Gesture Recognition For Traffic Control Based On Thinning Algorithm And Template Matching

With the rapid development of social economic, traffic congestion problem has become globalized. Gesture recognition for traffic control is able to relieve traffic problems caused by weather condition and also it means a lot to driverless vehicles. As the computer techonology develops, traffic control gesture recognition is becoming an active research area because of its important research value. This paper focuses on vision-based gesture recognition for traffic control.Some related knowledge of human action recognition is outlined, as well as some traditional thinning algorithms and template matching methods. What's more, a new skeletonization algorithm based on contour-deleting and a new template matching method based on weighted Hausdorff distance is presented, as well as its application on gesture recognition for traffic control. The result shows that using this approach to recognize traffic control gesture is feasible. The main content includes the follows:(1) With the regard of image problem, a pretreatment is added before the recognition happens, this pretreatment contains: Gray processing; Background differencing; Binarization; Digital morphological close operation and Normalization. This pretreatment is able to create standard gesture image.(2) Traffic control gesture is a set of gestures with obvious periodicity, although the duration of each gesture is random. In order to get rid of the interference caused by redundant frames, a step is added to select key frames from the video sequence by calculating the numbers of contour pixels for every selected frame. This is able to improve the efficiency of recognition.(3) After analyzing the topology of traffic control gesture, a skeletonization algorithm based on contour-deleting is presented. First, extract the contour of digital image and narrow the scope of thinning target from the entire image down to the contour; then delete non-skeleton pixels from the contour continuously using a method called 8-neighbors determination until the skeleton is extracted. This skeletonization algorithm improves the efficiency of thinning algorithm and is able to maintain the connectivity of the skeleton.(4) After analyzing the characters of different traffic control gestures, a template matching method using weighted Hausdorff distance is presented. This method assigns different weights to different areas and calculates all the distances between key frames and template frames, and takes the template with the smallest distance as the matching one.The result shows that using this approach to recognize traffic control gesture is feasible and has its important research value and broad application prospect.