Research on Faster R-CNN has recently witnessed the progress in both accuracy and execution efficiency in detecting objects such as faces, hands or pedestrians in photograph or video. However, constrained by the size of its convolution feature map output, it is unable to clearly detect small or tiny objects. Therefore, we presented a fast, deep convolutional neural network based on a modified Faster R-CNN. Multiple strategies, such as fast multi-level combination, context cues, and a new anchor generating method were employed for small object detection in this paper. We demonstrated performance of our algorithm both on the KITTI-ROAD dataset and our own traffic scene lane markings dataset. Experiments demonstrated that our algorithm obtained better accuracy than Faster R-CNN in small object detection.
Keywords: Computer Vision; Deep Learning; Image Processing; Intelligent Transportation Systems; Lane Marking Detection.