Research On Target Detection Technology Of Surveillance Scene Based On YOLOv2

The intelligent gas station is an important application of intelligent video monitoring.The use of computer vision technology makes gas station management efficient and concise,greatly saving human and material resources.According to the actual demand,this paper designs a vehicle and pedestrian video surveillance system based on YOLOv2 detection algorithm for the detection of vehicles and pedestrians in gas stations.In this paper,designs a Controller Class to solve the data.The main thread is used to collect data to the management class,and YOLOv2 is used to detect image data concurrently,which improves the processing speed of the system.The rationality of the design is verified by field tests,and the data parallel processing system based on YOLOv2 can be applied to other monitoring scenarios.Compared YOLOv2 with Faster R-CNN and SSD,it is concluded that YOLOv2 has good accuracy and detection speed under ambient illumination and target posture changes.,which is suitable for the target detection under this monitoring scene.Using the self-built gas station database to finetune the YOLOv2 network model,the accuracy of vehicle and pedestrian detection is increased by 0.5% and 1.7%.The recall rate is increased by 8.3% and 6.7%.It is proved that the fine tuning has a better optimization effect on the YOLOv2 model in the monitoring scene.The rationality of the design is verified through field tests,and the system can be applied to many other monitoring scenarios.Meanwhile YOLOv2 is loaded into the multi-camera monitoring system.it takes 0.046 seconds to process an 1920×1200 image,and the vehicle and pedestrian detection accuracy is 98.2% and 96.5%,and the recall rate is 93.9% and 92.6%.Therefore,the multi camera monitoring system loaded with YOLOv2 detection algorithm can complete real-time detection of vehicle and pedestrian tasks.