Detecting and Tracking Insects in the InsectEye Device

Restricted (Penn State Only)
- Author:
- Raju, Anand
- Area of Honors:
- Computer Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Vijaykrishnan Narayanan, Thesis Supervisor
Vijaykrishnan Narayanan, Thesis Honors Advisor
John Joseph Hannan, Faculty Reader - Keywords:
- Tracking
Detection
Neural Networks - Abstract:
- Insect population tracking and gaining metrics of their populations poses a significant problem to conventional tracking methodologies, which do not function well with changing shapes, sizings, and erratic movements of objects. However, understanding the motions and being able to quickly canvas species of insects is a critical problem to tracking populations of individual species in an insect habitat, particularly for understanding counts and tracking invasive species. The main point of creating the device is to provide a method of data collection on insects that does not end in the death of the insects. A further complication to this problem is handling this near real-time, as insects move throughout areas extremely rapidly. This work focuses on solving this problem in an edge-style fashion, deployed directly at the site of insects. This work takes a look at using kernelized correlation filters to track objects in a Re-Identification fashion, and associate IDs with object detections in video of insects. In particular, this model detects, tracks, and counts insects in videos received from the InsectEye device. This paper covers various options for object detection and tracking, but focuses in-depth on YOLO detection, KCF tracking, and Re-Identification tracking methods. With increased processing power, considering that KCF and YOLOv5 are both used for real-time detection and tracking, the hope is that our solution can run real-time on the edge within the InsectEye device.