OBJECT DISTANCE ESTIMATION WITH LOW-COST HETEROGENY STEREOVISION CAMERA BASED ON MASK R-CNN CONTOUR CENTROID METHOD
Camera is commonly used for computer as tools to see. The critical component in computer vision is image, which represents a scene into array of pixels that contain RGB and alpha every pixel. Stereovision is often used to estimate object’s distance, however, using low-cost camera hinders the a...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/58014 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Camera is commonly used for computer as tools to see. The critical component in
computer vision is image, which represents a scene into array of pixels that contain
RGB and alpha every pixel. Stereovision is often used to estimate object’s distance,
however, using low-cost camera hinders the accuracy and precision of the distance
estimation. Sometimes the low-cost cameras that we used are also unidentical and
contain less to none specs information so it’s quite hard to gain information about
focal length. In this thesis, the author will develop a method to estimate object’s
distance with low-cost heterogeny stereovision camera based on Mask R-CNN.
Mask R-CNN not only detects object type and position, but also predicts an object
shape / mask (Instance Segmentation) therefore the center point of the object could
be determined using the mask’s shape. This method is tested with 3 objects inside a
frame: suitcase (220 cm away), car (280 cm away), and books (150 cm away). The
object’s distance results are 220.98 cm ? 7.52 cm for suitcase, 253.75 cm ? 9.85
cm for suitcase and 148.61 cm ? 4.69cm for books. This results in accuracy of
99.56%, 90.63% and 99.07% for suitcase, car, and books respecctively
|
|---|