Monocular Multi-Pose Pedestrian Ranging Algorithm Based on Key Point Detection

A monocular multi-pose pedestrian ranging algorithm is proposed for the problem that when existing ranging algorithms perform pedestrian ranging from monocular images, the multi-scale variation caused by the different height, body size, posture and angle of pedestrians usually produces huge deviations in the detection results. Firstly, the key point information of the person in the monocular image is acquired using the human key point extraction algorithm openpose, the key point coordinate information of the shoulder is filtered out, the shoulder width is pre-set, the information is fused using C++ API, then the depth distance between the pedestrian and the camera is measured using the improved pinhole imaging ranging model, and finally the data output is processed using the mean filter. Through experimental validation, the algorithm is finally verified on the collected data set to have a very high accuracy within 12m, with an error within 0.2m, and a high improvement in robustness for multi-attitude and multi-angle pedestrian ranging.