Monocular Cyclist Detection with Convolutional Neural Networks

Cycling is an increasingly popular method of transportation forsustainability and health benefits. However, cyclists face growing risks,especially when encountering large vehicles on the road. This study aims toreduce the number of vehicle-cyclist collisions, which are often caused by poordriver attention to blind spots. To achieve this, we designed astate-of-the-art real-time monocular cyclist detection that can detect cyclistswith object detection convolutional neural networks, such as EfficientDet Liteand SSD MobileNetV2. First, our proposed cyclist detection models achievegreater than 0.900 mAP (IoU: 0.5), fine-tuned on a newly proposed cyclist imagedataset comprising over 20,000 images. Next, the models were deployed onto aGoogle Coral Dev Board mini-computer with a camera module and analyzed forspeed, reaching inference times as low as 15 milliseconds. Lastly, theend-to-end cyclist detection device was tested in real-time to model trafficscenarios and analyzed further for performance and feasibility. We concludedthat this cyclist detection device can accurately and quickly detect cyclistsand has the potential to improve cyclist safety significantly. Future studiescould determine the feasibility of the proposed device in the vehicle industryand improvements to cyclist safety over time.