The visual coupling between neighbors explains 'flocking' in human crowds

Patterns of collective motion or 'flocking' in birds, fish schools, and human crowds are believed to emerge from local interactions between individuals. Most models of collective motion attribute these interactions to hypothetical rules or forces, often inspired by physical systems, and described from an overhead view. We develop a visual model of human flocking from an embedded view, based on optical variables that actually govern pedestrian interactions. Specifically, people control their walking speed and direction by canceling the average optical expansion and angular velocity of their neighbors, weighted by visual occlusion. We test the model by simulating data from experiments with virtual crowds and real human 'swarms'. The visual model outperforms our previous overhead model and explains basic properties of physics-inspired models: 'repulsion' forces reduce to canceling optical expansion, 'attraction' forces to canceling optical contraction, and 'alignment' to canceling the combination of expansion/contraction and angular velocity. Critically, the neighborhood of interaction follows from Euclid's Law of perspective and the geometry of occlusion. We conclude that the local interactions underlying human flocking are a natural consequence of the laws of optics. Similar principles may apply to collective motion in other species.

How Is Urban Greenness Spatially Associated with Dockless Bike Sharing Usage on Weekdays, Weekends, and Holidays?

Dockless bike sharing plays an important role in residents’ daily travel, traffic congestion, and air pollution. Recently, urban greenness has been proven to be associated with bike sharing usage around metro stations using a global model. However, their spatial associations and bike sharing usage on public holidays have seldom been explored in previous studies. In this study, urban greenness was obtained objectively using eye-level greenness with street-view images by deep learning segmentation and overhead view greenness from the normalized difference vegetation index (NDVI). Geographically weighted regression (GWR) was applied to fill the research gap by exploring the spatially varying association between dockless bike sharing usage on weekdays, weekends, and holidays, and urban greenness indicators as well as other built environment factors. The results showed that eye-level greenness was positively associated with bike sharing usage on weekdays, weekends, and holidays. Overhead-view greenness was found to be negatively related to bike usage on weekends and holidays, and insignificant on weekdays. Therefore, to promote bike sharing usage and build a cycling-friendly environment, the study suggests that the relevant urban planner should pay more attention to eye-level greenness exposure along secondary roads rather than the NDVI. Most importantly, planning implications varying across the study area during different days were proposed based on GWR results. For example, the improvement of eye-level greenness might effectively promote bike usage in northeastern and southern Futian districts and western Nanshan on weekdays. It also helps promote bike usage in Futian and Luohu districts on weekends, and in southern Futian and southeastern Nanshan districts on holidays.

A Tool for Annotating Homographies from Hockey Broadcast Video

In order to develop solutions for automatic ice rink localization from broadcast video, a dataset with ground truth homographies is required. Hockey broadcast video does not tend to provide camera parameters for each frame, which means that they must be gathered manually. A novel tool for collecting ground truth transforms through point correspondences between each frame and an overhead view of the ice rink is presented in this paper. Through collaboration with the users of the tool, we have added features to improve accuracy and efficiency, especially in frames with few lines on the playing surface visible. A dataset of 4,262 frames has been collected, which will be used for research into automatic camera calibration techniques.

Convolutional neural network–based person tracking using overhead views

In video surveillance, person tracking is considered as challenging task. Numerous computer vision, machine and deep learning–based techniques have been developed in recent years. Majority of these techniques are based on frontal view images/video sequences. The advancement of convolutional neural network reforms the way of object tracking. The network layers of convolutional neural network models trained on a number of images or video sequences improve speed and accuracy of object tracking. In this work, the generalization performance of existing pre-trained deep learning models have investigated for overhead view person detection and tracking, under different experimental conditions. The object tracking method Generic Object Tracking Using Regression Networks (GOTURN) which has been yielding outstanding tracking results in recent years is explored for person tracking using overhead views. This work mainly focused on overhead view person tracking using Faster region convolutional neural network (Faster-RCNN) in combination with GOTURN architecture. In this way, the person is first identified in overhead view video sequences and then tracked using a GOTURN tracking algorithm. Faster-RCNN detection model achieved the true detection rate ranging from 90% to 93% with a minimum false detection rate up to 0.5%. The GOTURN tracking algorithm achieved similar results with the success rate ranging from 90% to 94%. Finally, the discussion is made on output results along with future direction.

Vehicle Detection in Aerial Images Using a Fast Oriented Region Search and the Vector of Locally Aggregated Descriptors

Vehicle detection in aerial images plays a significant role in civil and military applications and it faces many challenges including the overhead-view perspective, the highly complex background, and the variants of vehicles. This paper presents a robust vehicle detection scheme to overcome these issues. In the detection stage, we propose a novel algorithm to generate oriented proposals that could enclose the vehicle objects properly as rotated rectangles with orientations. To discriminate the object and background in the proposals, we propose a modified vector of locally aggregated descriptors (VLAD) image representation model with a recently proposed image feature, i.e., local steering kernel (LSK) feature. By applying non-maximum suppression (NMS) after classification, we show that each vehicle object is detected with a single-oriented bounding box. Experiments are conducted on aerial images to compare the proposed method with state-of-art methods and evaluate the impact of the components in the model. The results have proven the robustness of the proposed method under various circumstances and the superior performance over other existing vehicle detection approaches.