Research On The Fire Engine Access Vehicle Detection Technique Based On Multi-feature Cascade Classifier

With the rapid development of economy and living standards, the number of vehicles has increased, but the number of risks is also increased. One of the risks is the fire hazard. Fire engine access is occupied by private cars which bring huge dangers in fire safety. To reduce such hazards, a scientific fire engine access vehicle detection system should be utilized.Fire engine access vehicle detection system is based on the Internet of Things. It aims at preventing the fire accidents that are caused by the occupation of fire engine access. Once the fire engine access is occupied, the system can identify the car and contact to people in charge to deal with the occupation in time. Because the vehicle detection algorithm is the core of the fire engine access vehicle detection system, an applicable vehicle detection algorithm is needed.This thesis does some researches of the common methods include methods of feature-based, optical flow, machine learning of vehicle detection. It proposes the method of vehicle detection that bases on adaptive threshold detection of saliency vehicle bottom shadow segmentation and multi-feature fusion cascade classifier. Here are the details:This thesis indicates to use the adaptive threshold to segment the shadow of the bottom of a vehicle. First, the saliency detection method is used to extract images of a road area of the vehicle and it can exclude the impacts of objects in a non-lane area. Second, the gray sampling is made for the extracted road area. The statistics from the sampling are used to calculate segmentation threshold of the vehicle bottom shadow. Third, the edge extraction method that bases on the pixel change rate is used to extract the vehicle bottom shadow edges. Finally, based on the bottom shadow edges, a vehicle region of interest is built.Based on the multi-feature fusion cascade classifier vehicle detection method, the thesis first fuses the Haar and LBP features which are based on the Fisher criterion. Then it utilizes the AdaBoost cascade classifier to detect vehicles. In the first step, the HOG feature classifier which is better in anti-lighting is used to eliminate the easily distinguished non-vehicular samples. In the next step, the multi-feature fusion classifier is used to eliminate non-vehicular samples that are difficult to distinguish.Experimental results show that the proposed method can adaptively complete vehicle detection in the fire engine access under different climatic conditions, compared with the traditional vehicle detection method, it has higher detection rate and lower false positive rate.