Bus Passenger Flow Collection System Based On TOF Camera

Artificial intelligence industry has developed rapidly in China in recent years,computer vision is a very important application field of artificial intel igence and has become the new darling of the times,more and more researchers have invested a lot of time to study the technology,it promotes the development of all walks of life.many people began to flood into cities with the acceleration of urbanization,intelligent transportatio n facilities have also become a research hotspot as an important way for people to travel.Bus passenger flow information has attracted many scholars’ research as an important information of traffic intelligence big data.Traditional methods of passenger flow statistics generally use color cameras for passenger flow analysis,however,RGB images captured by color cameras are heavily affected by light,especially at night and dusk,when it is particularly difficult to capture clear RBG images during this time.this paper designs a bus passenger flow counting system based on time of flight technology and depth camera in order to solve the above problems,this paper includes three parts: passenger target detection,passenger target tracking and passenger target counting.In the aspect of passenger detection,a maximum stable extreme region algorit hm based on gradient hierarchy is proposed based on the feature of depth image,the algorit hm accurately divides the head area of the human body from the area of the passenger’s human body,and use the classifier trained by linear support vector machine to make decision,the characteristic region of human head is identified by this method.In the aspect of passenger target tracking and counting,this paper selects the passenger tracking algorithm based on the center of gravity of human head to monitor passengers,and proposes a counting method based on two counting lines and three counting states,which solves the crowd congestion,It solves the repeated counting problem caused by the same passenger repeatedly triggering the counting line,and greatly reduces the counting error.Finally,through the analysis of a lot of bus scenes,it is proved that the system has high recognition rate and strong robustness.