Full-time vehicle detection and tracking method and system in video based on Kalman filter and deep learning

Abstract

The present invention provides a full-time vehicle detection and tracking method in video based on Kalman filtering and deep learning. The method first trains the deep learning CNN model offline, and establishes vehicle daytime and nighttime infrared classifiers. Secondly, the system collects a frame of image and detects the vehicle: during the day, the moving and stationary targets are obtained by the moving and static target detection method, and the vehicles are separated by the daytime classifier; at night, the car light method is used to detect the position of the car light, and the night classifier is used to detect the vehicle position. Then, a Kalman filter is initialized for each detected vehicle object. Then in the next frame, detect the vehicle position in the current frame, and use the Kalman filter of the vehicle target in the previous frame to predict the position; compare the detected position with the predicted position, and then compare the vehicle target in the previous frame with the vehicle position in the current frame Connect to get the tracked path of the vehicle. Finally, the system draws the vehicle's path through successive frames in the video.

Description

technical field [0001] The invention belongs to the fields of image recognition, video analysis and vehicle recognition, and in particular relates to a full-time vehicle detection and tracking method and system in video based on Kalman filtering and deep learning. Background technique [0002] With the development of society and the continuous improvement of people's living standards, automobiles have gradually become popular. But what follows is traffic accidents, especially the trend of vicious traffic accidents continues to rise. In recent years, the growth rate of vehicles has been far higher than that of roads and other traffic facilities. Therefore, traffic accidents continue to occur, casualties are increasing day by day, and property losses are serious. The main causes of traffic accidents are speeding, lane-occupied driving, drunk driving, and fatigue driving. Therefore, for the traffic management department, video surveillance is used to monitor the running statu...

AI Technical Summary

Problems solved by technology

[0013] Patent CN201220337489 "Intelligent traffic control device based on image processing" is based on the method of motion detection. When the vehicle stops, the vehicle will blend into the background. If the motion method cannot detect the vehicle, it cannot be tracked; at the same time, it does not describe the situation of night detection

[0014] Patent CN201210357874 "A Vehicle Tracking Method and System" is a method based on feature detection. This method traverses the entire image or a partial image. The search strategy requires a large amount of calculation, and the target features cannot be updated in time. Changes in posture can easily lead to tracking loss , at the same time, due to occlusion, error elimination, etc., the temporarily lost target cannot continue to track; at the same time, the situation of night detection is not described

Images