Research on Pedestrian Attitude Detection Algorithm from the Perspective of Machine Learning

Abstract

In the rapid development of science and technology today, the intelligence of the visual system has been highly valued. The recognition and detection of pedestrian attitudes in a complex environment have become the application trend of intelligent video. The widely used of camera machine does not have such a function. Therefore, this article deeply discusses the relevant algorithms of pedestrian gesture detection and recognition based on machine learning. The traditional HOG feature detection can only achieve the relevant detection of the upright walking crowd target. While when the pedestrian makes different gestures, its detection effect is directly affected and challenging to be recognized. So, this article uses the checking methods of the deformable part model (DPM) to check the target pedestrian gesture and elaborate pedestrian's attitude estimation algorithm for the deformable parts principle. Finally, it combines the algorithm with HOG+SVM principles to simulate with the MATLAB and gets the experimental results to show that this approach can make a pedestrian posture test implemented to achieve high precision accuracy.