An improved YOLOv5 algorithm for obscured target recognition

Abstract

Target detection is an important problem in computer vision and has important research value in the fields of pedestrian tracking, license plate recognition, and unmanned vehicles [1].The Viola-Jones algorithm is used to detect frontal face images, which improves the speed of face detection by tens or hundreds of times while obtaining the same or even better accuracy, but for special and HOG captures local shape information better and has good invariance to both geometric and optical changes, SVM solves machine learning in small sample cases, but its feature descriptor acquisition process is complex and has high dimensionality, leading to poor real-time performance, and the support vector machine algorithm is difficult to implement for large-scale training samples, while the deep learning-based YOLOv5 target detection algorithm combines the advantages of Viola-Jones algorithm and HOG+SVM algorithm to make up for the shortcomings of the above two algorithms, which is not only very stable for occlusion and complex case processing, but also can be implemented for large regular training samples, but the accuracy of YOLOv5 for target detection is not ideal, and this paper adds SENet mechanism in YOLOv5, which can make the network better to learn the locations in the images that need attention. Therefore, this paper first introduces the traditional target detection algorithm, then introduces and analyzes the Yolov5 algorithm, and improves and optimizes it, and compares it with the traditional target detection algorithm, and the results show that the improved Yolov5 algorithm has better results for target detection.