Designing a Lidar Sensor Classifier Using a MATLAB Framework

A deep learning-based classifier is an important component in the environment perception of a moving vehicle, and classified output from the classifier is essential to proceed further in the vehicle data chain. There are several sensors in the vehicle network to provide the environment inferencing. We focused on LIDAR-based sensors in our experimentation to design a classifier for object classification from LIDAR sensor data. We targeted our approach to reveal needs for the production environment and realize an open-source deep network module using MATLAB^® and Deep Learning Toolbox™. We also generated production code using Embedded Coder^® to deploy into the embedded platform. We used this deep learning kit to design the DNN network and fine-tune and debug the missing layers. Once this model was executed in a MATLAB environment, we proceeded to generate production code using Embedded Coder. The generated code was then ready for compiling and deployment on NVIDIA–XAVIER hardware.