Research And Implementation On Intelligent Camera System For Pedestrian Protection

Nowadays,with the constant development of intelligent transportation,The Advanced Driver Assistance System(ADAS,in short)has been widely used in vehicle driving.It provides safety and convenient driving auxiliary functions for people.In addition,the ADAS executes centralized processing that bases on videos captured from each camera.Due to factors such as transmission bandwidth,processing power,storage capacity and computing speed,the above systems cause extreme stress on terminals as well as high cost.Due to the constant improvement of performance and reduction of volume coupled with decrease of energy consumption,dedicated digital processing chips provide a new solution for high-speed image processing.In this paper,we have discussed and implemented an intelligent camera system in regard to pedestrian protection.Thereinto,intelligent cameras integrate image acquisition,image processing and data transmission and other functions,process the images,then transmit the processed results to terminal equipment.At the same time,the embedded intelligent camera also has characteristics such as low cost,low power consumption and easy extensibility.First,the paper introduces the intelligent camera system based on pedestrian protection and its overall module structure and module function.In accordance with the requirements of the core module of the intelligent camera,we select the multi-core parallel processor(ICP)CV2201-350CBI.But,APEX is the main parallel processing module of the image recognition processor of CV2201 series,as the core unit of image processing module.The intelligent camera through apex to improve the parallel accelerate ratio and parallel pedestrian detection algorithm suitable for apex has been selected to detect pedestrians.This article is mainly through the analysis of APEX and its parallel basic unit CU,and combines with the process of pedestrian detection algorithm,thus selecting the Haar+ Adaboost pedestrian detection algorithm which is appropriate to the platform.Through parallelizable analysis of the pedestrian detection algorithm,it is concluded that the three can be parallelized algorithm.We can come to the conclusion that the integral image,the 45° rotating integral image and Adaboost classifier can be parallel implemented on the APEX.Meanwhile,in the APEX parallel implementation phase,we have made the corresponding improvement of key algorithms.In this paper,the difficulty is the algorithm implementable of parallelization and modular on the APEX with 96 working CU(Computational Unit)in parallel.Finally,the overall design of the intelligent camera system is introduced,and the intelligent camera system is evaluated in performance under different weather and illumination conditions.The evaluation results show that the system has great detection effect and nice real-time performance and robustness.