Microsoft Kinect based real-time segmentation and recognition for human activity learning
MetadataShow full item record
Lower body pain and injury have become common in this technical world, especially in elderly people. It is quite difficult to recover from these injuries leading to problems in performing daily routine activities like walking, running, sitting etc. Although there are many activity recognition models present today, there has been relatively little multiple activities recognition study of lower limbs. Most of the previous researchers focused on single activity recognition using various machine learning algorithms. Researchers have evolved with the learning of gait using different methods and techniques for upper and lower body using the sensors and different camera systems. This research has two main sections, one is for segmenting the motion and another is recognizing those movements. In this research, multiple activities were performed by the patients in a random manner without stopping and these activities were recognized in different groups stating the performed activity if the part of the multiple activities is walking or running or leg raising activity. The first goal of this dissertation is to plot the human gaits as a skeleton using MATLAB with a camera sensor second goal is to segment those derived gaits using the on-line aligned cluster analysis and dynamic time alignment kernel method and the last goal is to recognize the segmented gaits using the support vector machine algorithm. This is done by tracking and learning the person’s lower limb data points and finding the exact action performed by a dynamic time alignment Kernel method for segmentation and comparison of different algorithms like Support Vector Machine, K-Nearest neighbors for recognition. The experimental results collected in this research show that the Support Vector Machine performs higher recognition accuracy.
Institutional Repository URIhttp://hdl.handle.net/10657.1/825