Development of an object detection and mask generation software for dynamic beam projection in automotive pixel lighting applications

Nowadays there are many contributions to the automotive industry and the field is developing fast. This work can be used for some real-time autonomous driving applications. The goal was to add advanced functionality to a standard light source in collaboration with electronic systems. Including advanced features may result in safer and more pleasant driving. The application fields of the work could include glare-free light sources, orientation and lane lights, marking lights, and symbol projection. On a real-time source, object detection and classification with a confidence score is implemented. The best model is obtained by intending to train the model with varying parameters. The most accurate result which is mAP value 0.572 was obtained by distributing the training dataset with learning rate 0.2 and setting the epochs to 300. Moreover, a basic implementation of a glare-free light source was done to avoid the drivers from being blinded by the illumination of the beams. The car and rectangle shape masks were generated as image files and sent as CSV files to the pixel light source device. As a result, the rectangle shaped mask functions more precisely then car shaped.