Human motion segmentation and recognition using machine vision for mechanical assembly operation

The observation, decomposition and record of motion are usually accomplished through artificial means during the process of motion analysis. This method not only has a heavy workload, its efficiency is also very low. To solve this problem, this paper proposes a novel method to segment and recognize continuous human motion automatically based on machine vision for mechanical assembly operation. First, the content-based dynamic key frame extraction technology was utilized to extract key frames from video stream, and then automatic segmentation of action was implemented. Further, the SIFT feature points of the region of interest (ROIs) were extracted, on the basis of which the characteristic vector of the key frame was derived. The feature vector can be used not only to represent the characteristic of motion, but also to describe the connection between motion and environment. Finally, the classifier is constructed based on support vector machine (SVM) to classify feature vectors, and the type of therblig is identified according to the classification results. Our approach enables robust therblig recognition in challenging situations (such as changing of light intensity, dynamic backgrounds) and allows automatic segmentation of motion sequences. Experimental results demonstrate that our approach achieves recognition rates of 96.00 % on sample video which captured on the assembly line.