A pedestrian detection method

Abstract

The invention provides a method for detecting pedestrians, which adopts an obliquely installed fisheye camera to collect images, and ensures that pedestrians can be detected at an optimal monitoring angle without fisheye distortion correction. The obtained sample set is used for training, the aggregated channel features are extracted by the ACF method, and the classifier is trained by the soft cascade AdaBoost method, which can improve the accuracy of pedestrian statistical features without fisheye distortion correction; in the detection In the process, after quickly extracting 8 features, use a fixed-size pyramid aggregation template to aggregate 8 feature images to form pyramid aggregation channel features, which can avoid using all pyramid channel features, thereby reducing the amount of computing data. Improve the calculation speed, and finally merge the overlapping candidate windows through the greedy non-maximum value suppression method to improve the robustness and accuracy of pedestrian detection.

Description

technical field [0001] The invention relates to the technical field of visual detection, in particular to a pedestrian detection method. Background technique [0002] With the rapid development of related technologies such as computer vision, security intelligent monitoring and security fields, there are more and more occasions that require intelligent monitoring. In the current traditional monitoring technology, the monitoring equipment with ordinary lens generally has a small imaging field of view and a limited monitoring range. Multiple monitoring devices are needed to solve the problem of video blind spots, which increases the cost and also increases the complexity of the system (such as : The amount of video data brought by multiple monitoring devices has increased sharply, and too many monitoring devices affect the appearance, and are not conducive to installation, deployment and management, etc.); and although the fisheye lens has a large imaging field of view, it als...

AI Technical Summary

Problems solved by technology

However, this method uses ordinary cameras, the range of image acquisition is limited, and blind spots are prone to appear, and a large amount of calculations are required in the detection process. Although the idea of ​​the integral map speeds up the feature extraction, it still cannot meet the detection speed requirements.

Images