Obstacle Avoidance using LiDAR Sensors and Artificial Intelligence

Abstract: In advanced driver assistance systems, avoiding obstacles is an important feature focused on providing correct, timed and reliable alerts prior to an impending collision (with objects, vehicles, pedestrians, etc.). In order to address the design and evaluation of obstacle detection in a cyber-physical transportation system, the obstacle identification library has been developed and implemented. Library is then built into a system for co-simulation that facilitates the interface between MATLAB / Simulink and SCANeR applications. Next the modelling and simulation of virtual on chip light detection and range sensors in a cyber-physical system is explored for traffic scenarios.  In SCANeR, the cyber-physical device is planned and enforced. Secondly, using a visual sensory library supplied by SCANeR, three unique AI - based approaches for obstacle detection libraries are also developed and implemented.  A MLP - NN, multi-layer perceptron neural network; a SOM, map of self-organization; and an SVM, support vector machine are the three methods for obstacle detection and identification in the library. Finally, a contrast is made between these approaches under varying weather conditions, with very good findings in terms of precision. Using the multi-layer perceptron in clear skies and low visibility conditions, the support vector machine in the weather of rain and the self-organized map in the weather of snow are the best results obtained. Keywords: Loop - sensors; simulation framework; cyber - physical system; on - chip LiDAR; obstacle detection and recognition libraries.