Moving trace optimization tracking for adult of Bactrocera Dorsalis (Hendel) based on Kalman filter algorithm

Abstract: Bactrocera Dorsalis (Hendel) are invasive pests which occur frequently and are seriously harmful to the growth of fruit trees, and they have been ranked an important quarantine object in many countries and regions. The regular manual survey used as the routine predicting method for Bactrocera Dorsalis (Hendel) has not accomplished the requirement of real-time and precise monitoring and warning by means of the adult trapping and monitoring device deployed in orchards. With the development of science and technologies, the method of the automatic machine monitoring for pests has been studied including detection of sound characteristics, radar monitoring and spectral monitoring. Considering the characteristic with randomness, migratory and hiding for Bactrocera Dorsalis (Hendel), there were some problems such as timing, processing and costs in monitoring pests with the aid of combining the above monitoring and the traditional method. In order to accomplish precise monitoring for Bactrocera Dorsalis (Hendel), machine vision technologies were used as an in-field automatic detecting method for the Hendel adults in this paper. Considering the problem with tracking Bactrocera Dorsalis (Hendel) object disappearance in multi-objects with more closer condition by means of the mean shift algorithm in color space according to previous machine vision technology research results, the fusion algorithm based on mean shift and Kalman filter theories for moving objects was proposed for optimizing multi-objects moving trace tracking by means of colorful analysis for moving objects and background in monitoring zones. The recurrence relation of the adults moving trace was obtained, and position coordinate, X-component and Y-component of speed in the 2D plane were extracted by image processing and matching technologies in this algorithm. By analyzing the state sequence linear minimum variance estimate theory of dynamic system and recurrence relation of the adults moving trace, the model of state estimate based on a Kalman filter was built to achieve the position estimation of the adults using the prediction and modified equation of the model. The experiment of the adults tracking under the condition of single object and the condition of multi-objects with scatter and gathering indicated that the mean shift algorithm was adaptive to track the adults in the condition of single object with monitoring precision of 100%, was not adaptive to the condition of multi-objects with scatter and gathering since corresponding monitoring precisions were 86% and 76% respectively. The cooperation of mean shift and Kalman filter algorithm estimating of moving objects' approximate location could achieve the stable and continuous tracking in the condition of multi-objects with scatter and gathering with corresponding monitoring precision of 96% and 93%. The real-time tracking experiment of the adults moving trace in pest monitoring zones by the aid of machine vision further validated the practicability of the Hendel adults population density monitoring for providing a theoretical and practical basis for in-field Hendel adults automatic monitoring technology studies.