Optical Flow for Collision Avoidance in Autonomous Cars

Autonomous cars, robotic platforms and other devices capable of unassisted movement are becoming widely considered as superior to human-based control in many areas. Such platforms, however, often are constructed using expensive equipment. We investigate the possibility of using simple setup consisting of an embedded computer module, such as smartphone, single camera and inertial measurement unit along with the concept of the optical flow to detect possible collisions given real-time, onboard processing as an alternative to compound systems based on radar and lidar devices. While most optical flow algorithms are not applicable for real-time processing, our findings prove that those which sacrifice accuracy to gain speed can still be upgraded while remaining accurate enough for the field of collision avoidance. We propose modifications to further enhance optical flow for given context. Our findings prove that using proposed setup consisting of both hardware and software allows for omitting expensive sensors in the field of collision avoidance.