Lane Departure Warning System using Deep Learning

딥러닝을 이용한 차로이탈 경고 시스템

Abstract

As artificial intelligence technology has been developed rapidly, many researchers who are interested in next-generation vehicles have been studying on applying the artificial intelligence technology to advanced driver assistance systems (ADAS). In this paper, a method of applying deep learning algorithm to the lane departure warning system which is one of the main components of the ADAS was proposed. The performance of the proposed method was evaluated by taking a comparative experiments with the existing algorithm which is based on the line detection using image processing techniques. The experiments were carried out for two different driving situations with image databases for driving on a highway and on the urban streets. The experimental results showed that the proposed system has higher accuracy and precision than the existing method under both situations.

최근 인공지능 기술이 급격히 발전하면서 첨단 운전자 지원 시스템 분야에 딥러닝 기술을 접목하여 기존의 기술보다 뛰어난 성능을 보여주기 위한 여러 연구들이 진행 되고 있다. 이러한 동향에 맞춰 본 논문 또한 첨단 운전자 지원 시스템의 핵심 요소 중 하나인 차로이탈 경고시스템에 딥러닝 기술을 접목한 방법을 제안한다. 제안하는 방법과 기존의 차선검출 기반의 경고시스템과의 비교 실험을 통해 그 성능을 평가 하였다. 고속도로 주행영상과 시내 주행영상을 이용한 두 가지의 서로 다른 환경에서 모두 제안하는 방법이 정확도 및 정밀도 부분에서 더 높은 수치를 보여주었다.

Fig. 1 The comparison of data processing flow for lane departure warning a) existing method, b) proposed method

Fig. 2 Overview of the line detection algorithm[6]

Fig. 3 Determination of vehicle’s position between two lines a) center b) leaning to left c) leaning to right [6]

Fig. 4 The overall structure of the proposed method

Fig. 5 Examples of Training Data. a) Normal driving data, b) Lane departure data

Fig. 6 Annotation of driving data without top-view transformation a) on a straight lane and b) on a curved lane

Fig. 7 Annotation of driving data with top-view transformation a) on a straight lane and b) on a curved lane

Fig. 8 The proposed CNN architecture.

Fig. 9 The validation accuracy in relation to the different number of epoches

Fig. 10 Sample images of experimental results

Table 1 The number of frames in the dataset for training CNN and validation

Table 2 The result of a comparative experiment with a dataset consisting of highway driving images

Table 3 The result of a comparative experiment with a dataset consisting of street driving images

Table 4 The total result of experiments

Table 5 Average processing time per frame