing Low-Cost Sensors for High Precision tasks using HPC

Extract

The main aim of this project was to somehow make use of modern day, high end High Performance Computers to increase the performance and efficiency of low cost sensory system.

In particular RANSAC sequential algorithm was point of interest which in our application is used as a Line Fitting Algorithm to detect a wall beside an autonomous Robot. Throughout the project work various optimization methods were tried in order to bring the efficiency and performance level of low cost sensors to a level where they can match the levels of some high quality, expensive sensors.

A dummy model of a Low Cost Sensor and High Cost Sensor was considered and both were employed to detect a wall with maximum precision possible. Only a couple of seconds were taken by the High Quality, High Cost sensory system while Low Quality, Low Cost sensory system took ten’s of seconds for computing the same wall with equal precision.

This timing factor was targeted using High Performance Computers where various parallel programming methods were tried in order to bring down the computation time. Astounding results were obtained where most effectively CUDA implementation reported a 100x speed up. Other implementations also gave more than satisfactory results where OpenMP reported an amazing 10x speedup, and Matlab PCT and MPI version reported around 7x speedup. Optimization Flags were also included in experimenting which gave a 3x speedup.

This Project came out with flying colors and proved that performance of Low Cost sensors can be improved significantly just by employing effective parallel computing methods on the processing part of the sensory system and hence by using a High Performance Computation model.