The Technology Research Of Vehicle Detection Based On HOG Features

With the pace of economic development and urbanization, traffic jams are frequent,resulting in frequent accidents, and technology-related intelligent transportation isincreasingly attentions related object detection technology is also booming stand up. In manyobject detection technique, the detection of a vehicle is the relatively difficult because of thecomplexity of the vehicle itself, for example, a variety of different models, different colors,and even the shape along with the presentation of the individual visual angle differ. Existingvehicle detection algorithms because of its limitations in all aspects, and with the conversionapplication scenarios, and can not have reached a very good detection results. Thus paper alsoconducted a related inquiry, the main work is as follows:(1) Implemented based on HOG features and SVM classifier vehicle detection. HOGedge because of its good characteristics, and reflects the robustness, so this feature to describethe vehicle using HOG profile. Support Vector Machine SVM due to its good dichotomycharacteristics, combined with easy to use, so use it with incorporated herein HOG. However,in the actual scene, the situation is more complex road traffic, the vehicle will be showing theperformance of different shapes, so this training in a number of different shapes classifierswere detected.(2) From the detection time analysis of some of these parallel optimization algorithms.The use of multiple classifiers, combined with a large number of the shortcomings of HOGfeature dimension itself, making this the computation time will seem longer, so this paper on atwo-stage vehicle detection for parallel processing of the learning phase HOG algorithmparallelization, depending on the calculation steps HOG features, respectively, in the gradientcalculation, cell histograms and normalized process optimization, in addition, also in thedetection phase scanning and image scaling and window merger parallelism, This willcertainly reduce the amount of calculation, make up a certain amount of time consumption.Based on the above work to achieve the detection of vehicles, tested tested to faster,more accurate identification of the target image in most vehicles.