Object Detection And Tracking Based On The Technology Of Improved Adaboost

As the fast development of the society, research on the tracking of specific object has become prominent.Nowadays, under different disciplines, object detection and tracking is studied in pathological cell detection, accessory defect detection, house security and artificial intelligence. If the object detection and tracking can be realized in a specific way, there can be convenice in the field of biology, military and safety guard.The popular algorithm, Adaboost, is famous for high detection rate, real time detection and robust to illumination variance and view variance. This paper studied algorithms based on Adaboost, and fully developed it to conquer the problems of eyes, face features and car detection and tracking.Main research work is listed below:1. A two-layer Adaboost classifier for eye detection and tracking is proposed. The two-layer structure is composed of a single-eye and double-eye classifiers. Compared to traditional YCbCr eye map algorithm, this algorithm is much more robust to light variance, with its advantage in high detection and tracking rate and low false detection rate. Through the analysis of relationship between training sample number, training stages and classifier's false detection rate, the final training efficiency has been increased.2. Human feature detection based on YCbCr is weak in the condition of illumination variance. A fusion algorithm composed of YCbCr and Adaboost is proposed, which is constructed based on each classifier's trust weight. This algorithm can be applied in the condition of gray image, color image and illumination variance circumstance.3. Modified Adaboost algorithm. Patch-like feature is proposed and Haar-like feature is projected to each Patch-like feature through the first turn feature selection. Finally, each Patch-like feature is weighted to be a strong classifier to realize the car detection. This algorithm can be developed to the use of multiple object detection.4. Different kinds of cameras are analyzed to find out their effects on detection rate. Basler641fc camera is developed for human face tracking and recognition system.