Simulated kalman filter (SKF) based image template matching for distance measurement by using stereo vision system
2018
Distance measurement sensor such as Kinect and Laser Range Finder (LRF) are already been implemented in industrial applications. One of the important issues for sensors is the accuracy of distance measurement. This thesis explains the development of new algorithm for distance measurement using stereo vision sensor. Stereo vision sensor consists of two stereo cameras, mounted parallel in stationary position. Stereo vision sensor can provide color and texture information for easy data and feature extraction. Based on literature, stereo algorithm is already being implemented to solve the distance measurement problem. Stereo algorithm consists of camera calibration, stereo mapping (or disparity mapping) and 3D point cloud data. From the algorithm, disparity mapping algorithm such as Semi-global block algorithm is found out to be inaccurate and complex. This research introduces a new approach to measure distance by using stereo vision systems. The new approach is by using image template matching application which is by finding one best matching pixel from stereo images, and then, calculate the distance between two pixels using stereo depth equation. Based on this approach, searching the best pixel in the images is a challenging task. It is because the process need high memory and expensive computational time when dealing with image pixels. As reported in literature, conventional algorithm for image template matching such as correlation between two images took very long processing time. That is why, image template matching problem is now considered as an optimization problem. By implementing optimization algorithm in image template matching, it is expected that the computation time can be reduced. In addition, it is expected that it can be applied in real-time application. In this study, Simulated Kalman Filter (SKF) is applied to image template matching application as the optimization algorithm. SKF is compared with conventional algorithms for image template matching which are performance index value (PIM) and correlation by using DC components of image (TMC) and by using power of images (TMP) methods. The findings showed that computational time for SKF is lower than others, which is 1.5 seconds within 25 runs. Meanwhile, the computational time for PIM, TMC and TMP methods are 2.0, 2.2 and 3.3 seconds respectively. After that, SKF is tested to find the most accurate image template matching and compared with Particle Swarm Optimization (PSO) and Bat Algorithm with Mutation (BAM). The result obtained is 40% successful image matching for SKF compared with PSO and BAM which are only 12% and 20% respectively. In addition, to ensure the robustness of SKF algorithm, the algorithm is tested under vision problems, occlusion and illumination-invariant. For both problems, SKF showed the good performance in correct image matching compared to PSO and BAM with the average successful image matching result of all cases for SKF is 15.2% for illumination-invariant problem and 33.33% for occlusion. The next experiment is the application of image template matching for distance measurement using stereo vision system. SKF is compared with PSO, BAM, stereo algorithm for stereo vision system and ground truth data for distance measurement of 24 different cases. Each case involved different distance between stereo camera and interested object in the image. The result shows that the accuracy of estimate error model, SKF, PSO and BAM are 83.50%, 87.36%, 61.31% and 34.00%, respectively respect to the ground truth value. The highest accuracy is by using SKF compared to other methods. Therefore, the new approach for distance measurement by using SKF based image template matching on stereo vision system is accurate, efficient and robust.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI