Research On Key Technologies Of Vehicle Mounted Thermal Imaging Night Vision System

The infrared thermal imaging system is less affected by external lighting and adverse weather conditions.As a driver assistance system,it can effectively enhance the driver's perception for the environment,at the same time,it enhances the safety and security of drivers and vulnerable road users at night or in rain and snow conditions.Based on this reason,in this thesis a low power,small size,low cost vehicle-mounted thermal imaging infrared night vision system was designed.The system in this thesis was implemented on a Xilinx Field Programmable Gate Array(FPGA)and a low power multi-core heterogeneous processor.The functions such as infrared imaging,image preprocessing,pedestrian detection has realized on this system.In this thesis,a design scheme for a vehicle-mounted infrared thermal imaging system which based on a domestic uncooled infrared detector was proposed and highperformance,low-power,small-size packaged core chips were selected.In addition,a scheme for inter-board cascading was designed for the need of miniaturization and low power consumption.The system hardware platform was consisted of an infrared detector driver board,an FPGA image preprocessing board,and a signal comprehensive processing board.To improve the vehicle-mounted night vision system's performance,the nonuniformity correction(NUC),dynamic range compression and detail enhancement methods of infrared images were studied in this thesis,which based on the analysis of infrared image features.Aiming at the shortcomings of the traditional non-uniformity correction method,an improved neural network-based correction method was proposed.This method uses non-local mean filtering to obtain the desired image and updating the correction parameters by judging the motion of the scene,which mitigates the gosting artifacts effect better when the images stays motionless for a long time.Experiments show that this algorithm featured good correction effect and fast convergence speed.For the high dynamic range raw infrared image,an improved algorithm based on bilateral filter was proposed in this paper.After dynamic range compression and local contrast retention of the base component image,an adaptive gain control approach was used on detail component image to strength the edges and suppress the noise,Finally,the processed layered image is weighted and superimposed to obtain an enhanced 8-bit image.Then experiments were conducted on infrared images of different scenes and comparison was made between this algorithm with others commonly used enhancement algorithms.Experiments showed that this algorithm working effectively in improving the contrast of the image and retaining the edge details.What's more,a new simplified image enhancement algorithm based on bilateral filter was proposed,which used detail enhancement and plateau histogram equalization techniques.The image non-uniformity correction algorithm and the simplified enhancement algorithm were completed on the system's FPGA image preprocessing board.The image preprocessing effect has improved better as well as the real-time performance of the entire system.Pedestrian detection of vehicle-mounted infrared night-vision system was an important research content in this thesis.An improved aggregated channel features detection algorithm for infrared pedestrian was proposed.The experimental results showed that this algorithm has good robustness and real-time performance.Finally,this algorithm was optimized and implemented on the multi-core heterogeneous processor DM8148 platform.In the process of transplantation,the characteristics of each kernel of the processor and the communication and synchronization between cores were fully considered.The algorithm was divided into different modules and implemented on each processor,which improved the running efficiency of the pedestrian detection.Finally,the performance index of the vehicle-mounted thermal imaging night vision system designed in this thesis was tested.Tests verified that the power consumption is about 3W,in addition,stable and clear imaging,far detection distance and good pedestrian detection effect was achieved in the vehicle infrared thermal imaging system,which was up to the standards on low power consumption,low cost,and high stability.