Object Detection Binary Classifiers methodology based on deep learning to identify small objects handled similarly: Application in video surveillance

The capability of distinguishing between small objects when manipulated with hand is essential in many fields, especially in video surveillance. To date, the recognition of such objects in images using Convolutional Neural Networks (CNNs) remains a challenge. In this paper, we propose improving robustness, accuracy and reliability of the detection of small objects handled similarly using binarization techniques. We propose improving their detection in videos using a two level methodology based on deep learning, called Object Detection with Binary Classifiers. The first level selects the candidate regions from the input frame and the second level applies a binarization technique based on a CNN-classifier with One-Versus-All or One-Versus-One. In particular, we focus on the video surveillance problem of detecting weapons and objects that can be confused with a handgun or a knife when manipulated with hand. We create a database considering six objects: pistol, knife, smartphone, bill, purse and card. The experimental study shows that the proposed methodology reduces the number of false positives with respect to the baseline multi-class detection model. •We propose and evaluate a two level methodology called ODeBiC, based on the use of deep learning, to improve the detection of small objects that can be handled similarly. The first level uses a detector to select from each input frame the candidate regions with a specific confidence about the presence of each object. Then, the second level analyses these proposals using a binarization technique to identify the objects with higher accuracy. ODeBiC methodology maintains a good accuracy for the detection of large objects as well.•We analyse the potential of binarization techniques such as, OVA and OVO, to improve the detection of small objects, manipulated with hand, that can be confused with a weapon. As far as we know, this is the first study in analysing such potential.•We build a new dataset called Sohas_weapon (small objects handled similarly to a weapon, dataset) for the case study of six small objects that are often handled in a similar way to a weapon: pistol, knife, smartphone, bill, purse and card. We used different camera and surveillance camera technologies to take the images. 10% of the images were downloaded from Internet. All these images were manually annotated for the detection task. This useful dataset will be available for other studies (http://sci2s.ugr.es/weapons-detection).