Desarrollo de un procedimiento para detectar carriles en vías no señalizadas utilizando visión artificial

Abstract

Traffic accidents are mainly caused by human errors such as inattention, misbehavior or distraction. Many companies have developed techniques to improve driving safety and reduce road accidents. The traffic context is the main stimulus for driver intention and can be used to predict future action. This work develops a procedure that allows detecting the lane on unmarked roads using artificial vision. A video camera in a vehicle is used to obtain images of the driver's field of vision. In the obtained frames, five regions included within the braking distance of the vehicle are established. The first region is located to the left of the car. The next three in the center and the last one on the right. Then, deep learning libraries are incorporated in charge of analyzing these images. The histogram divergence operator is used to quantify textures between regions. The values obtained allowed classifying similarities between regions. For example, in the case where the car travels in the middle lane, five similar textures are obtained. In the cases of transfer in the right or left lane, the divergence increased with respect to the central regions. In this way, it is proposed to estimate the location of the vehicle lane using the values of the divergence between regions as a basis for comparison. This work contributes to technologies related to traffic perception, which include obstacle detection, road structure and lane detection.