基于并行局部特征金字塔的弱势道路使用者检测

VULNERABLE ROAD USERS DETECTION BASED ON PARALLEL LOCAL FEATURES PYRAMID NETWORK

针对现实环境中弱势道路使用者(Vulnerable Road Users,VRU)检测准确率低的问题,提出并行局部特征金字塔检测模型。该模型以SSD(Single Shot Multibox Detector)为基础,针对SSD中默认框过大不能涵盖小尺度VRU的问题,重新设计默认框来匹配更多尺度的VRU;使用两阶段回归策略解决正负样本数不平衡的问题;通过并行的局部特征金字塔结构将高层的语义特征融合到底层特征中来增强模型对VRU的检测能力。相比SSD网络,提出方法的mAP(mean Average Precision)提升了7%,且检测速度满足实时性要求。由此验证了该模型对VRU检测的有效性。

A parallel local features pyramid model is proposed to solve the problem of low detection accuracy of VRU(Vulnerable Road Users)in real environment.This model is on the basis of SSD(Single Shot Multibox Detector).In order to solve the problem that the default box in SSD is too large to cover small scale VRU,the default box was redesigned to match more scale VRU;two-stage regression strategy was used to solve the problem of imbalance between positive and negative sample;through the parallel local features pyramid structure,the semantic features of the higher level were integrated into the lower level to enhance the detection ability of the model to the VRU.Compared with SSD network,the mAP(mean Average Precision)of this method is improved by 7%and the detection speed satisfies the real-time requirement.The validity of this model for VRU detection is verified.