Research And Implementation Of Artificial Fruit Fly Visual Collision Early Warning System Based On DSP

Fruit flies are a kind of insect with the neurons of vision and smell.They can not only use smell neurons to determine the distance between themselves and food,but also judge the approximate orientation of the food in terms of their visions neurons.This biological mechanism provides an important biological resource for us to study fruit fly optimization.On the other hand,they are highly sensitive to the moving target in the visual field.Their behaviors of obstacle avoidance can give us another new way to study computer vision,whereas few work has been done up to now.Therefore,this study,based on the behavioral characteristics of fruit flies' visual systems,designs an improved artificial fruit fly visual collision alarm model and explores a fruit fly visual collision alarm system with TI's DSP chip TMS320DM642.The results are helpful for accelerating the application of biological metaphors of fruit fly's visual systems in computational intelligence;they also help us investigate the possibility of integrating artificial fruit fly visual models with hardware.The main research work and the achievements acquired are summarized below:A.As associated to the characteristics of DSP development platform and CCD's video image format,some relative hardware knowledge for the development of a collision alarm system was introduced,and meanwhile a special hardware development environment was determined.Further,introduced the relation between RGB and YCrCb color spaces,and analyzed the influence of threshold value on image binarization.Formulated the principles of noise processing and object extraction in image processing,and then obtained a denoising and object extraction approach suitable for the DSP hardware environment after analyzing their advantages,disadvantages,differences,relationship and applicable scopes.Finally,the simulative television was introduced in order to design the system of fruit fly visual collision alarm,which lays the foundation for further research and development.B.The structural characteristics and biological principles of fruit fly visual systems were formulated,while the principles,design ideas and faults of the reported artificial fruit fly visual collision detection model were summarized and analyzed.An improved artificial fruit fly visual collision detection model was proposed by improving the design of the reported medulla layer and a scheme of collision detection.Finally,with the help of video sequences of glass ball movement and vehicle driving,the computational model was examined whether to be able to effectively and rationally send correct alarms.Numerical results showed that the model is effective.C.Based on the TMS320DM642 chip,CCD camera,video image output device and LED lights,a hardware platform of fruit fly visual collision alarm was developed.As related to the requirement of hardware design,a DSP collision alarm system was designed by improving the efficiency of the proposed model.It bases on the idea of marking the area of which the target is located,while the model is transplanted into the hardware platform.Finally,its warning effect and real-time performance were detected by means of video sequences in real scenes.Experimental results have validated that it is preliminarily effective in real-time environments and can effectively send alarm signals.