Fast and accurate lightweight model for real-time tiger detection in the wild

In the realm of computer vision, the need for fast and accurate inference has become increasingly critical, particularly in wildlife conservation efforts where real-time detection and monitoring are essential. Lightweight models have gained prominence for their efficiency, but the quest for even...

全面介紹

Saved in:
書目詳細資料
主要作者: Ong, Gregory Chong Jun
其他作者: Deepu Rajan
格式: Final Year Project
語言:English
出版: Nanyang Technological University 2023
主題:
在線閱讀:https://hdl.handle.net/10356/171907
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
機構: Nanyang Technological University
語言: English
實物特徵
總結:In the realm of computer vision, the need for fast and accurate inference has become increasingly critical, particularly in wildlife conservation efforts where real-time detection and monitoring are essential. Lightweight models have gained prominence for their efficiency, but the quest for even greater efficiency and precision without sacrificing their lightweight nature continues. This paper addresses this challenge by proposing an enhanced computer vision model tailored for wildlife detection, with a specific focus on tiger detection as a case study. The primary objective of this research is to develop a more efficient computer vision model that surpasses the existing lightweight models in terms of both inference speed and precision while still maintaining a lightweight profile. To achieve this goal, a comprehensive comparative analysis is conducted, examining the use of popular lightweight models such as MobileNetV2 and ShuffleNetV2 as the backbone for the model. In the forthcoming sections, we delve into the process of crafting an optimal backbone for tiger detection, elucidating the approach through systematic experimentation and rigorous testing. Findings reveal that my proposed backbone outperforms MobileNetV2 and ShuffleNetV2 in the context of tiger detection. The enhanced backbone not only demonstrates superior inference speed but also exhibits higher precision while maintaining a low model complexity. The implications of these results can extend beyond tiger detection, offering a valuable contribution to the broader pursuit of efficient and accurate computer vision models for real-world applications.