The purpose of this research is to investigate novel object detection and localization algorithms for autonomous robots. The average of overlaps and the percentage of positive detections are calculated to compare the accuracy between the Scale Invariant Feature Transform and the Speeded-Up Robust Features. Dense features are used as additional features to increase the precision in detection. Features are clustered into visual words to form a histogram. Images are trained based on the histogram and their corresponding labels with Support Vector Machine. The bounding box with the highest output from the Support Vector Machine represents the location of a target class. |