Vehicle detection method based on improved YOLOv3

Abstract

The invention belongs to the field of deep learning and intelligent vehicle road detection, and specifically relates to a vehicle detection method based on improved YOLOv3. The method comprises the following steps: redesigning a convolutional neural network structure between a Darknet layer and three yolo layers, and designing a YOLO-TN network by referring to an idea of TridentNet weight sharing;carrying out model pruning on the YOLO-TN convolutional neural network; constructing a vehicle detection data set, and labeling vehicle position information in the data set; respectively training vehicle detection models based on the YOLO-TN and the YOLOv3, completing a vehicle detection task, and comparing detection results of the YOLO-TN and the YOLOv3. The method has high average precision while guaranteeing real-time detection, and is lower in omission ratio and more accurate in positioning when used for detecting distant vehicles and small-scale targets.

Description

technical field [0001] The invention belongs to the field of deep learning and intelligent vehicle road detection, and in particular relates to a vehicle detection method based on improved YOLOv3. Background technique [0002] In recent years, with the continuous improvement and development of artificial intelligence (AI) theory and technology, Advanced Driving Assistant System (ADAS) has played a pivotal role in the development of the automotive industry, and major car companies have also It is gradually transitioning from traditional cars to smart cars. [0003] Target detection is one of the research hotspots and difficulties of smart cars. Most of the traditional target detection methods are based on manual features, such as the detection method based on HOG features, which traverse the entire image through a sliding window (Sliding Window) to obtain HOG features. Input the extracted features into the SVM classifier for target detection. This method can achieve good res...

AI Technical Summary

Problems solved by technology

This method can obtain better detection results when detecting pedestrians, but there are also some shortcomings: (1) Only one fixed-scale candidate frame is used. If multi-label classification is required, the detection effect and accuracy will deteriorate, and the generalization The ability is poor; (2) Although modifying the aspect ratio of the candidate window to 1:2 can obtain more accurate positioning, if the number and diversity of samples cannot be guaranteed, the false detection rate will increase; (3) when detecting distant targets and small-scale targets, the missed detection rate will increase

However, this method still cannot obtain better detection results when detecting small targets.

Images