Object Location Methods In Complicated Background

Object location is defined as finding the exact location of target objects in a given image which contains several or no targets.In some complex scenes,object location becomes a big challenge due to complex background,deformation of objects,uneven brightness and occlusion issues.Specific object location methods often provide better performances than universal methods in specific application scenarios.Object location is widely applied in remote sensing image processing,medical image analysis,industrial vision and other fields.In this paper,three typical complex scenarios are studied based on theoretical research of object location methods:1)Label location on the bottles in the industrial production;2)Barcode location in images captured by smartphones;3)Bottle location in boxes in packaging industry.Three methods are proposed in this paper according to the above scenarios:1.A label location method based on feature points matching is proposed according to object location based on image matching.The proposed method can give the exact position of target labels and reject bottles with missing labels,wrong labels or defective labels at the same time.The location of target labels also facilitates subsequent steps such as defect detection.Experiments has proved the high detection accuracy and rate of the proposed method even in the images with complex backgrounds such as brushes in the images,uneven brightness,over exposure.2.A barcode location method based on machine learning and Hough Transform is proposed inspired by image classification methods.The proposed technique performs well even when the barcodes are twisted,occluded or partially illegible due to reflection without any prior information and man-made annotation.Meanwhile,the method obtains the exact bounding boxes of the barcodes in test images,which can abate searching field and cost when decoding and improve decoding accuracy.3.A bottle location method in boxes based on deformable part models(DPM)is proposed inspired by image segmentation and object detection algorithms.The proposed method avoids exhaustive search of the sliding window method and the construction of the multi-layer image pyramid by the introduction of Maximally Stable Extremal Regions(MSER)algorithm,which greatly improves the detection rate.The rotation of segmented objects brings great improvements to the low detection accuracy of the DPM models when processing rotated objects.Experiments prove that the proposed method can meet the requirements of industrial production in terms of detection accuracy and detection efficiency.