Subscriber Discussion

Toy Or Tool? Video-Based Firearm Detection

UE
Undisclosed End User #1
•Jun 18, 2018

Does anyone have experience using video analytics to detect the presence of firearms in video? If so:

- What software did you use/try?

- Was it more of a toy or tool?

Use Case 1: Video Triage. User wants to scan video archives and tag any files/frames that contain firearms. All tagged videos will be reviewed by a human operator. User has *no* control over the format/resolution of the video. User strongly prefers pre-trained recognition models.

Use Case 2: Live Monitoring. User wants to monitor live feeds and alert operators if a firearm is detected. User upgrades cameras (i.e., a mix of EO/IR assets) approximately every 5 years and is not vendor locked. User has onsite operators but would consider external Cloud-based services. Radar-based systems and shot-detection technologies that rely on audio are out-of-scope.

UE
Undisclosed End User #1
•Jun 18, 2018

Ideally, I’d like to identify at least 3 video analytic options for each use case. So far, based on vendor materials and IPVM articles, I have only identified 1 possible option for each…

- Use Case 1: Video Triage… Possible Option: Pixel Forensics, + ?

- Use Case 2: Live Monitoring… Possible Option: Deep Science, + ?

Any other video analytic options I should have under consideration?

UE
Undisclosed End User #1
•Jun 19, 2018

Came across these product references* in 2014 and 2015 IPVM articles:

- ReardentIT's SafeZone Firearm Detection analytic (https://ipvm.com/reports/gun-detection-analytics-safezone-examined)

- NanoWatt Design's GunDetect (https://ipvm.com/reports/gun-detecting-smartcam-startup-gundetect#)

*Of the two, GunDetect appears to still be marketed. I have not, however, uncovered any evidence to suggest that the product is in use or effective.

Given the Video Triage and Live Monitoring use cases discussed, what video analytic options should I consider?

JH
John Honovich
•Jun 19, 2018
IPVM

I am pretty skeptical of it and know of no well established company in that space.

Btw, a recent random think on LinkedIn that went semi-viral, a student with an open sourced video based gun detector:

UM
Undisclosed Manufacturer #2
•Jun 20, 2018

This is Suresh and represent Duranc. we have been training for various custom objects using machine learning frameworks and guns is one them. Here is a sample video of a store robbery which shows different objects being detected and to the left prediction confidence.

The algorithm/network can be run on archived videos for post processing as well as on live video streams to detect objects using Duranc platform.

(1)
UE
Undisclosed End User #1
•Jun 22, 2018
In reply to Undisclosed Manufacturer #2

Thanks for your response. When you have a moment, I have two follow-up questions:

  • Is this capability available today or planned for a future release? The Duranc website (http://www.duranc.com/) does not appear to reference this capability.

  • Is the Duranc platform leveraging YOLO (you only look once)? The example video closely resembles the output one receives when running YOLO from the command line.
UM
Undisclosed Manufacturer #2
•Jun 22, 2018
In reply to Undisclosed End User #1

Please reach out to me @ sureshy@duranc.com for further questions and clarifications.

Avatar
Skip Cusack
•Jun 20, 2018

It certainly seems like a solvable problem, and one worth doing. Looks like a good opportunity for a tech company... 

MS
Mohamed Salah
•Jun 23, 2018

You may want to try VSBLTY

http://vsblty.net

 

I believe they are working on a live monitoring solution.  Was shown a couple of clips a while back.

New discussion

Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.

Newest discussions