A method for improved pedestrian gesture recognition in self-driving cars
Ying Zheng, Hong Bao, Cheng Xu · Australian Journal of Mechanical Engineering · 2018
In the pursuit of self-driving vehicles, pedestrian recognition plays an integral role. The following work proposes a pedestrian gesture recognition method based on a k-nearest neighbour algorithm combined with pyramid residual module to reduce computation and improve real-time performance of gesture recognition. The method was formulated using a data from Udacity, and subsequently compared with other recognition methods. The experimental results showed that the accuracy of the new method was as high as 92%, which is an improvement over the conventional histogram of oriented gradient method. The proposed method was further verified by other indicators, demonstrating its robustness and generality.