search

kaledhoshme123 / Early-Detection-of-Collective-or-Individual-Theft-Attempts-Us-ing-Long-term-Recurrent-Convolutional-

I designed an intelligent system capable of analyzing movement within the videos and detecting suspicious movement that precedes the occurrence of shoplifting crimes. The proposed system can analyze the movement into two primary classifications: the natural movement, and the suspicious movement (with the percentage of each of them being determined.” Thus, the system appears, depending on the percentage of the type of movement, whether the possibility of theft is high or low, or the Confusion movement, which are branched cases depending on the percentage percent accuracy of smart model classification"). The system is integrated with surveillance camera systems that are placed in stores, and the system can at that time alert security personnel in cases where the movement of people in the monitored area appears to be suspicious. The system can also help in cases where it is required to search within a large number of video clips recorded by the surveillance cameras to determine the time moments before the theft crimes. The compressed file contains several video clips on which the system has been tested (the system is waiting for 160 frames to pass, “that is, approximately 3 seconds on average, depending on the frequency of the frames within the video clips or the live broadcast”). I sent you a detailed study of how the system works, and if you like the system and find that it can complement your software systems, I will send you the code and the smart trained model.
MIT License
16 stars 5 forks source link

some discuss #1

Open zbt1356660588 opened 1 year ago

zbt1356660588 commented 1 year ago

Dear Professor, I'm sorry to disturb you. I am currently studying deep learning, especially object detection and posture recognition. The content we learned in class is related to the identification of criminal behavior. My idea is to recognize behavior through 3D convolution. My approach is to divide the criminal stage into two parts and create two datasets respectively: one for suspicious behavior before a crime, and the other for successful criminal behavior. These two behaviors only last about 3-5 seconds. Then, by preprocessing the video data of these different categories into 16-frame images, along with normal behavior videos, there are three categories in total for training. Finally, a three-classification behavior division will be achieved to predict crime and issue an alarm. However, I'm very interested in your pre-processing part. Did you use OpenCV? Can I discuss it with you in more depth? Sorry for taking up your precious time. Thank you again. Sincerely Email: zbt1356660588@gmail.com

kaledhoshme123 commented 1 year ago

Dear Professor, I'm sorry to disturb you. I am currently studying deep learning, especially object detection and posture recognition. The content we learned in class is related to the identification of criminal behavior. My idea is to recognize behavior through 3D convolution. My approach is to divide the criminal stage into two parts and create two datasets respectively: one for suspicious behavior before a crime, and the other for successful criminal behavior. These two behaviors only last about 3-5 seconds. Then, by preprocessing the video data of these different categories into 16-frame images, along with normal behavior videos, there are three categories in total for training. Finally, a three-classification behavior division will be achieved to predict crime and issue an alarm. However, I'm very interested in your pre-processing part. Did you use OpenCV? Can I discuss it with you in more depth? Sorry for taking up your precious time. Thank you again. Sincerely Email: zbt1356660588@gmail.com

Welcome Yes, you need to use OpenCV as a default in many of the video pre-processing methodologies included in the dataset. I'm available to help with pre-processing steps and more. I wish you success.