Hello Friends,

Incident / Accident Management is requirement for many guidelines and regulations and needed to be properly managed within an organization. If an incident is not properly managed it results in loss of information, chances for incident avoidance in future and monetary challenges. Deep Learning Techniques like Tensorflow / Keras along with OpenCV and other libraries can be used to automatically manage incidents / accidents within an organization.

melted food on the floor
Photo by Christopher Thoms on Pexels.com

The framework for the analysis consists of following scenario. A series of web cams are deployed within your industry so that they can stream in real time the situation of the plant to a logging system. Either the system is designed to analyse images at pre – defined intervals to retrieve the images and do real-time classification to identify if there is incident. The various incidents categories can be like whether spill happened or not or if someone fallen etc. Based on the image analysis detailed information about the different persons involved and their key attributes like department to which they belong and other attributes like age, supervisor information, emergency contact information is retrieved. Once details about the incident and person is obtained an notification is triggered to incident logging system like SAP EHSM Incident Management or even to other systems like ETQ Reliance etc. Once incident is approved and classification and detailed incident investigation is accomplished a series of analysis are carried out to retrieve more insights from the incidents.


figure 2

Below are the various scenarios which are part of the E2E incident management system. The various components consist of  

        Basics of Image Analysis using different libraries, basics of image analysis is performed like displaying an image, analysis of image, changing the alignment, format, brightness, hue, rotation, feature mapping etc.

        Incident / No Incident classification system – which is based on OpenCV / Keras based image / video analysis in identifying an incident within an organization using series of image analysis streamed from web Cams. Detailed classification techniques using Keras are implemented to identify the pattern.

        Facial detection during incident / accident – Automatic logging of persons involved during the incident / accident will be carried out by personal identification using image analysis. The key attributes like name of the persons their department, their age and emergency contact information etc are retrieved from the system automatically once the person is identified.  

        Incident Analysis – Based on incident notification information and person involved a series of descriptive, prescriptive analysis on incidents to gain insights about the personal behaviour and incident statistics etc.

In my upcoming blogs I would like to cover in depth analysis of each of the above steps like right from image analysis using various techniques like skimage or CV2 libraries, image transformation techniques and image / video classification using markers for frames analysis. Retrieval of personal information once a person is identified and triggering initial notification system. Finally, I would like to provide in depth analysis of exploratory data analysis of incident / accident to retrieve insights about the incident. 


This Post Has 3 Comments

  1. Hello Jak, would this approach take into consideration the different causal factors inherent in most incidents such as human behaviour, procedural failures etc which would not be obvious from the various analytical methods?

    1. Hi Wi
      Here the intention is to capture an event and person involved. Based in person involved we can retrieve his past behaviour and do analysis. Detailed analysis of what that person was doing and procedures etc is too much for image classification and can best managed during incident investigation or root cause assessment


      1. Hi Jak ,

        thanks for clarifying, awesome write, I will take a more in-depth look since I am new to Data Driven Safety.

        Keep the good work up.

Leave a Reply

Close Menu
%d bloggers like this: