Research On Pedestrian Detection For Assistance System On Vehicle

Pedestrian detection is a research focus and front direction in Computer Vision field. It has great application value and significance in the Vehicle Driving Assist System, by means of detecting pedestrian before vehicle and timely early warning, the driver could take effective measures to protect pedestrians, which can ensure road traffic security. This thesis focuses on studying and improving pedestrian detection algorithm, which is designed and simulated based on MATLAB platform, then an embedded pedestrian detection system is expected to be built, the algorithm is transplanted to the DSP platform.During the pedestrian detection algorithm design and simulation, pedestrian detection is regarded as a binary classification problem, and based on statistical learning—AdaBoost algorithm is mainly researched and analysed in the thesis. In order to improve the detection accuracy of algorithm, a combined algorithm based on Support Vector Machine and AdaBoost is used for pedestrian detection. Feature Selection is a key technique in pedestrian detection, Histogram of Oriented Gradient (HOG) features are input to the classifier, at the same time. Multi-scale HOG features are put forward by absorbing the variable scale characteristics of Harr-Like feature presented by Viola, which can describe different size areas of the characteristics of person and get more detail information. Then, non-person sample false positive rate is introduced in the process of weight-updating, a cascade structure of the pedestrian classifier is trained. Experimental results show that detection accuracy and calculation time of modified AdaBoost algorithm are 99.8% and 577.66s,99.8%和970.73s separately under the conditions of 5,8 selected features, which is better than the detection accuracy of traditional AdaBoost algorithm. A six-level cascade classifier is trained by using a total of 46 features, it can achieve 96.74% detection accuracy with 3.26‰false positive rate on INRIA person database.TI's TMS320DM642 chip is a high-performance digital signal processor for digital media, and is a core component of video processing. ICETEK-DM642-PCI development board is used as hardware development platform, real-time operating system DSP/BIOS tool is to configure and schedule reasonably tasks. With RF5 reference standard framework, pedestrian detection system is constructed based on DSP and realizes the functional design of video capture module and video display module.