Violent Event Detection: An Approach Using Fusion GHOG-GIST Descriptor

Violent Video Event Detection Based on Integrated LBP and GLCM Texture Features

“…GHOG+GIST feature descriptor is a fusion of gradient and texture feature descriptor, the method used scale, magnitude and orientation and gives adequate result but compare to our proposed method GHOG+GIST features vector dimension is substantial and required large time computation cost. [35] 87.8 -HOF [35] 83.5 -HNF [35] 87.5 -LTP [35] 71.90±0.49 -ViF [23] 81.60±0.22 88.01 OViF [7] 84.20±3.33 90.32 ViF+OViF [7] 86.30±1.57 91.93 DiMOLIF [34] 88.6±1.2 93.23 GHOG+GIST [10] AUC HOG [35] 57.43±0.37 61.82 HOF [35] 58.53±0.32 57.60 HNF [35] 56.52±0.31 59.94 LTP [35] 71.53±0.17 79.86 ViF [23] 81.20±1.79 88.04 OViF [7] 76.80±3.90 80.47 ViF+OViF [7] 86.00±1.41 91.82 DiMOLIF [34] 85.83±4.2 89.25 GHOG+GIST [10] 88.86±5. In experimentation, our proposed features descriptor performed on an 8GB RAM, Intel core i7 computer running Windows 10.…”

“…In this section, we have presented our proposed texturebased features extraction technique performance. Five-fold cross-validation technique [7,10,34] have been used for experimentation. Consequently, for each one of the two datasets is separated into five halves.…”

“…The fusion of three features called the MoBSIFT descriptor used to detect violent and non-violent events. Recently, Lohithashva et al [10] introduced gradient and texture-based feature descriptors. The fusion features descriptor extracted prominent features and fed to support vector machine (SVM) classifier to detect crowded and uncrowded violent event scenes in the video.…”

“…Recently, some of the surveys [11][12][13] have published different feature extraction techniques used to detect violent events in videos. The most existing methods based on the spatio-temporal interest points [8], features fusion [9,10], optical flow [14], textures [15,16], trajectories [17], descriptors and deep learning techniques [18]. Some of the researchers have also been focusing on effective segmentation [19], subspace techniques [20,21], and classifiers [1,10,22] used to detect violent events.…”

“…The most existing methods based on the spatio-temporal interest points [8], features fusion [9,10], optical flow [14], textures [15,16], trajectories [17], descriptors and deep learning techniques [18]. Some of the researchers have also been focusing on effective segmentation [19], subspace techniques [20,21], and classifiers [1,10,22] used to detect violent events. Although, they are facing difficulty due to complex background, illuminations, scale variation, slow motion in video surveillance.…”