Night traffic flow detection method based on headlamp trajectory tracking and dynamic matching

Abstract

The invention provides a night traffic flow detection method based on headlamp trajectory tracking and dynamic matching. The method comprises the steps of: video collection; parameter input and calibration; static extraction: through preprocessing, acquiring image brightness information and segmenting out all light spots and preliminarily screening out headlamp light spots based on geometric features; dynamic filtering: tracking the light spots obtained through static extraction and removing disturbing light sources based on the geometric features in the tracking process: vehicle lamp matching: smoothing original tracking trajectories and performing vehicle lamp matching based on smoothed trajectory feature similarity and trajectory point space position relationships; vehicle counting: when vehicle lamps are tracked to a traffic flow detection line position, counting vehicles according to a vehicle lamp matching result; traffic information output. The method extracts headlamp light spots according to the time domain stability of the brightness feature and the geometric features of headlamps at night and performs vehicle lamp matching based on vehicle lamp tracking trajectories, thereby being capable of counting vehicles in complicated night road scenarios.

Description

technical field [0001] The invention relates to the technical fields of computer vision, image processing and intelligent traffic data collection, in particular to a nighttime traffic flow detection method based on headlight trajectory tracking and dynamic matching. Background technique [0002] Traffic flow data acquisition is the basis and important link of intelligent transportation system (ITS). Video-based detection technology is compared with traditional traffic flow data acquisition methods, such as geomagnetic induction coil detection, ultrasonic detection, infrared detection, manual counting, etc. , has the following advantages: 1. Easy to install, maintain and upgrade the system; 2. The purchase cost of testing equipment is low; 3. It can provide real-time traffic scene image data on site, or reproduce past traffic scenes; 4. Environmentally friendly; 5. No interference between devices. Therefore, the traffic data acquisition method based on computer vision has a ...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the disadvantage that existing traffic flow video detection methods are difficult to apply to complex night road scenes, the present invention provides a nighttime traffic flow detection method based on headlight trajectory tracking and dynamic matching, using headlights as the basis for identifying the existence of vehicles at night , make full use of the time domain stability of the headlight geometric features to accurately extract the headlight spots, and then realize the headlight pairing based on the similarity of the headlight tracking trajectory features and the spatial position relationship of the headlights, and finally achieve the vehicle counting target in the complex road scene at night

Images