Computer Vision Tracking of Crow Flight

Tan, Chee Sheng (2024) Computer Vision Tracking of Crow Flight. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text TAN CHEE SHENG_Full Text.pdf Restricted to Registered users only Download (18MB)

Abstract

The population of crows in urban cities had gradually increased recently and the number of crows-related complaints also increased. Thus, a Computer Vision solution was used to track crows in order to monitor and control their population. In the Computer Vision solution, the object detection algorithm used was YOLOv8 while the object tracking algorithm used was BoT-SORT. For the YOLOv8 model trained, it achieved mAP50 of 0.852 and mAP50-95 of 0.551 while precision of 0.7970, recall of 0.8690 and F1-score of 0.8314 for the test results. For predictions on other flying objects, the classes of pigeon, sparrow, drone, plane and helicopter achieved an accuracy of 81.48%, 82.73%, 77.42% and 86.67%, respectively. This paper also shows that the BoT-SORT algorithm was able to overcome some occlusions and ensure smoother tracking of crows. Meanwhile, the effects of image processing techniques applied and different perspective views on the tracking performance were also discussed. For tracking speed, it was evaluated with different input image size and different computing units such as Raspberry Pi 5.