EAGER- A Cloud-assisted Framework for Improving Pedestrian Safety in Urban Communities using Crowd-sourced Mobile Device Data
pdf
Pedestrian distraction continues to be a significant pedestrian safety concern in urban communities. Existing pedestrian safety frameworks and applications are unable to simulataneously detect complex distraction-related activities (often focusing on detecting only specific activities and contexts) and ignore the hazards posed by the distracted pedestrian to fellow pedestrians and drivers. Moreover, a majority of existing complex activity recognition schemes are either computationally impractical for real-time implementation on mainstream mobile and wearable devices, or employ specialized auxiliary hardware, or both. Thus, there is an urgent need for usable and accurate pedestrian safety solutions that can be efficiently implemented (and used) on commercial off-the-shelf (or COTS) mobile and wearable devices. This project proposes a cloud-assisted pedestrian safety framework that detects several commonly observed activities resulting in pedestrian distraction by using multi-modal and multi-source data from users’ mobile and wearable devices, and provides appropriate on-device and community notifications, with the objective of achieving a favorable balance between responsiveness, computational efficiency, detection accuracy and usability.
The first phase of the project will focus on a customized hierarchical complex activity recognition model to accurately and efficienty detect (in real-time) complex activities that are a cause of pedestrian distractions. The proposed detection and recognition framework will employ multi-modal and multi-source data from sensors on-board COTS mobile devices carried by the pedestrians, such as, smartphones and smartwatches. The proposed framework will be evaluated against existing state-of-the-art complex human activity recognition proposals. Furthermore, the practical feasibility of the proposed framework will be demonstrated by development of a prototype system and application for popular mobile and wearable device environments (e.g., Android Wear), and by conducting a comprehensive evaluation of the same. The second phase of the project will involve integration of a privacy-preserving and context-aware cloud-based notification service within the pedestrian safety framework to alert pedestrians and drivers about distracted pedestrians in the vicinity, thus allowing them to prevent impending accidents or safety hazards. Here, the hazardous contexts are identified in the spatio-temporal continuum through real-time and privacy-preserving analysis of crowd-sourced event data. This will be followed by a thorough quantitative and qualitative evaluation of the proposed pedestrian safety framework by means of a campus-wide testbed deployment.
Submitted by Murtuza Jadliwala
on
Pedestrian distraction continues to be a significant pedestrian safety concern in urban communities. Existing pedestrian safety frameworks and applications are unable to simulataneously detect complex distraction-related activities (often focusing on detecting only specific activities and contexts) and ignore the hazards posed by the distracted pedestrian to fellow pedestrians and drivers. Moreover, a majority of existing complex activity recognition schemes are either computationally impractical for real-time implementation on mainstream mobile and wearable devices, or employ specialized auxiliary hardware, or both. Thus, there is an urgent need for usable and accurate pedestrian safety solutions that can be efficiently implemented (and used) on commercial off-the-shelf (or COTS) mobile and wearable devices. This project proposes a cloud-assisted pedestrian safety framework that detects several commonly observed activities resulting in pedestrian distraction by using multi-modal and multi-source data from users’ mobile and wearable devices, and provides appropriate on-device and community notifications, with the objective of achieving a favorable balance between responsiveness, computational efficiency, detection accuracy and usability.
The first phase of the project will focus on a customized hierarchical complex activity recognition model to accurately and efficienty detect (in real-time) complex activities that are a cause of pedestrian distractions. The proposed detection and recognition framework will employ multi-modal and multi-source data from sensors on-board COTS mobile devices carried by the pedestrians, such as, smartphones and smartwatches. The proposed framework will be evaluated against existing state-of-the-art complex human activity recognition proposals. Furthermore, the practical feasibility of the proposed framework will be demonstrated by development of a prototype system and application for popular mobile and wearable device environments (e.g., Android Wear), and by conducting a comprehensive evaluation of the same. The second phase of the project will involve integration of a privacy-preserving and context-aware cloud-based notification service within the pedestrian safety framework to alert pedestrians and drivers about distracted pedestrians in the vicinity, thus allowing them to prevent impending accidents or safety hazards. Here, the hazardous contexts are identified in the spatio-temporal continuum through real-time and privacy-preserving analysis of crowd-sourced event data. This will be followed by a thorough quantitative and qualitative evaluation of the proposed pedestrian safety framework by means of a campus-wide testbed deployment.