Research And Implementation Of Specified Pedestrian Detection Based On Machine Learning

The field of artificial intelligence has developed rapidly in recent years,thanks to the improvement of computer processing capabilities and advanced machine learning algorithms.Therefore,in many problems that are difficult to solve using traditional algorithms,the use of artificial intelligence algorithms is gradually being solved.Pedestrian recognition has always faced difficulties that are difficult to solve with traditional algorithms,such as illumination,occlusion,and background changes.Using a single algorithm to extract pedestrian features in a changing environment is also difficult to achieve.Because the machine learning-based algorithm can acquire the ability to adapt to changes in the outside world through learning and training,this paper proposes to use the convolutional neural network and recurrent neural network method in machine learning to identify the specified pedestrians.First of all,for a frame in a pedestrian video,the feature of one frame is defined by the method of extracting the pedestrian pose.For a person's posture is through the 16 joints(right ankle,right knee,right hip,left hip,left knee,left ankle,pelvis,chest,upper neck,head,right wrist,right elbow,right shoulder,Left shoulder,left elbow,left wrist)defined.Because extracting features through this method can be weakened,pedestrians wear,background changes,and the effects of light.The method is to use the convolutional neural network to identify each part of the human body,and the identified area is found by heat map to find the point that can best represent the part.In addition,because the training data set also marks the part of the pedestrian's body being occluded,the algorithm also has a certain ability to recognize partial occlusion.Secondly,for the characteristics of pedestrians,this paper adopts a series of gesture features of pedestrians in video frames.This method is based on the issue of time-series feature extraction.The specific method is implemented using a recurrent neural network with a certain memory function.According to the order of the video frames,each frame is used as an input for a time of the recurrent neural network,and the recurrent neural network selects the memorized portion and the forgotten portion according to its own parameters,and after such a multi-frame iterative operation,the final output will be obtained.That is,the characteristics of pedestrians.Finally,this paper uses the above algorithm to implement a pedestrian recognition system.The system includes a pedestrian positioning module,which is based on the principle of gradient histogram and support vector machine.This is because the gesture recognition algorithm in this paper can only identify the picture in the center of the image.Therefore,for a practical application,a pedestrian positioning module can be used to find the pedestrian and then perform gesture extraction.The system establishes a pedestrian database and compares features to achieve the basic functions of pedestrian recognition.