Research On Vessel Detection And Recognition Fusing Radar And Vision Data

The target detection and recognition technology of the surface unmanned vessel(USV)has important applications in such areas as surface monitoring and reconnaissance,water protection and vessel traffic management.Through the current research on the detection and recognition of targets at domestic and international,it is found that using LIDAR to detect objects has long range detection distance,high precision ranging,high reliability and strong environmental adaptability,and Lidar can work in any weather conditions,whether it is day or night.However,the original data is large and it has serious data noise.Low resolution of radar can lead to missing rate and high false detection rate.In addition,radar cannot identify the target vessel.Using visual system to detect objects,rich feature information can be obtained,such as color,brightness,texture and so on,which is easy to classify and identify objects and visual system has a lower cost.However,the image processing calculation is relatively large and vulnerable to environment impact,and monocular camera cannot get the depth information of the target.The vision sensor is also sensitive to weather and can only work during the day and is affected by strong light,shadow and other disturbances.When the depth information is detected by the binocular camera,the detection range is limited and the accuracy of ranging is not high.Comprehensive consideration of advantages and disadvantages of Lidar and camera,it is better to detect targets fusing the information of the two sensors,which not only reflects the advantages of Lidar detection,such as long range detection distance,high precision ranging,high reliability and so on,but also embodies the advantages of vision detection that obtains the object feature information and analyze and process it so as to identify targets accurately.Therefore,this paper presents a ship target detection and recognition method fusing visual and radar information.This paper mainly includes the following research contents:1)The overall scheme of detection and recognition system for target ship and coastline was designed.Based on the lake experiment scene,the system function of ship detection and recognition was analyzed,and the key technologies to realize the scheme were defined,the relationship among which technologies was clarified.2)This paper studies the ship and coastline detection algorithm based on Lidar.For the vastness and complexity of the lake experiment scene,and the large amount of data scanned by Lidar,if the clustering algorithm is not appropriate,the clustering calculation will be large and the real-time effect will be bad.For this problem,the clustering algorithm based on the spatial angle is put forward,which divides the spatial angle into several parts evenly.By clustering the point cloud data in a small range and in a large range,the feature extraction is used to extract the target and coastline.Under the premise of correct clustering,the experimental results show that the method improves the real-time property of object clustering.3)Ship target detection based on the vision.The vision-based vessel detection methods used in this paper are SVM classifier and saliency detection in parallel.The HOG feature is firstly extracted from positive and negative samples,which is used to train SVM classifier.Secondly,the characteristics of the color,brightness in the lake are consistent,which can be detected with visual saliency method.The lake area can be extracted by the methods of canny edge detection and Hough transform.Finally,a comprehensive judgment is performed to achieve the detection result of ship target by SVM classifier,visual saliency and Lidar.The vision uses online compression tracking method to track the target vessel.After detecting the location of the target vessel,the Haar-like feature of the positive sample and the background negative sample is acquired online.Then the Bayesian classifier is trained to estimate the target vessel.4)Data fusion method based on laser radar,vison and heading reference system(HRS).HRS is used to compensate the angle deviation caused by the shaking of USV due to the influence of waves.This paper presents a combining approach of cross-verification and saliency detection verification radar to fuse radar and vision data for ship detection and recognition.