Okutama-Action: An Aerial View Video Dataset for Concurrent Human Action Detection

Abstract

Despite significant progress in the development of human action detection datasets and algorithms, no current dataset is representative of real-world aerial view scenarios. We present Okutama-Action, a new video dataset for aerial view concurrent human action detection. It consists of 43 minute-long fully-annotated sequences with 12 action classes. Okutama-Action features many challenges missing in current datasets, including dynamic transition of actions, significant changes in scale and aspect ratio, abrupt camera movement, as well as multi-labeled actors. As a result, our dataset is more challenging than existing ones, and will help push the field forward to enable real-world applications.

Mini-Unmanned Aerial Vehicle-Based Remote Sensing: Techniques, applications, and prospects

“…pedestrians, bikers, cars and buses, to understand pedestrian trajectory and their interact with the physical space as well as with the targets that populate such spaces. This could make a great contribution to pedestrian tracking, target trajectory prediction and activity understanding [238]. In [186], researchers adopt a camera-equipped UAV to record naturalistic vehicle trajectory and naturalistic behavior of road users, which is intended for scenario-based safety validation of highly automated vehicles.…”

Convolutional Neural Networks for Aerial Multi-Label Pedestrian Detection

“…The state of the art deep detectors like SSD, due to the limited computational resources, cannot be trained on high resolution images, therefore, the aerial action detection seems impossible using these types of networks. Authors in [4] show that the SSD512 can detect pedestrian (not their actions) with the good accuracy of Mean Average Precision (mAP@0.5) of 72.3% which is as good as the performance of SSD on the frontal-view datasets like VOC2017 [7]. However, using even a larger input of 960x540, they only report mAP=18.18% for detecting pedestrians' actions.…”

“…For instance, in an urgent situation, on the basis of people's tweets, it might be very important to search for a person who is running and carrying something in a street. In this paper, we assume that objects of interest can be discriminated based on their single or multiple actions, and evaluate the proposed framework on the Okutama-Action dataset [4], which is an aerial dataset for concurrent human single and multiple action detection.…”

“…Blue and Green boxes are respectively the targets for the first and second question, while Red boxes are the wrong answers. Image is a frame from the Okutama-Action dataset [4]. detectors like Faster R-CNN [5], YOLO [6], SSD [7] as well as deep object classifiers like VGG [8] usually work with input images of less than one mega pixel (e.g., 300x300 and 500x500), which makes the problem even worse.…”

Convolutional Neural Networks for Aerial Vehicle Detection and Recognition

“…We propose a framework in which, first, an SSD [3], which has shown promising performances in the aerial image object detection literature [2] and [17], generates a number of objects of interest proposals for an input aerial image. These proposals might contain vehicle, background, or other objects.…”