Implementation of Face Detection and Eye Tracking with Convolution Neutral Network and Edge Computing on Electromobility

The eye tracking obtains the position data of both eyes; the fuzzy intent decision maker chooses the desired action for embedded vehicle. The action operation instructions are sent to edge computing to control the appliance and algorithm. The proposed article adopts eye aspect ratio and facial landmark detection with built-in pop count of XNOR-net to judgment driver monitoring system (DMS) on APEX2 and ARM NEON for NXP S32V234 platform. Verification of facial landmark detection and human behavior based on the human face image feature by camera is studied. Implementation network of the computing time upgrade to 59 ms from original 1624 ms under the accuracy loss is smaller than 3%. The overall algorithm has been improved to 16 fps from 0.5 fps.