Image Processing:
I developed many image processing tasks.
Especially using Deep Learning and Machine Learning, develop many object detection, classification,
segmentation, keypoint detection tasks.
In fact, deep learning can carry out several kinds of image processing, but deep learning needs more
processing time and high device’s requirement.
I can develop image processing tasks with short processing time and low device’s requirement, this is
most important.
Bucket Detection using Deep learning in industrial environment.
I developed Bucket’s 4 apex detection project in industrial application.
I can detect many kinds of Buckets in many environments using one application.
I can also classify the bucket kind. Processing time is 100ms on cpu, not gpu.
This task is developed for real-time processing in industrial environment.
Electric Accessories Detection and Classify using Deep learning
There are more then 20 kinds of Electric Accessories.
I can detect , count and classify them.
Directional Object detection
Original Object Detection
Especially, I can detect object and predict its’ direction.
Many Original Object detection algorithms only detect object position, cannot detect direction.
But with predict detection, also can predict object’s size.
This task is developed for real-time processing in industry environment.
Face Detection with KeyPoint using Deep Learning
Processing time for Face Detection with KeyPoint is 10ms on cpu.
With Face KeyPoint, we can develop many Face-related projects, such as face recognition, emotion
recognition, age prediction and so on.
In such projects, KeyPoint position’s accuracy is very important.
And such projects are running on edge device, so fast running time is very important.
This task is developed for biometric recognition.
Input Image
Detect Face Region
Detect Result
Crop Face Region
Detect Face KeyPoint
Car Plate Recognition using Deep Learning
I developed car plate recognition using Deep Learning.
Running time is 100ms/frame on cpu.
In this task, most important is high accuracy and fast real time processing.
I developed light deep learning model and implement fast car plate recognition using deep learning.
This task is developed for fast car plate recognition in edge device.
Fake Hand Detection using Machine Learning
In many image processing tasks, fake detection is a serious problem. Specially, in Biometric
Authentication, it is a very important problem. In many situations, hackers using fake fingerprint or fake
hand to verify Biometric Authentication System.Using Machine Learning and image processing, Classify
fake hands. This task is developed for biometric recognition.
Fake
True
Meter Recognition using Deep Learning
-
Recognition Flow
Input Image
Detection and classification of Meter Region
Detection of the graduation and Needle
-
Recognition Result
Crop Meter Region
In the 1st time, detect meter region and classify kind of meter.
After detection and classification, crop meter region.
Using crop image, detect 2 end points of meter gradation and needle.
Meter Recognition time is 800ms/frame on cpu.
This task is developed for real-time processing in industrial environment.
Fingernail Segmention
Using the segmentation method of deep learning, implement segmentation of fingernail.
After getting segmentation model, transform the model and implement using c++.
Object detection with direction
original object detection
object detection with direction
I developed directional object detection using deep learning.
The weakness of original object detection is that the detection accuracy is poor when many objects are
overlapped.
Using directional object detection, this problem can be solved.