Vehicle Detection Algorithm Based On Fusion Of Laser/Infrared And Multi DSP Software Implementation

Vehicle detection and tracking have great significance in military and civil fields.In some special scenes,the algorithm in infrared image shows its limitations.For example,when the vehicle is static and the infrared radiation of the vehicle is similar to the environment,it is difficult to detect the vehicle by the infrared information;when the vehicle is occluded and the infrared algorithm cannot extract the complete contour of the target,it is also difficult to identify the target.In this thesis,we studied the vehicle detection algorithm in laser image,and the vehicle detection algorithm based on fusion of laser/infrared,in the end the whole algorithm is implemented on DSP.The main research of this thesis is as follows:Focus on the high time complexity of detection algorithm in laser image,this thesis studies the low time complexity and anti-occlusion detection algorithm in laser image.In the algorithm,Ground Model is established firstly,then according to which,frame integral based on elevation filtering is carried out,finally the region of interest(ROI)is recognized in the integral image.In the Ground Model building stage,the algorithm reduces the time complexity by using the row histogram statistics.Moreover,the algorithm parameters are calculated adaptively while the algorithm running.The experimental results show that the algorithm can detect the vehicles in different distance and different attitude,and it can also detect the occluded vehicle target.Focus on that the existing fusion theory is difficult to apply in practice,this thesis studies the fusion algorithm based on D-S evidence theory,and proposes a fusion method which is easy to implement: in the region of interest,Basic Probability Assignment is established by the probability output of the original algorithm,and then the subsequent evidence combination and decision-making are carried out.Through this method,we realized the vehicle detection algorithm based on fusion of laser/infrared,which has the ability of laser precise positioning and infrared texture recognition,and we realized the fusion algorithm of tracking algorithm and laser detection algorithm,which achieves the goal of tracking rectangle scale update.The experimental results show that the fusion algorithm retains the advantages of meta algorithm and widens the applicable scenarios of a single algorithm: when the vehicle is static or occluded,the fusion algorithm's F1 Score achieves 96.55%,which is higher than 37.83% for infrared algorithm and 96.00% for laser algorithm.Focus on the feature of our system,which is multi-DSP parallel,software/hardware parallel and IP parallel in FPGA,this thesis designs a set of stable DSP Software.DSP Software deals with concurrent interrupt,and large-scale data transmission.Based on DSP Software,we transplant the whole algorithm to DSP.The experiments show that DSP Software can run stably,and the algorithm can run on it in real time.