Monocular vision-based position sensor using neural networks for automated vehicle following

This paper presents a new position sensor with a CCD camera based on neural networks that measures the distance and direction angle to and the pose angle of the lead vehicle in automated vehicle following. A picture image of lamps mounted on the lead vehicle is obtained with the CCD camera. Lamp positions are established in a rectangular coordinate system by means of graphic data processing. The measuring process of the proposed position sensor is developed by neural network learning with backpropagation. The number of lamps can be reduced four to three without sacrificing sensor accuracy. This reduction in the number of lamps shortens acquisition time in graphic data processing. Experimental results show that the distance, direction angle and pose angle are sufficiency accurate for practical use in automated vehicle following.