No abstract

Many modern approaches for object detection are two-staged pipelines. The first stage identifies regions of interest which are then classified in the second stage. Faster R-CNN is such an approach for object detection which combines both stages into a single pipeline. In this paper we apply Faster R-CNN to the task of company logo detection. Motivated by its weak performance on small object instances, we examine in detail both the proposal and the classification stage with respect to a wide range of object sizes. We investigate the influence of feature map resolution on the performance of those stages.Based on theoretical considerations, we introduce an improved scheme for generating anchor proposals and propose a modification to Faster R-CNN which leverages higher-resolution feature maps for small objects. We evaluate our approach on the FlickrLogos dataset improving the RPN performance from 0.52 to 0.71 (MABO) and the detection performance from 0.52 to 0.67 (mAP).

In this paper we consider the problem of human pose estimation in real-world videos of swimmers. Swimming channels allow filming swimmers simultaneously above and below the water surface with a single stationary camera. These recordings can be used to quantitatively assess the athletes' performance. The quantitative evaluation, so far, requires manual annotations of body parts in each video frame. We therefore apply the concept of CNNs in order to automatically infer the required pose information. Starting with an off-the-shelf architecture, we develop extensions to leverage activity information -in our case the swimming style of an athlete -and the continuous nature of the video recordings. Our main contributions are threefold: (a) We apply and evaluate a fine-tuned Convolutional Pose Machine architecture as a baseline in our very challenging aquatic environment and discuss its error modes, (b) we propose an extension to input swimming style information into the fully convolutional architecture and (c) modify the architecture for continuous pose estimation in videos. With these additions we achieve reliable pose estimates with up to +16% more correct body joint detections compared to the baseline architecture.

In this paper we address the problem of automatic event detection in athlete motion for automated performance analysis in athletics. We specifically consider the detection of stride-, jump-and landing related events from monocular recordings in long and triple jump. Existing work on event detection in sports often uses manually designed features on body and pose configurations of the athlete to infer the occurrence of events. We present a two-step approach, where temporal 2D pose sequences extracted from the videos form the basis for learning an event detection model. We formulate the detection of discrete events as a sequence translation task and propose a convolutional sequence network that can accurately predict the timing of event occurrences. Our best performing architecture achieves a precision/recall of 92.3%/89.0% in detecting start and end of ground contact during the run-up and jump of an athlete at a temporal precision of ±1 frame at 200Hz. The results show that 2D pose sequences are a suitable motion representation for learning event detection in a sequence-to-sequence framework. CCS CONCEPTS• Computing methodologies → Visual content-based indexing and retrieval; Neural networks; Object detection. KEYWORDS event detection, video indexing, computer vision in sports, convolutional sequence modeling

In this work we propose a novel approach to automatically detect a swimmer and estimate his/her pose continuously in order to derive an estimate of his/her stroke rate given that we observe the swimmer from the side. We divide a swimming cycle of each stroke into several intervals. Each interval represents a pose of the stroke. We use specifically trained object detectors to detect each pose of a stroke within a video and count the number of occurrences per time unit of the most distinctive poses (so-called key poses) of a stroke to continuously infer the stroke rate. We extensively evaluate the overall performance and the influence of the selected poses for all swimming styles on a data set consisting of a variety of swimmers.