Search

Real-time video analytics on small autonomous drones poses several difficult challenges at the intersection of wireless bandwidth, processing capacity, energy consumption, result accuracy, and timeliness of results. In response to these challenges, this paper describes four strategies to build adaptive computer vision pipelines for domains such as search-and-rescue, surveillance, and wildlife conservation. Our experimental results show that a judicious combination of drone-based processing and edge-based processing can save substantial wireless bandwidth and thus improve scalability, without compromising result accuracy or latency.

The generation of high-quality training data has become the key bottleneck in the use of deep learning across many domains. In this paper, we describe Eureka, an interactive system that leverages edge computing and early discard to greatly improve the productivity of experts in the construction of a labeled dataset. Our experimental results show that Eureka reduces the labeling effort needed to construct a training set by two orders of magnitude relative to a brute-force approach.