Motion Detection

Monitoring the track through CCTV and detect if a train is passing by

An OpenCV based PoC Project

Use OpenCV and Python to compare before and after frames directly from CCTV live video.

Reduce the background image noise (ignoring minor movements such as trees, grass)

Detect big object appearance such as a train. Green boxes indicates a train is detected 

Mobirise Website Builder

Safety Monitoring in Action

Site Safety Supervision System (4S) for Construction

Vision based AI solution to monitor unsafe acts and situations and hence to reduce accidents:
1. Unauthorised access to restricted zones, danger zones, lifting zones and no-parking zone
2. Workers near site vehicles or plant
3. Potential collisions between workers and site vehicles or other plant
4. Monitoring of fatigue, distraction, inattentive behaviours of site vehicles drivers and plant operators
5. Workers and other personnel not wearing the required personal protective equipment (PPE), including safety helmet and safety vest
6. Workers working at height either without wearing safety belt

Visual AI Recognition Procedure

1

Data Collection

The first step in visual AI recognition training is to gather a diverse set of training images (or video frames) relevant to the task at hand. This data can be collected from various sources, including in-house image databases, public datasets, or custom image captures.
2

Data Annotation

The collected images then need to be labelled or annotated to assign meaningful tags or descriptors that indicate what specific objects, scenes, or actions are present in the images. This labelling process is an essential step to help the AI model learn how to recognize specific patterns.
3

Model Selection

Based on the specific task and dataset, a suitable AI model architecture needs to be chosen. This may involve selecting a pre-trained model from a library, such as VGG, ResNet or YOLO, or building a custom model from scratch if no existing models are suitable for the task.
4

Training Setup

Once the model is selected, the training data is partitioned into training and validation sets, and data preprocessing is applied to ensure that the images are in a consistent format and size. Then, various training parameters, such as learning rate, batch size, and optimizer, are selected based on the model architecture, dataset size, and computing resources.
5

Model Training

Using the labeled data and the chosen AI model architecture, the training process begins. During this process, the model learns how to recognize the patterns in the data. The training process usually involves several epochs or iterations until the model reaches a satisfactory level of accuracy on both training and validation sets.
6

Evaluation and Fine-tuning

Once the training is complete, the model's performance needs to be evaluated using independent test data. Based on the evaluation results, the model may need to be fine-tuned further to improve its accuracy and reduce generalization error.
7

Deployment

Once the model's performance meets the project's requirements, the model can be deployed in production mode and integrated with the existing system. The deployment process ensures that the model can perform its intended task within the expected time and resource constraints.
8

Ongoing Maintenance

After deployment, the model's performance should be continuously monitored and evaluated to ensure that it remains accurate and up-to-date with changing data patterns and operational requirements.

© Copyright 2023 Big Blocks Limited - All Rights Reserved