Lane and Bump Detection Based on Computer Vision and Deep Learning Methods.

In recent years, Artificial intelligence (AI) is witnessing great progress, which in turn has led to the development of the implementation of computer vision and deep learning in real life. As accidents have been increasing due to leaving the cars off their tracks or not seeing a bump in front of the cars and cause different types of accidents caused because of the lack of attention of the drivers. Therefore, there should be technical developments to reduce the frequency of accidents and keep us safe. In This paper, we worked on both speed bump detection and lane detection. So, the inherent synergies between lane and bump detection are made use to enhance overall computing efficiency while maintaining the robustness of both tasks to fulfil the core aim of safety. Firstly, each system is worked on separately so the result can be seen of each one. For Lane Departure Warning System (LDWS), it warns the driver with a timely signal that the vehicle has left the lane. Lane boundaries are detected using a V2 camera installed at the front of the car that captures a view of the road. This proposed system effectively combines the OpenCV’s canny edge algorithm with perspective warping algorithm and Histogram algorithm to improve the image. The detection speed deployed on the embedded platform Jetson Nano reaches 15 fps and it showed that the proposed plan was robust, and it can be applied on both curved and straight roads. For Speed bump detection, it warns the driver when there is a bump at a distance up to 40 m in front of the car. For bump dataset, images are extracted from videos collected manually in Ismailia streets in Egypt. Many algorithms, such as Faster R-CNN, VGG-16, MobileNet-SSD, YOLOv3 Nano, and YOLOv3 Lite-Mobilenet, are retrained on our dataset, then the results are recorded and the proposed algorithm that achieved the highest speed and accuracy with 89.17% was YOLOv3 Lite-Mobilenet.