Development of a Vision-Based Mobile Robot Navigation System for Golf Balls Detection and Location
A significant challenge in the design of an autonomous mobile robot is the reliable detection of targets, obstacles and targets tracking. Many types of sensor are used for that purposes such as infrared, sonar, vision sensor and laser. Monocular vision is one of the methods used due to simplicity...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2007
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/587/1/600413_fk_2007_11_abstrak_je__dh_pdf_.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | A significant challenge in the design of an autonomous mobile robot is the reliable detection of targets, obstacles and targets tracking. Many types of sensor are used for that purposes such as infrared, sonar, vision sensor and laser. Monocular vision is one of the methods used due to simplicity and computational cost compared to stereo vision. Based on current trends the autonomous mobile robot development, vision sensor is used as different functions such as target recognition, obstacles avoidance, and navigation.
To fulfill such demands the mobile robot should be able to estimate the distance of the detected targets and their angles from its current location. From the extracted information, the motions of the mobile robot can be done efficiently for targets retrieval task. This thesis addresses issue on golf balls localization. The sensor used for localization is a single color webcam. The experiment involves stationary golf balls localization at indoor and outdoor scene.
The objective is to localize golf balls at various locations to be retrieved by the mobile robot. The distance towards the golf balls are estimated based on their diameter. This is based on the perspective view concept where the golf ball sizes are inversely proportional to their distance from webcam. Golf balls detection is done using color segmentation in RGB (red, green and blue) color space. A vector, a, that represents mean value of the target sample is calculated. Then the mean and standard deviation of each color component is calculated. The threshold value lies in the range μ ± σ which represents a square bounding box in RGB color space with a center at a. Every pixel in the test image is tested whether it lies within the bounding box which contributes to target pixel. The technique for segmentation can avoid high computation time for color image processing. The simple features such as diameter, x-y ratio and area are used as its inputs to the k-nearest neighbors (K-NN) classifier.
The software is developed in Visual Basic 6 with a laptop computer acts as a controller and for handling image acquisition and processing. The localization process takes less than one second to be completed. The technique has been tested at indoor and outdoor environment. The efficiency of the estimation is more than 90 percents with a condition that the targets are less than 50 percents occluded.
|
|---|