On the feature detection of nonconforming objects with automated drone surveillance

Recently, drones have illustrated some more attentions in the relationship between surveillance and militarization, despite the several constraints such as the endurance, weather conditions and the lack of the harmonized standard on the take-off weight. By the supposed accuracy and precision, drone systems may be used as the replacement of troublesome daily works. In our study, the authors tried to propose the seeking method on the finding of unwanted or broken dumping bikes on the street. The proposed reinforcement method is based on the improvement of key points matching problem by SIFT algorithm. Generally, the more key points being matched between two images mean the much similar between them. On the contrary, the more key points could confound the features either. To strength the advantage of key points, our proposed method focused on the multiple refinement rounds during the matching progress, that is our reinforcement matching. According to our experiment, we not only illustrate the improvement of clustering accuracy in the face recognition and aerial photography database, but also we testify the possibility of retaining the clustering purity.