A Dataset and System for Real-Time Gun Detection in Surveillance Video Using Deep Learning

Qi, Delong; Tan, Weijun; Liu, Zhifu; Yao, Qi; Liu, Jingfeng

doi:10.1109/smc52423.2021.9659207

Cited by 13 publications

(5 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Later, [20] published a dataset with 51,000 annotated images for gun detection, and most of these images were selected from IMFDB [1], and some were from previously published datasets such as [19]. They fine-tune a Centernet with Mobilenet backbone pretrained on Pascal VOC.…”

Section: Firearm Detectionmentioning

confidence: 99%

“…The studies in [9,29] train networks to classify whether an image has a handgun but skip the critical step of gun detection. [19,20] train Faster R-CNN models to detect handguns, but the images in their datasets are neither from a CCTV perspective nor a real-world surveillance scene. The images in [20] are in ideal conditions and high resolution but mostly are from movie scenes.…”

Section: Firearm Detectionmentioning

confidence: 99%

“…[19,20] train Faster R-CNN models to detect handguns, but the images in their datasets are neither from a CCTV perspective nor a real-world surveillance scene. The images in [20] are in ideal conditions and high resolution but mostly are from movie scenes. The dataset constructed by [8] can be used to pre-train a gun detection model, but they still need to evaluate their model on real crime scene data or analyze SOTA models on the dataset.…”

Section: Firearm Detectionmentioning

confidence: 99%

“…Some preliminary studies of gun detection from images/videos have been conducted based on generic object detectors. Most of these studies (e.g., [9,19,20]) focus on well-processed gun images very different from images in Figure 1: Example surveillance images involving handgun violence from the UCF-Crime dataset [26]. Handguns in low-resolution images are tiny and seriously occluded by hands, so there are no salient textures and shapes for detection.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

CCTV-Gun: Benchmarking Handgun Detection in CCTV Images

Srikar¹,

Li²,

Doshi³

et al. 2023

Preprint

View full text Add to dashboard Cite

Gun violence is a critical security problem, and it is imperative for the computer vision community to develop effective gun detection algorithms for real-world scenarios, particularly in Closed Circuit Television (CCTV) surveillance data. Despite significant progress in visual object detection, detecting guns in real-world CCTV images remains a challenging and under-explored task. Firearms, especially handguns, are typically very small in size, nonsalient in appearance, and often severely occluded or indistinguishable from other small objects. Additionally, the lack of principled benchmarks and difficulty collecting relevant datasets further hinder algorithmic development. In this paper, we present a meticulously crafted and annotated benchmark, called CCTV-Gun, which addresses the challenges of detecting handguns in real-world CCTV images. Our contribution is three-fold. Firstly, we carefully select and analyze real-world CCTV images from three datasets, manually annotate handguns and their holders, and assign each image with relevant challenge factors such as blur and occlusion. Secondly, we propose a new crossdataset evaluation protocol in addition to the standard intra-dataset protocol, which is vital for gun detection in practical settings. Finally, we comprehensively evaluate both classical and state-of-the-art object detection algorithms, providing an in-depth analysis of their generalizing abilities. The benchmark will facilitate further research and development on this topic and ultimately enhance security. Code, annotations, and trained models are available at https://github.com/srikarym/CCTV-Gun.

show abstract

Section: Firearm Detectionmentioning

confidence: 99%

Section: Firearm Detectionmentioning

confidence: 99%

Section: Firearm Detectionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

CCTV-Gun: Benchmarking Handgun Detection in CCTV Images

Srikar¹,

Li²,

Doshi³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The second dataset we discuss here was assembled by Qi et al (2021) and is by far the largest we consider. It features 51,882 images of firearms including images from movies, CCTV, and stock images.…”

Section: Review Of Firearm Detection Datasetsmentioning

confidence: 99%

Firearms on Twitter: A Novel Object Detection Pipeline

Harvey

Lebret

Massonnet

et al. 2023

ICWSM

View full text Add to dashboard Cite

Social media is an important source of real-time imagery concerning world events. One subset of social media posts which may be of particular interest are those featuring firearms. These posts can give insight into weapon movements, troop activity and civilian safety. Object detection tools offer important opportunities for insight into these images. Unfortunately, these images can be visually complex, poorly lit and generally challenging for object detection models. We present an analysis of existing gun detection datasets, and find that these datasets to not effectively address the challenge of gun detection on real-life images. Following this, we present a novel object detection pipeline. We train our pipeline on a number of datasets including one created for this investigation made up of Twitter images of the Russo-Ukrainian War. We compare the performance of our model as trained on the different datasets to baseline numbers provided by original authors as well as a YOLO v5 benchmark. We find that our model outperforms the state-of-the-art benchmarks on contextually rich, real-life-derived imagery of firearms.

show abstract