The AEB-P (Autonomous Emergency Braking Pedestrian) system has the functional requirements of avoiding the pedestrian collision and ensuring the pedestrian’s life safety. By studying relevant theoretical systems, such as TTC (time to collision) and braking safety distance, an AEB-P warning model was established, and the traffic safety level and work area of the AEB-P warning system were defined. The upper-layer fuzzy neural network controller of the AEB-P system was designed, and the BP (backpropagation) neural network was trained by collected pedestrian longitudinal anti-collision braking operation data of experienced drivers. Also, the fuzzy neural network model was optimized by introducing the genetic algorithm. The lower-layer controller of the AEB-P system was designed based on the PID (proportional integral derivative controller) theory, which realizes the conversion of the expected speed reduction to the pressure of a vehicle braking pipeline. The relevant pedestrian test scenarios were set up based on the C-NCAP (China-new car assessment program) test standards. The CarSim and Simulink co-simulation model of the AEB-P system was established, and a multi-condition simulation analysis was performed. The results showed that the proposed control strategy was credible and reliable and could flexibly allocate early warning and braking time according to the change in actual working conditions, to reduce the occurrence of pedestrian collision accidents.

To improve the accuracy of lane detection in complex scenarios, an adaptive lane feature learning algorithm which can automatically learn the features of a lane in various scenarios is proposed. First, a two-stage learning network based on the YOLO v3 (You Only Look Once, v3) is constructed. The structural parameters of the YOLO v3 algorithm are modified to make it more suitable for lane detection. To improve the training efficiency, a method for automatic generation of the lane label images in a simple scenario, which provides label data for the training of the first-stage network, is proposed. Then, an adaptive edge detection algorithm based on the Canny operator is used to relocate the lane detected by the first-stage model. Furthermore, the unrecognized lanes are shielded to avoid interference in subsequent model training. Then, the images processed by the above method are used as label data for the training of the second-stage model. The experiment was carried out on the KITTI and Caltech datasets, and the results showed that the accuracy and speed of the second-stage model reached a high level.

With the aim to achieve an accurate lateral distance between vehicle and lane boundaries during the road test of Lane Departure Warning and Lane Keeping Assist, this study proposes a recognition model to estimate the distance directly by training a deep neural network, called LatDisLanes. The neural network model obtains the distance using two down-face cameras without data pre-processing and post-processing. Nevertheless, the accuracy of recognition is disrupted by inclination angle, but the bias is decreased using a proposed dynamic correction model. Furthermore, as training a model requires a large number of label images, an image synthesis algorithm that is based on the Image Quilting is proposed. The experiment on test data set shows that the accuracy of LatDisLanes is 94.78% and 99.94%, respectively, if the allowable error is 0.46 cm and 2.3 cm when the vehicle runs smoothly. In addition, a bigger error can be caused when inclination angle is greater than 3°, but the error can be reduced by proposing a dynamic correction model.