Community Blog Physical Threat Detection: Enhancing Security for Businesses

Physical Threat Detection: Enhancing Security for Businesses

This project is from team GenY-AI, which was awarded the Second Prize in the Alibaba Cloud Singapore AI Hackathon.

This project is from team GenY-AI, which was awarded the Second Prize in the Alibaba Cloud Singapore AI Hackathon.


Project Introduction

The project aims to develop a disrupting application that leverages AI, GenAI, and machine learning to enhance detection capabilities and provide real-time VoIP alerts to security personnel on site for suspicious behaviors. This will allow businesses to proactively respond to security threats efficiently.

Project Value

  1. Wide range of suspicious behavior recognition: Existing solutions are limited in the number of types of suspicious behaviors that can be detected. Our solution attempts to expand this range by training it on significant amounts of CCTV footage.
  2. Real-time alerts: Existing solutions perform suspicious behavior recognition on past CCTV footage and do not incorporate any preventive attempts. Our solution incorporates real-time alerts such that immediate actions can be taken.
  3. On-demand modifications: Regarding deployment, clients may require changes to the application based on various reasons (e.g. lighting condition of the site, differing requirements, etc). Our solution provides a co-pilot that allows clients to perform such changes themselves, thereby removing the need for continuous engineer involvement.


  1. Action recognition module: We utilize a 3D-convolution based classifier to classify a person's action based on a sequence of frames of that person. The model is trained to classify based on a wide range of suspicious behavior. We also made optimizations to ensure the model's ability to run in real time.
  2. Real-time alert system: Following the classifications from 1, we also provide real-time alerts when a person's behavior is detected as suspicious/threatening. An image-to-text model (PromptCap) is also utilized to provide text description of the detected suspect. Finally, we utilize a text-to-speech model (Intelligent Speech Interaction) to convert the textual descriptions to audio, which is then sent to security personnel on-site for immediate actionability.
  3. On-demand modifications: We incorporate a text-to-python module (Qwen-7B) to allow on-demand modifications to our application to suit the client’s needs. The user simply needs to insert a textual prompt (e.g. add "running" to the list of suspicious behavior) and the text-to-python module will make changes to the application to incorporate the requested feature without interrupting execution.

Technology Highlights

We utilized PAI to deploy our Qwen instance. While our image-to-text (Promptcap) and action recognition models are not from Alibaba, we also utilized PAI for those models. In particular, we utilized PAI to train our action recognition model, as well as to host Promptcap to perform inference.

Alibaba Cloud Products Used

About the Team

Hendricks Corp is committed to more than just delivering software or services. We are dedicated to our client’s mission and business needs. We adopt your goals and make them our own to drive success for our clients. This personal commitment, combined with our own values, allows Hendricks Corp to offer a comprehensive portfolio of innovative and intelligent AI and video solutions and services for the commercial, government, and private sectors.

With over 20 years of experience, our quality starts with the people providing our solutions. Hendricks Corp’s culture is to take personal pride and consistently strive for excellence in all our endeavors.

0 1 0
Share on

Alibaba Cloud Project Hub

113 posts | 21 followers

You may also like


Alibaba Cloud Project Hub

113 posts | 21 followers

Related Products