I am working on contour detection based object detection approaches. But all the contour detection techniques are highly influenced by the image noise and the environmental conditions.
I wanted to know is there any robust approaches other than contour detection.
Deep learning based object detection is another approach, but I am expecting some other methods.
For example 'snakes, active contour models'. They are like memory wire and adapt to the contour by gradient descent. Or Hough transform. Originally for finding straight lines, but the idea works also for squares, circles, and arbitrary figures. Or you can find contours by histograms, e.g. by a histogram of directional vectors perpendicular to the outer border of a figure, which you've found by watershed method or with a neural net.
Regards,
Joachim
Dear Ahmed,
You can start by studying this paper
A Review of Supervised Edge Detection Evaluation Methods and an Objective ...
par B Magnier · 2018 · Cité 4 fois · Articles associés
31 mai 2018 · Useful for human visual perception, edge detection remains a ... One of the most challenging goals in contour detection is to operate ... Iqbal et al., Symmetry, 2019.
Dear Joachim Pimiskern Saadane Rachid
Thank you very much for your responses. The environment where I am working is very dynamic in nature. And the object that I am trying to detect would be on constant motion.
Assume a standard number plate detection and recognition problem where a vehicle will be either stable or moving inline with the camera. In my case the number plate like objects moves across the camera. Because of this movement I am getting a little noisy images.
Instead of classifying objects by vectors perpendicular to the border you can also choose a set of pairs of points that lie on the object and determine the distances. It could be advantageous to restrict to lines that completely lie on the object. That can be tested by going stepwise from one point to the other and determine the colors of the intermediate points. In the end you have a histogram of lengths which can be used to identify an object.
Regards,
Joachim
- Badges
- Science topic