Design Of Real-time Target Detection System Based On Dual-band Sensor

Target detection technology is one of the important topics in the field of computer vision,artificial intelligence and other fields of research.It plays an important role in the field of security monitoring and defense industry.However,traditional target detection algorithms usually run on PC or server clusters,which have poor portability and high power consumption,and cannot satisfied the needs of outdoor real-time detection tasks.What`s more,the traditional target detection system usually uses a single-band sensor for data collection,which can only compatible with one type of data input of a visible light sensor or an infrared sensor.When the lighting is sufficient and the scene is simple,the field of view can be clear,the visible light image detection works well,but when there is interference such as occlusion and shadow in the scene,the detection result of the visible light image is missed or get the wrong result.Compared with the visible light image,the infrared image has a greater advantage for scenes with strong temperature differences in the infrared wave band,such as life objects and engines at night,but due to the limitations of the detector,the infrared image usually has a lower resolution and edges The distortion is severe and is not suitable for all work scenarios.In order to solve the problem that the single-band detection system cannot meet the needs of different working scenarios,this paper designs and implements a real-time target detection system for dual-band sensors.The system integrates visible light and infrared dual-band sensor input interfaces and uses current mainstream method in target detection.Compared with the single-band detection system,the detection algorithm can better meet the needs of different scene detection tasks.This system adopt FPGA and embedded GPU heterogeneous architecture for system integration,which makes full use of the flexibility of FPGA and the advantages of parallel processing of GPU.,the detection system achieve low power consumption and miniaturization while ensuring the detection accuracy and real-time performance of the system.The main work contents are as follows:1.Hardware system design based on information infrared and visible light dual sensors:In order to solve the problem that the single visible light sensor or infrared sensor has limited information and insufficient detection capability,this paper proposes a design scheme of a target detection system that uses a combination of visible light sensor and infrared sensor.And around the system design scheme,the modular circuit is used to design the hardware circuit of the system.The image acquisition input module based on CMOS sensor,the image processing module based on FPGA+GPU and the real-time display module based on SDI are designed respectively.Finally achieves a visible light camera with SDI interface can be used independently and a circular processing board with a radius of 14 cm.It meets the system technical indicators and functional requirements while ensuring a good heat dissipation environment and engineering application development.2.Real-time target detection algorithm design:This paper analyzes and studies the current mainstream target detection algorithms,weighs the computational complexity of the algorithm and the accuracy of the detection results,and selects the Tiny YOLO v3 algorithm as the dual-band target detection algorithm of the target detection system.In response to the Tiny YOLO v3 algorithm cannot effectively deal with occlusion targets and weak targets,etc.,increase the YOLO network shallow feature output layer and target local feature learning samples to reduce the omission and false detection probability of occlusion targets and weak targets according to the characteristics that shallow convolution is more sensitive to the local characteristics of the target and the small target.Since the Tiny YOLO v3 algorithm is mainly designed for visible light images,this paper uses infrared images to train the Tiny YOLO v3 algorithm,so that the algorithm can meet the needs of infrared image detection,and real-time processing of 1080p@60hz visible light and 480p@50hz infrared input images.3.FPGA-based logic control.In order to meet the real-time requirements,this paper uses FPGA to realize the transmission path of image data,which mainly includes the acquisition input function,data transmission channel,and video output function.The acquisition input function realizes the configuration of the CMOS image sensor and the reception of highspeed LVDS signals using FPGA.And format conversion of various types of image data received for easy transmission.The data transmission channel completes the logical design of the DDR3 image cache based on the AXI bus and the PCIe high-speed data transmission channel by using the IP core inside the FPGA.The video output function enables the processed image to be displayed in SDI format.The target detection system of the dual-band sensors implemented in this paper can meet the real-time processing requirements of 1080p@60Hz visible light image and 480p@50Hz infrared image.At the same time,the system sloves the problems such as the limited information amount and poor detection ability of the current monochrome sensorbased target detection system.What`s more,the system also has the characteristics of low power consumption,strong portability,high integration and so on.It has broad application prospects in traffic supervision,security monitoring,etc.