2018/03/06 by Stefan Zernetsch, Viktor Kreß, Zernetsch, Stefan +5
Engineering · #Autonomous Vehicle Technology and Safety #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Traffic Prediction and Management Techniques #Traffic and Road Safety
paper · pdf · doi:10.48550/arxiv.1803.02242
openalex publication_date 2018/03/06 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
In this article, we present a novel approach to detect starting motions of\ncyclists in real world traffic scenarios based on Motion History Images (MHIs).\nThe method uses a deep Convolutional Neural Network (CNN) with a residual\nnetwork architecture (ResNet), which is commonly used in image classification\nand detection tasks. By combining MHIs with a ResNet classifier and performing\na frame by frame classification of the MHIs, we are able to detect starting\nmotions in image sequences. The detection is performed using a wide angle\nstereo camera system at an urban intersection. We compare our algorithm to an\nexisting method to detect movement transitions of pedestrians that uses MHIs in\ncombination with a Histograms of Oriented Gradients (HOG) like descriptor and a\nSupport Vector Machine (SVM), which we adapted to cyclists. To train and\nevaluate the methods a dataset containing MHIs of 394 cyclist starting motions\nwas created. The results show that both methods can be used to detect starting\nmotions of cyclists. Using the SVM approach, we were able to safely detect\nstarting motions 0.506 s on average after the bicycle starts moving with an\nF1-score of 97.7%. The ResNet approach achieved an F1-score of 100% at an\naverage detection time of 0.144 s. The ResNet approach outperformed the SVM\napproach in both robustness against false positive detections and detection\ntime.\n