Research On Object Detection Algorithm In Road Panoramic Image Based On Deep Learning

Combined with artificial intelligence algorithm,object detection of panoramic images has become a promising visual application with the development of panoramic technology and deep learning strategy.At the same time,the optimized deep learning network model provides a theoretical basis for the research of artificial intelligence in all aspects of life,which greatly promotes the research and application of unmanned autonomous systems in some Internet companies(such as Baidu’s automatic driving department)and car companies(such as Tesla).Accurate detection of road objects in a panoramic environment can be used as the basis for autonomous driving,ensuring the safety of driving,and it is also an important basis and necessary condition for achieving L5-level driverless driving.Most training images have too high pixel precision.After deep learning training through the model,the sensitivity of the model during the detection process is relatively low.When the model is used to process images with low pixels,an extremely high rate of missed detection is caused.However,most of the network models have problems in balancing detection accuracy and detection speed,this is an urgent problem to be solved.Firstly,this paper proposed a new CAMnasNet algorithm based on the Faster-RCNN algorithm to improve the road panoramic image object detection algorithm.Combining the lightweight network SENet and MNASNet to optimize the CAMB Block module,the algorithm can simultaneously take into account the advantages of less parameters,fast speed and high precision.Compared with other algorithms,the detection performance has been significantly improved due to the detection of small objects.In the data set test,the detection speed of this algorithm has been largely improved and basically reached one-stage in terms of time performance.Secondly,this paper proposed a new PAEDNet algorithm for road panoramic image object detection.The algorithm improved the recall rate through improved anchor matching rules and enhanced the feature expression ability of the details,thus achieving more accurate target classification and position localization.The detection performance of small and distorted objects in panoramic images has been improved,too,especially for categories such as person,sign,and line.By comparing other algorithms with experiments,the overall detection performance of the PAEDNet algorithm is currently optimal.Finally,this paper designed and developed the object detection software under the panoramic image.The object detection algorithm based on CAMnasNet network and PAEDNet network proposed in this paper was used to build a real-time application system platform,which realized the visualization and tooling of the object detection task flow.