Publication: Optimisation of object detection and measurement using stereo vision
Loading...
Date
2021-07-01
Authors
Jairan, Muhd.Hafizrah
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In the detection and measurement of objects, camera sensors can be a powerful sensor that can replenish infrared (IR) and ultrasonic sensors. The computer vision sensor
may provide additional knowledge about the environment, such as the distance between an object and its features. Stereo vision is being selected and implemented in this project
for an application that requires both object detection and measuring functions. Three stereo matching algorithms are examined and optimised to find the optimal balance of disparity map refinement for object detection and measurement. Normal Matching appears to provide the best results for object detection as compared to Semi-Global and Block Matching since it provides nearly all descriptions of the object in the screen. Thus, chosen as the stereo matching algorithm for this project. The optimised equations for the width, height and depth measurement were created through the best fit line method. The performance of the object detection and measurement were analysed. The object detection using the image thresholding gave good results although some of the detection had problem either over detect at the width or the height of the object. In term of the object measurement, the height and width measurement had percent of error less than 10% while the width measurement had the higher percent of error for 10.10%. The processing time for the object detection and measurement only took less than 3% from the overall processing time for the system. Most of the time was taken from the image acquisition and disparity mapping.