People Fighting Recognition

People Fighting Detection

Nodeflux People Fighting Recognition

Nodeflux People Fighting Recognition addresses the action recognition instances of people fighting and monitors the emergence of riots event, promoting improved public security and surveillance within monitored environments. This solution integrates advanced Large Vision Models and Visual Transformers directly into the Nodeflux Visionaire platform. The analytics system is capable of automatically detecting individuals fighting each other, through the inference process from multiple sequential frames, in real-time scenarios.

Designed Applications

Nodeflux People Fighting Recognition is primarily intended for use in law enforcement and surveillance applications. It monitors the occurrence of individuals fighting each other and aids in mitigating or even preventing potential risks or threats associated with riot events that may be impacting in mortality cases.

Disclaimer: In a sense of large vision models and visual transformers technology, the performance of this analytic might slightly differ in your environment, depending on several variables such as camera specs, camera height, camera angle, weather conditions, etc. We highly recommend you to test our analytics and run benchmarks on your own images, with ground truth or your quality expectations prepared beforehand. Please contact us for more info.

Architecture

Visionaire People Fighting Recognition utilizes a combination of other services:

1. Postgres - For database,

2. Docker Snapshot - For action recognition related,

3. Visionaire Docker Stream (must be v4.57.21 and above) - For video stream processing,

4. Visionaire Dashboard (Optional) - Built-in dashboard for visualization.

Additional Necessary Deployment (VLM)

Since Nodeflux People Fighting Recognition utilizes the snapshot platform, an additional docker-compose.yml file needs to be deployed alongside the other aforementioned services. The following directive link outlines the installation process for deploying the docker-compose.yml file specific to the Large Visual Language Model (VLM) pipeline of snapshot platform:

VLM Installation

Analytics Configuration

When you are assigning this analytics into the stream through the dashboard, you should configure some parameters setting below:

{
  "address": "http://192.168.103.122:4008",
  "always_dump": false,
  "dump_interval": 2,
  "images_num": 1,
  "interval_capture": 0.5,
  "is_dump_collage": false
}

Parameter	Explanation
address	The IP address and its port where the VLM snapshot service is deployed
always_dump	The setup to configure whether the event dump should be dumping for positive event only (the value is false) or both positive & negative events (the value is true). Default value is false.
dump_interval	How frequent you want the stream as the multiple sequential frames/snapshots to get the inference result. Unit in second
images_num	Number of sequential frames/snapshots that generated to be analyze in the inference process. Value range between 1 to 6 images. If the value is 1, then only single image frame will be sent to the inference process.
interval_capture	Interval time between 2 frames/snapshots within the same sequential frames/snapshot. Unit in second. The configuration only works when images_num is above 1.
is_dump_collage	If the images_num value is above 1, then the event dump may be configured to only dump the first single image of the sequence of images_num (the value is false) or all images in the sequence of images_num (the value is true).