Vehicle-mounted monocular camera-based vehicle detection and location method

Abstract

The invention discloses a vehicle-mounted monocular camera-based vehicle detection and location method. The method comprises the following steps of: S1, obtaining front view information from a monocular camera mounted on a running vehicle; S2, determining a target frame image corresponding to the front view information according to a preset ROI area; S3, inputting the target frame image into a classifier to carry out detection, and determining whether an image in a the target frame is a vehicle or not; S4, when the image in the target frame is the vehicle, carrying out regressive correction onthe target frame; and S5, displaying the target frame after the regressive correction, and determining the position of the front vehicle according to the target frame. By adoption of the method, thevehicle location correctness is improved.

Description

technical field [0001] The invention relates to the technical field of vehicles, in particular to a vehicle detection and positioning method based on a vehicle-mounted monocular camera. Background technique [0002] At present, edge information is usually used to detect vehicle targets. In complex urban road conditions, targets are sometimes serialized together with the background, resulting in missed detection. At the same time, it is used to complicate the background, resulting in a large number of false detections. Another method uses The cascade classifier detects the target, but because the classifier is trained in the same proportion of length and width during training, the length and width of the final detected target frame are also in the same proportion, but because in actual places, such as cars, SUVs, vans, etc. are not Equal proportions exist, and the same scale is not the same under different viewing angles of the same type of vehicle. Eventually, although the t...

AI Technical Summary

Problems solved by technology

[0002] At present, edge information is usually used to detect vehicle targets. In complex urban road conditions, targets are sometimes serialized together with the background, resulting in missed detection. At the same time, it is used to complicate the background, resulting in a large number of false detections. Another method uses The cascade classifier detects the target, but because the classifier is trained in the same proportion of length and width during training, the length and width of the final detected target frame are also in the same proportion, but because in actual places, such as cars, SUVs, vans, etc. are not Equal proportions exist, and the same scale is not the same under different viewing angles of the same type of vehicle. Eventually, although the target frame can be detected, it is not at the exact position. The red is the actual position of the rear of the vehicle, and the blue is the detection frame. 1:1 ratio detection position

At the same time, many existing methods do not consider the vehicle context in the video sequence, resulting in the detection results of the same target being extremely unstable in the video sequence.

Images