Research On Intelligent Video Surveillance Algorithm Of Escalator And Implementation Of Jetson TK1

Video surveillance technology has always played an important role in the field of public administration and decision-making due to its accuracy,intuition and low cost.Escalator is one of the main applications of this kind of technology.In recent years,with the rapid development of technologies such as artificial intelligence and embedded,the video surveillance system that takes intelligence and efficiency into consideration has gradually become possible.Video surveillance algorithm in the escalator scene was studied and a surveillance system based on Jetson TK1 was designed and implemented.By using three USB cameras with different shooting angles,the floor board and handrails were monitored.Pedestrian and object were detected and tracked in order to achieve pedestrian retrograde detection,passenger flow statistics,congestion detection and object retention detection.At the same time passenger presence detection on the escalator as well as foreign body detection on the handrail were achieved.The main work of this article includes four aspects:1)Aiming at the problem of target segmentation and multi-scale that cannot be solved in the traditional pedestrian detection algorithms,the heads overlooked data set and pedestrians overlooked data set were first established in the escalator scene.And then the HOG features were extracted to train Adaboost cascade classifiers,while classifier parameters suitable for escalator scenes were selected in the control experiment.Finally the sliding window strategy was used to complete the pedestrian detection.2)Aiming at the problem of target renewal that is difficult to solve in existing target tracking algorithms,a tracking strength model for moving targets was first proposed,which can adaptively add new targets and discard the outdated targets that should be deleted.And then the minimum distance strategy was used to match the observation sequence and tracking sequence.Finally the Kalman filter was used to complete the pedestrian tracking and object tracking.3)Aiming at the problem of excessive computation in existing behavioral analysis algorithms,the simple criteria of motion speed and tracking strength were first used to complete the floor board pedestrian retrograde detection,passenger flow statistics and congestion detection,as well as the escalator passenger presence detection.And then a fast background updating strategy based on the foreground mask was proposed,which was combined with the pedestrian tracking to complete the floor board object retention detection.Finally a small area monitoring method based on the Gaussian mixture model was proposed to complete the handrail foreign body detection.4)Aiming at the problem of hardware implementation of the system,Jetson TK1 was chosen as the embedded hardware platform,while Qt cross-platform development framework was adopted to design the software interface.OpenCV open source visual library was adopted to design the algorithm,and CAN bus was used to communicate with the escalator BAS controller.The running state of the escalator will be adjusted according to the surveillance result,which realizes intelligent control.The experimental results on industrial testing of the cooperative enterprise show that the system can track pedestrians and objects accurately and stably.It can accomplish the above six surveillance tasks.The proposed system meets the intelligent video surveillance system robustness,real-time and accuracy requirements,which has important engineering application value.