江苏大学学报（自然科学版）2026，Vol.47Issue(2)：151-157,7.DOI:10.3969/j.issn.1671-7775.2026.02.004
基于LSTM网络的前车行为识别
Recognition of front vehicle behavior based on LSTM
摘要
Abstract
To solve the problems of traditional method with poor practicability and low recognition rate for the front vehicle behavior,the new recognition method based on LSTM network model was proposed.In the process of vehicle detection,due to the difficulty in extracting features of vehicle targets and occluded targets in remote scenes,the phenomenon of wrong detection and missing detection occurred.The vehicle detection algorithm YOLOv5-BA improved by YOLOv5 in complex background was adopted.The feature fusion idea of weighted bidirectional feature pyramid network(BiFPN)was introduced,and the adaptive feature fusion module ASFF was introduced in the detection part to improve the detection performance.The results show that the high detection accuracy of the improved algorithm reaches 97.3％on KITTI data set.On this basis,combined with DeepSort algorithm,the vehicle detection and tracking are realized.By the forward vehicle behavior recognition model based on LSTM network,the average accuracy of the model on the data set is 92.3％,which can be used in the practical application of vehicle behavior recognition.关键词
目标检测/目标跟踪/行为识别/YOLOv5/LSTMKey words
object detection/target tracking/behavior recognition/YOLOv5/LSTM分类
信息技术与安全科学引用本文复制引用
朱宝全,马长旺,赵强,唐佳乐..基于LSTM网络的前车行为识别[J].江苏大学学报（自然科学版）,2026,47(2):151-157,7.基金项目
国家重点研发计划项目(2017YFC0803901) （2017YFC0803901）
中央高校基本科研业务费专项资金资助项目(2572016CB18) （2572016CB18）
