| Popis: |
Nowadays, circuited televisions are installed in many places, and it is been monitored by the authorities to detect the crimes. The authorities carelessness may prone to miss the important crimes that are been recorded. There are two existing architectures for detecting crimes and human activities such as 3D CNN and two-stream CNN. The drawbacks of the existing techniques are that it is unable to capture the complete local features and the accuracy is less. In order to overcome these problems, this paper proposes long short-term memory based convolutional neural network with feature map merge block to capture the complete features of the scene. The proposed architecture detects the weapons such as guns and knives, robberies, and if anybody is fighting. The experiments are conducted by using 500 crime scene videos collected from Internet. The experimental results shows that proposed architecture yields 90% of accuracy. Further, the results are compared with existing architectures. |
