Nighttime vehicle detection method and system based on image conversion and contour neighborhood difference

Abstract

The invention provides a nighttime vehicle detection method and system based on image conversion and contour neighborhood difference, and the method comprises the steps: carrying out the preprocessingof the image data collected by a camera, and obtaining a brightness block comprising vehicle lamps; screening out a part of non-vehicle-lamp region according to the brightness of the vehicle lamps, the area and the distance between the vehicle lamps; converting a perspective image into an aerial view, detecting a brightness region in the aerial view, and deleting the brightness region from the perspective image; eliminating a brightness block of a reflecting light source according to the difference between the mean gray scale value of contour points and the mean gray scale value of the contour; and finally carrying out the matching of the vehicle lamps of the same vehicle according to the heights of the vehicle lamps and the distance between the vehicle lamps. The invention proposes a low-cost solution based on the characteristics of a common white-light industrial camera, achieves the deletion and selection of the image data through the brightness features and symmetry of the vehiclelamps at night, achieves the further constraint through the difference between the light of the vehicle lamps and the speed and direction of ground reflected light and the difference between the contour of the light source and the brightness change information in a reflection light contour neighborhood region, and achieves the purpose of the vehicle detection.

Description

technical field [0001] The invention relates to the field of traffic information detection in the automatic driving industry, in particular to a nighttime vehicle detection method and system based on image conversion and contour neighborhood differences. Background technique [0002] Vehicle detection is more common in vision applications, especially in security and road camera analysis. In the above-mentioned scenarios, the pose of the camera is generally in a fixed state, but in the vehicle industry, the camera moves with the vehicle, and the pose is constantly changing, which further increases the complexity of the project. Target detection in common ADAS products, such as PD (Pedestrian Detection) and VD (Vehicle Detection), are widely used in daytime scenes, based on the characteristics of people and vehicles such as pedestrian outline gradient features, vehicle bottom shadows, license plates, colors, Based on ground estimation and calculation of Stixel to calculate pe...

AI Technical Summary

Problems solved by technology

In the above scenarios, the general camera pose is in a fixed state, but in the automotive industry, the camera moves with the vehicle, and the pose is constantly changing, which increases the complexity of the project.

Target detection in common ADAS products, such as PD (Pedestrian Detection) and VD (Vehicle Detection), are widely used in daytime scenes, based on the characteristics of people and vehicles such as pedestrian outline gradient features, vehicle bottom shadows, license plates, colors, Based on ground estimation and calculation of Stixel to calculate pedestrians and vehicles, etc., the target features are relatively rich, but in night scenes, the ambient light is not very ideal. On urban roads, there is very little feature information that can be used for vehicle detection. The influence of the vehicle has caused a lot of interference to the vehicle detection and recognition, how to distinguish the vehicle from the traffic flow is not a small difficulty

Many solutions use special equipment such as infrared night vision devices to detect pedestrian heat, but the cost of this solution (only the sensor) is about 40,000 to 60,000, which is easily affected by the environment and is not suitable for mass production.

Images