Drafting involves cycling so close behind another person that wind resistance is significantly reduced, which is illegal during most long distance and several short distance triathlon and duathlon events. In this paper, a proof of concept for a drafting detection system based on computer vision is proposed. After detecting and tracking a bicycle through the various scenes, the distance to this object is estimated through computational geometry. The probability of drafting is then determined through statistical analysis of subsequent measurements over an extended period of time. These algorithms are tested using a static recording and a recording that simulates a race situation with ground truth distances obtained from a Light Detection And Ranging (LiDAR) system. The most accurate developed distance estimation method yields an average error of $0.46$ m in our test scenario. When sampling the distances at periods of 1 or 2 s, simulations demonstrate that a drafting violation is detected quickly for cyclists riding at 2 m or more below the limit, while generally avoiding false positives during the race-like test set-up and five hour race simulations.

In this paper, we describe a robust method for compensating the panning and tilting motion of a camera, applied to foreground–background segmentation. First, the necessary internal camera parameters are determined through feature-point extraction and tracking. From these parameters, two motion models for points in the image plane are established. The first model assumes a fixed tilt angle, whereas the second model allows simultaneous pan and tilt. At runtime, these models are used to compensate for the motion of the camera in the background model. We will show that these methods provide a robust compensation mechanism and improve the foreground masks of an otherwise state-of-the-art unsupervised foreground–background segmentation method. The resulting algorithm is always able to obtain $F_1$ scores above $80\%$ on every daytime video in our test set when a minimal number of only eight feature matches are used to determine the background compensation, whereas the standard approaches need significantly more feature matches to produce similar results.

We describe a method for detecting crossing pedestrians and, in general, any object that is moving perpendicular to the driving direction of the vehicle. This is achieved by combining video snapshots from multiple cameras that are placed in a linear configuration and from multiple time instances. We demonstrate that the proposed array configuration imposes tight constraints on the expected disparity of static objects in a certain image region for a given camera pair. These regions are distinct for different camera pairs. In that manner, static regions can generally be distinguished from moving targets throughout the entire field of view when analyzing enough pairs, requiring only straightforward image processing techniques. On a self-captured dataset with crossing pedestrians, our proposed method reaches an F1 detection score of 83.66% and a mean average precision (MAP) of 84.79% on an overlap test when used stand-alone, being processed at 59 frames per second without GPU acceleration. When combining it with the Yolo V4 object detector in cooperative fusion, the proposed method boosts the maximal F1 scores of this detector on this same dataset from 87.86% to 92.68% and the MAP from 90.85% to 94.30%. Furthermore, combining it with the lower power Yolo-Tiny V4 detector in the same way yields F1 and MAP increases from 68.57% to 81.16% and 72.32% to 85.25%, respectively.