Research On Multi-object Detection And Tracking Algorithm For Complex Scenes Of Intelligent Vehicle

With the continued growth in vehicle ownership,the automotive industry is facing increasingly severe problems such as energy depletion,environmental pollution,traffic congestion and accidental casualties.As an effective way to solve the above problems,intelligent vehicle have achieved phased results in recent years.The vision-based intelligent vehicle environment perception system is used as the basic part of intelligent vehicle decision-making planning to realize the recognition and tracking of the vehicle's surrounding environment by sensing the traffic environment around the vehicle,so as to realize optimal path planning and decision control of the vehicle.Recently,the researches related to the environment perception have become a more promising aspect.The main research contents of the paper are as follows:(1)Focusing on the problem of communication between vision sensor and laser sensor,data transmission between sensors and computer through Socket communication protocol have been established.In order to derive the monocular camera's internal and external parameters and object distance assessment,the calibration of internal and external parameters of the monocular camera are implemented using the Zhang calibration method.Finally,camera parameters have been utilized to propose image transformation between the various coordinate systems.(2)Aiming at obstacle detection in traffic scenario,the shape of the rigid objects are considered as relatively stable and then the Hog feature is selected to extract the features of the vehicle object.Considering the Boost classifier has the advantages of accuracy,overfitting and computation,the Boost algorithm is used to establish a classifier for vehicle object detection.In order to improve the performance of vehicle object detection,the ROI extraction and the variable-scale sliding window are used to scan the image quickly to determine the position of the vehicle in the image sequence.(3)Aiming at pedestrian detection in urban traffic scenario,the change of appearance,behavior and posture are considered.The integral channel feature are used to carry out different levels of feature extraction for pedestrians object.In order to reduce the false detection rate and detection time in pedestrian detection,the cascade of Boost classifiers filter are utilized for the pedestrian object.As well,the image pyramid is used to sample the images up and down and eventually the pedestrians of different sizes in the image sequence are detected.(4)In order to improve the object detection accuracy of predicting the object trajectory and object behavior,a multi-object tracking algorithm based on spatio-temporal context is proposed.The multi-object tracking algorithm uses the target detection position information as the tracking input and the object's spatio-temporal context model as the basis to find the object maximum response position in the image.On the basis,in order to improve the real-time performance of multi-object tracking,Fourier transform is used to convert the image to the frequency domain for calculation.Therefore,the object position information is clustered and correlated to extract the object trajectory information.