Research On Pedestrian Pose Recognition Algorithm Based On Intelligent Driving

The intelligent automobile has become an inexorable trend.Furthermore,regarding the safety of self-driving vehicles,the high precision sensing of the pedestrians in the driving environment is valued as the key element,which is also an essential guarantee for automatic driving.It involves multiple concerns,including the high precision recognition of the pedestrians’ appearances,the complexity of identifying the posture forms of multi-person,the complicated background,the speed of discernment,and the expense calculation.According to the algorithms developed explicitly for the pedestrian appearance detection network adapted to the autopilot under complex surroundings,this article proposes the multitude Pose Detection-Net,known as MEPDN,which is a combination of the Multi-enhance neural network model and the Deep Resolution-Net,to improve the high-speed identification of the appearance of multiple people in a complicated environment,addressing two problems,the low accuracy of multitude pose recognition in driving conditions and the slow speed of detection.Key objectives and innovative points of the paper:(1)Propose the MEPDN to address the difficulty of achieving high accuracy and high speed multi-person appearance identification in a highly complex background while the vehicle is running.The MEPDN,a bottom-up multi-stage convolutional neural network structure,incorporating the Deep Resolution-Net,a progressive multi-level resolution network,allows high-speed pedestrian pose detection in complex backgrounds and detects the small and obscured targets of images.Compared with the current mainstream algorithms collected by the COCO database,the MEPDN achieves an average increase of about 5% in several key metrics such as m AP and AP[L] and about 4% on the MPII data compared to the current mainstream algorithms.(2)Suggest a lightweight convolutional neural network,Auto Drive,inspired by Squeeze-Net and YOLO,focusing on the lightweight intelligent driving network.By employing the single-stage detection channels with the Deep Resolution Net structure,the Auto Drive network simultaneously computes the Region Proposition and the classification in separate networks to enable high-precision autonomous driving under the small parameters.We combine the MEPDN with the Auto Drive network to examine the effectiveness of the MEPDN in providing the pedestrian detection capability of the Smart Driving Network.The integrated network is tested on the KITTI data and achieves an accuracy improvement of nearly 5% for pedestrian detection compared to current mainstream algorithms.Measured on the KITTI database,the integrated network has improved nearly 5 percent of the accuracy for pedestrian detection compared to the dominant algorithms.