Unconscious Behavior Detection For Pedestrian Safety

Pedestrian abnormal behavior detection is an important research topic in the field of pattern recognition,and has a broad application prospect in the field of intelligent transportation.At present,abnormal behavior detection is divided into three research directions,which are based on basic human posture,smart home and crowd-based abnormal behavior detection.We propose a new research direction for pedestrian abnormal behavior detection,which is called unconscious behavior detection.This behavior refers to pedestrian distraction caused by the use of electronic devices.In the process of occurrence of distraction behavior,it will be affected by pedestrians themselves or external factors,so it brings great challenges to behavior detection.Designing an efficient pedestrian distraction behavior detection algorithm is still a challenging task.As an intelligent portable tool,mobile phones play an important role in personal work and life,which leads to the use of mobile phones as one of the main reasons for pedestrians'distraction.According to the survey results,there are more and more pedestrians who play mobile phones on all occasions,which brings many traffic problems and attracts great attention from all walks of life.However,the current research on pedestrian distraction behavior is still in its infancy,so this paper studies pedestrian distraction behavior and puts forward corresponding solutions to this problem.The main work of this paper is as follows:Firstly,the pedestrian unconscious behavior data set(PWUM)is constructed,and the image in the data set is selected,annotated and processed for pedestrian privacy protection.A new pedestrian distraction behavior is added to the existing pedestrian data set,which can be used by researchers in related fields.Secondly,a pedestrian unconscious behavior detection algorithm is proposed.The main work includes pedestrian detection based on gradient and texture feature ensemble method,pedestrian sensitive parts extraction based on selective search method,and pedestrian classification based on feature ensemble method.We have done a lot of experiments on the proposed algorithm,and the experimental results verify the rationality of the algorithm framework and the effectiveness of the algorithm.Thirdly,a multi-feature pedestrian distraction detection algorithm based on convolution neural network is proposed.In this algorithm,we propose a G-VGGNet network model,which can be used to solve the problem that global features are easily ignored in CNN.The main work includes extracting global feature descriptors,constructing G-VGGNet model,training network through Finetune process and predicting pedestrian classification results.The experimental results show that the classification effect of the algorithm is better.