1. Spot, an Algorithm for Low-Resolution, Low-Contrast, Moving Object-Tracking with a Non-Stationary Camera
- Author
-
Crutchfield, Christopher Lee
- Subjects
Electrical engineering ,computer vision ,motion tracking ,moving object detection ,moving object tracking - Abstract
The ability to track moving objects in a video stream is helpful for many applications,from pedestrian and vehicle tracking in a city to animal tracking for ecology and conservation.This write-up introduces Spot, an algorithm for moving object tracking in low-resolution, low-contrast videos. This write-up will discuss two motivating examples to guide the development ofSpot–satellite-based surveillance of vehicles in cityscapes and animal tracking using drones forecological purposes.Spot uses image processing techniques to generate a pipeline to track moving objectsframe-to-frame. It then leverages Bayesian Filtering techniques to use the frame-to-frame motionxito track individual identity between consecutive frames.Each stage of Spot’s pipeline–both image processing and the Bayesian Filtering portionsof the pipeline–introduces many parameters. To determine which parameters are ideal for aparticular dataset, a design space exploration tool, dubbed Sherlock, is used to choose the optimalparameters. As part of this, we evaluate multiple possible objective functions and demonstratethe importance of selecting an appropriate one.Spot is competitive with other modern, moving object-tracking algorithms on cityscapedata, outperforming others in some of the metrics presented. For tracking animals from dronefootage, Spot demonstrated an ability to track wildlife at a similar rate to its performance insome cityscape videos.
- Published
- 2023