Pedestrian detection method for traffic environment based on human tree model

Abstract

The present invention discloses a pedestrian detection method for the traffic environment based on the human body tree model, and belongs to the field of road traffic pedestrian detection. The method comprises: selecting a data set with annotation information of the human body joint as a training sample of the model, and expanding the joint into the required human body part; based on the relative position relation between each parent part and child part, using principles of the relative distance of the sample, the mean value of the sample correlation difference and the mean value of the total correlation difference of the sample set, optimizing the initial clustering center of the K-means algorithm to realize the clustering of the various parts of the human body, and obtaining hidden variables of the training samples; using a coordinate reduction method to solve the hidden structure SVM problem, and training, obtaining, detecting and determining the models; in the detection phase, according to the constructed human tree structure, the part state transition equation and the off-line training model, merging the dynamic planning idea to realize the traversal of the pyramid of the test sample, obtaining the whole human body detection result of the image, and using a non-maximal suppression algorithm to obtain the final detection bounding box.

Description

technical field [0001] The invention belongs to the field of road traffic pedestrian detection. Combined with an improved human tree diagram model to encode the co-occurrence and spatial relationship between human parts, it involves a traffic pedestrian detection method based on part pose estimation algorithm. Background technique [0002] In recent years, due to the increase in car ownership and utilization rate, the occurrence of traffic accidents has shown an upward trend year by year. Among them, pedestrians are the main victims of traffic accidents, and ensuring their life and property safety is an important topic in the field of intelligent vehicle research. With the effective application of Advanced Driver Assistance Systems (ADAS), the functions of intelligent vehicles in terms of assisted driving and danger alarms have been gradually improved. How to quickly and accurately detect pedestrians in the driving process, effectively improve the robustness and safety of ...

AI Technical Summary

Problems solved by technology

However, the above work often regards the target pedestrian as a whole or divides the human body into multiple independent parts for training, ignoring the space and symbiotic relationship between the various parts of the human body, making it difficult to accurately distinguish the pedestrian's posture, which in turn affects The accuracy of the detection results and the coverage of the bounding box on the human body

Images