Efficient Gun Detection in Real-World Videos: Challenges and Solutions
Keywords: Image-augmented training, transfer learning
Abstract: Object detection in videos is a crucial task in the computer vision domain. Existing methods have explored different approaches to detect objects and classify the videos. However, detecting tiny objects (e.g., gun) in videos has always been a challenging and rigorous task. Moreover, the existing video analysis (detection and classification) models may not achieve high accuracy for gun detection in videos in real-world scenarios due to the lack of a large amount of labeled data. Thus, it is imperative to develop an efficient method to capture the features of tiny objects and train models that can perform accurate gun detection. To address this challenge, we make three contributions. First, we perform an empirical study of several existing video classification methods to identify the presence of guns in videos. Our extensive analysis shows that these methods may not achieve high accuracy in detecting guns in videos. Second, we propose a novel gun detection method with image-augmented training and evaluate the technique in real-world settings with different evaluation metrics. Third, our experimental results demonstrate that our proposed domain-specific method can achieve significant performance improvements in real-world settings compared to the other popular methods. We also discuss emerging challenges and critical aspects of detecting tiny objects, e.g., guns, using existing computer vision techniques, their limitations, and future research opportunities.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8347
Loading