Research On Pedestrian Trajectory Detection Method Applied To Assisted Driving

In order to improve the safety of road traffic,automobile safe driving has become one of the important directions of the development of automobile technology.Advanced driver assistance systems can help drivers improve driving safety by providing functions such as lane keeping,automatic emergency braking and blind spot monitoring.Although automobile assisted driving technology has been developed for a long time,it is still not perfect and cannot be well adapted to all types of roads.It is more effective to improve the safety of assisted driving than to estimate complex external environmental factors by warning in advance and paying attention to the movement direction and trend of external pedestrians.In order to more accurately estimate the movement behavior of pedestrians outside the car,this thesis adopts and builds a set of pedestrian trajectory detection system,so as to effectively improve the safety of assisted driving and reduce the probability of accidents.The details are as follows:First,aiming at the problem that the detection rate and accuracy of pedestrian detection are both low in the complex scene where pedestrians walk in disorder and block each other,this thesis adopts a double anchor frame target detection and recognition algorithm based on information fusion.This algorithm detects the information of each person’s body and face at the same time,and determines the binding relationship to improve the low recall rate of human body in complex scenes.Second,aiming at the problem of unclear features in the complex scene where the same pedestrian moves in disorder and blocks each other,this thesis adopts a feature extraction algorithm based on the fusion of clothing and head information.The algorithm extracts features in blocks from the human body detected in the first stage,fuses features of different areas of the human body and integrates contextual information.Moreover,the attention mechanism is added to the face,head and shoulder area,which makes the model pay more attention to the features of face and head and shoulder,so as to improve the accuracy of blocking pedestrians’ rerecognition.Third,aiming at the low accuracy of determining the relative relationship between pedestrians and vehicles in complex scenes,this thesis uses the stereo matching algorithm PSMNet for position estimation.By adding the spatial pyramid model,the algorithm improves the sensitivity field of the network model and expands the acquisition of context information.Finally,in order to improve the model’s control of global information,the global relationship modeling method is adopted to further obtain the global relationship between different pixels.Finally,based on the three algorithms used above,this thesis designed and built a set of pedestrian trajectory detection system,which realized real-time analysis of pedestrian status in complex scenes,accurate judgment of pedestrian trajectory,and estimation of the distance between vehicles and people.From the algorithm level,the detection system of pedestrian trajectory is mainly divided into pedestrian detection module,pedestrian weight recognition module,person-vehicle distance estimation module and pedestrian trajectory analysis module,so as to facilitate the subsequent improvement of various algorithm modules.The pedestrian trajectory detection system has the characteristics of low coupling,and the selection of corresponding computing resources is carried out for different algorithm modules.