Kestrel: Video Analytics for Augmented Multi-Camera Vehicle Tracking

Abstract

In the future, the video-enabled camera will be the most pervasive type of sensor in the Internet of Things. Such cameras will enable continuous surveillance through heterogeneous camera networks consisting of fixed camera systems as well as cameras on mobile devices. The challenge in these networks is to enable efficient video analytics: the ability to process videos cheaply and quickly to enable searching for specific events or sequences of events. In this paper, we discuss the design and implementation of Kestrel, a video analytics system that tracks the path of vehicles across a heterogeneous camera network. In Kestrel, fixed camera feeds are processed on the cloud, and mobile devices are invoked only to resolve ambiguities in vehicle tracks. Kestrel's mobile device pipeline detects objects using a deep neural network, extracts attributes using cheap visual features, and resolves path ambiguities by careful association of vehicle visual descriptors, while using several optimizations to conserve energy and reduce latency. Our evaluations show that Kestrel can achieve precision and recall comparable to a fixed camera network of the same size and topology, while reducing energy usage on mobile devices by more than an order of magnitude.

Video Caching, Analytics, and Delivery at the Wireless Edge: A Survey and Future Directions

“…The following systems leverage other mobile devices such as smartphones to carry out analytics. Qiu et al [185] design a system to track a car's path over a network of fixed surveillance cameras that uses computational resources on mobile devices (e.g., smartphones and cameras on-board vehicles) when necessary. Specifically, the tracking system uses a light-weight analytics pipeline on the mobile devices.…”

Identification of Warning Situations in Road Using Cloud Computing Technologies and Sensors Available in Mobile Devices: A Systematic Review

“…No [67] Dataset available at [68] It is a dataset of images and annotation, together with standardised evaluation software. No [69] Dataset available at [70] It is a dataset of 540 fingerprints representing 27 device types.…”

Coral-Pie

“…Color-histogram by itself is not capable of distinguishing between vehicles with similar colors. However, coupling this feature with the temporal and spatial locality of a given vehicle's movement between adjacent cameras, it is still possible to achieve a highly accurate cross-camera vehicle re-identification with color histograms [25]. Finally, a detection event is generated for the vehicle which is a JSON object that contains the name of the camera that detected the vehicle, the UTC timestamp of the detection, the features including moving direction and adaptive histogram, the ID of the vehicle generated by the Sort Tracker, and the ID of the corresponding vertex in the trajectory graph (see Section 4.2.1).…”

“…On the other hand, they found the network bandwidth limitation and extra administrator cost of processing live camera streams during their Cloud emulation, which motivates our work. Kestrel [25] is a multi-camera vehicle tracking system for both static cameras and cameras on mobile devices, where mobile cameras handle the ambiguity of vehicle trajectory constructed from static cameras. However, it is not a real-time system for tracking all vehicles and it processes all the static camera streams in the Cloud, which creates a network burden for large-scale camera systems.…”