Object detection using cnn

Details about: Object detection using cnn


The main idea of ​​this project is to fully comply with the European data protection law shortly GDPR. According to the privacy law, the human image captured by business cameras should not be exposed. Human identification should not be disclosed for commercial reasons. To detect people and make them pixels, we face several challenges, for example, an image size of 500 megapixels and a high signal-to-noise ratio. We use three programming languages ​​to solve this problem, such as java, python and php. We use java for the development of the user interface, Ptython to solve artificial intelligence algorithms and PHP to pixelize images with better quality.

How its work

Several steps are considered to detect the objects from the 500MP images. The steps are the following:

Step 1:

The images are captured from the specific location with very high resolution (500MP) and sent to the WMS server.

Step 2:

This high resolution image is resized up to (50MP to 5MP ) using the Linux bash script. Then, the images are open Java user interface.

Step 3:

In the Java user interface, images are cropped with specific regions to reduce the CPU overhead

Stage 4:

This cropped image is processed through various image processing algorithms, such as contrast enhancement, denosing, etc.

Step 5:

Pre-processed crop images are farther processes through faster recurrent neural network that uses the Python tensorflow and generates detected coordinates.

Step 6:

The detected coordinates are calculated through original image ratio and offset.

Step 7:

Finally, imagemagic is called through Linux bash and write the coordinates in the original image using php.

Key Information
DI Md Sarwar Zahan
Develop for
February 2019
Last Update:
July 2019
JDK 11, FRCNN(Faster Recurrent Neural Network), TensorFlow, numpy, scipy, threading, asyncio, CV2, json, imagemagick, subprocess, os
Responsive frontend
Firefox, Safari, Opera, Chrome, Edge
Object detection
Java, Python, php, bash, javascript, HTML, CSS3