| Autonomous driving technology has reshaped people’s cognition of travel and has great social and economic value.Vehicle detection,one of the basic directions of autonomous driving technology,provides technical support for the corresponding anti-collision warning system,and becomes the current research hotspot.However,due to low brightness,complex road conditions and obscure vehicle body contour in night,the existing vehicle detection methods have some limitations.To solve this problem,this paper proposes two kinds of vision-based nighttime vehicle detection methods: the algorithm based on taillight pairing and the deep learning algorithm combining Mobile Net v2 and YOLO v3.In the process of driving at night,it can effectively assist drivers to timely perceive the vehicles on the road and pay attention to the potential collision risk,which gurantees driving safety.The main research contents are as follows:(1)Aiming at the problem that the current nighttime vehicle detection method based on taillight is prone to misdetection when detecting parallel vehicles,a taillight pairing method is designed,and a nighttime vehicle detection method combining light pairing and feature fusion is proposed.The proposed taillight pairing method is used to determine the potential area of the vehicle,and then the multiple features of the area are extracted.After that,the trained SVM classifier is used to verify the potential area of the vehicle.Finally,the experiment proves the proposed detection method can better complete the nighttime vehicle detection task.But this method has some limitations in the complex scene,due to it relies on prior knowledge and artificial features.(2)In order to solve the problem that the detection method based on taillight has insufficient detection rate in complex scenes,a deep learning based nighttime vehicle detection algorithm combining Mobile Net v2 and YOLO v3 is proposed.Considering the requirement of real-time performance,a lightweight network called Mobile Net v2 is used as the backbone of feature extraction.At the same time,K-means clustering algorithm is used to recluster the dataset in this paper to get the appropriate size of vehicle anchor boxes.In addition,EIo U loss function is used as the location loss to optimize the model.Verification on public data set and personal test set shows that the algorithm can realize effective vehicle detection in complex night scenes.In this paper,two vision-based nighttime vehicle detection algorithms composed of traditional and deep learning approaches are proposed.The problem of wrong pairing of parallel vehicle lights in vehicle detection based on taillight at night can be effectively solved.The detection method based on deep learning can overcome the problems that traditional vision-based methods rely on prior knowledge and artificial features,and achieves an average precision of 88.11% and a detection speed of 23 frames per second.The research results of this paper lay a foundation for the realization of anti-collision warning system in intelligent driving at night. |
