Research On Target Detection Algorithm And Application Based On Intelligent Terminal

With the continuous development of economy and technology,the moving target detection technology has made a great progress.A common moving target detection algorithm is to obtain the moving target regions by subtracting the moving target images from the background images.Factors,such as the light intensity of the detection areas and the obstacle of moving targets,will influence experimental results.In addition,the quality of detected moving targets will have an effect on the recognition of moving targets,tracking,and video surveillance.When a camera is under the condition of low light or low visibility,such as fog,rain,midnight,etc.,the traditional target detection algorithms may have problems of low detection accuracy,low speed,and high missing rate.In order to solve these problems,this thesis initially uses an improved image enhancement algorithm to process images,and then the processed images are transmitted to real-time object detection(YOLO)and the residual network(Resnet)modified convolutional neural network for target detection.As a result,based on the smart terminals system,the final design can detect moving targets.The main points of this thesis are as follows:1.In order to solve the problem of poor recognition of the camera on severe weather conditions,such as low illumination and low visibility,the research compares Single Scale Retinex(SSR),Multi-Scale Retinex(MSR),Multi-Scale Retinex with Color Restoration(MSRCR),and White Patch Retinex,and finally an improved White Patch Retinex is proposed.This algorithm does not use the brightest point of the images as the light intensity of the incident light.However,it calculates the cumulative histogram of each channel in the original images and takes the accumulated value as the intensity of incident light in the images.Compared with other ones,the improved method performs incredibly better than the original one,because the color of the image is more realistic and clearer,and the image color saturation is extremely high.In conclusion,the experiment shows that this algorithm has a good detection effect under different conditions,such as low illumination,low visibility,etc.2.To deal with the potential problems of using the YOLO convolutional neural network when detecting moving targets,this thesis emphasizes that the Resnet network and YOLO network should be combined to detect moving targets.Due to the deeper level of the stray network,the targets can be improved with the guarantee of the detection accuracy of network.Therefore,in this thesis,YOLO's 24-layer convolutional layer classifier is deleted,and a 50-layer residual network is used to make it as a feature extractor for detecting moving targets.Consequently,the experiment proves that the detection accuracy of the algorithm has been significantly improved,since the detection rate is comparatively fast and the robustness is relatively high.3.Low-illumination and low-visibility images can be processed through computer terminals,and transmitted into the improved YOLO network to identify and track moving targets.At the same time,it can be applied into a wide range of areas,such as security,autonomous driving,the power industry,and urban management.This design of the upper system has a high detection efficiency and recognition.On the one hand,the image enhancement can be realized in short time.On the other hand,the target detection and recognition can be achieved.