Design And Implementation Of Target Detection System For Miniature Vehicles Based On Joint Adaptive Kalman Filter

Target detection for vehicles based on local environment perception by vehicle-sensors,then estimating the motion state information of the forwards target.The estimation for driving vehicles is an important research topic in intelligent vehicle auxiliary driving system,it mainly includes the Real-time measurement experimental car carrying sensor and relative horizontal longitudinal velocity,the relative target in front of the transverse longitudinal acceleration and the relative position of state variables.But the single sensor may be easily affected by the environmental change,so the the multi-sensor fusion will be introduced to the miniature intelligent vehicles,the paper proposed a multi-layer filter through extending Innovation-based Adaptive Estimation.Compared with traditional target detection in single sensor,this solution significantly improve accuracy and variety of target information.it adopts multi buffer pools to reduce time error and space error,thereby improving system efficiency.On the other hand,the miniature intelligent vehicles technology can reduce experiment costs and risks,make the experiment easy to maintain and realize simulation of mass vehicle motions,which provides good protection for the experiment.The paper based on miniature intelligent vehicles in the scene.So every miniature vehicle can get the motion state information of other vehicles and obstacle.The design and implementment of target detection systemis as the following two aspects:(1)Implementation of miniature car in front of the target detection: Miniature car use cameras,radar and in the process of driving vehicle photoelectric encoder to continually collect vehicle in front of local traffic conditions,then use the mini2440 processor of information from different sensors to centralized processing.In mini2440 the algorithm is as follows: Based on HOG + SVM pedestrian recognition to extract the image data of pedestrian targets,use the region segmentation method to extract the target movement information radar data and use the photoelectric encoder to obtain its own speed.In the end,the information which processing is completed will be sent to the server for data fusion display of local traffic environment.(2)A multi-sensorial Federated Kalman Filter algorithm based on Innovation-based Adaptive Estimation(IAE)is proposed(JAKF),which is used to estimate motion state of the vehiclesahead.In this system,Lidar and Radar become local filter,local filters adjust measurement noise covariance Rand system noise covariance Qadaptively,then global filter uses the measurement noise covariance and system noise covariance to calculate information allocation factor for Federated Kalman Filtering,finally,according to the information distribution factor to complete the optimal information fusion and the global filtering results feedback to each local filter.Federated Kalman Filter combine advantage of multiple sensors,so the global filtering's results is optimal than each local filtering's results,in addition,the feedback mechanism can improve the filter measurement accuracy.According to the result of experiment shows that the proposed miniature car function of target detection system is correct and can achieve the goal of real-time monitoring local traffic environment.Experiment results show that the algorithm has great adaptive capacity and error-tolerant capacity.When a sensor filtering performance is abnormal,the system of filtering result does not have exception.And the algorithm has the most accurate estimate than the standard one does and single sensor one does.