Unsupervised abnormality detection by using intelligent and heterogeneous autonomous systems( IEEE Signal Processing Cup )

the competition required to do Unsupervised abnormality detection by using intelligent and heterogeneous autonomous systems where the goal was to detect abnormalities in the behaviour of the ground and aerial systems based on embedded sensor data in real-time.

Autonomous systems experience failure when the sensors record abnormal data which leads to the crashing of the decision making module if it is not capable of detecting abnormalities. The design of abnormality detection models becomes an important step in preventing the autonomous systems from malfunctioning. The organisers of SP cup have provided the dataset consisting of data from various sensors mounted on a drone. After identifying the key challenges like limited training data, non-uniformity in sampling of data from different sensors, low FPS of camera, absence of ground truth annotations, we have used isolation forest method, CNN based timeforecasting method and ARMA model based timeforecasting method to detect the abnormality. We compare these methods using the performance metric of Excess Mass Curves

We were 6th internationally in this competition !