Day #139 - Video Analytics Case Study - Retail Scene Gun Detect Alert Situation

Summary of key points from papers studied, Sample projects, Code base to follow it further. This is a summary and bookmark. Credits to bookmarked papers / authors

Papers Studied
Paper #1 - Real Time Monitoring of CCTV Camera Images Using Object Detectors and Scene Classification for Retail and Surveillance Applications

Key Points

• AlexNet has 60 million parameters and 650,000 neurons, consists of five convolutional layers. Those layers are followed by max-pooling layers, and three globally-connected layers with a final 1000-way softmax layer

CNN Architecture Overview

• Kernels are learnable filters
• Pooling layer (sub-sampling) reduces the dimensionality of feature map
• Output of Softmax function categorical distribution

Datasets

• Knives Images Database
• Internet Movie Firearms Database
• EgoHands
• ImageNet
• Pistol-Detection-in-Videos
• Knife Image Database 

Application Architecture
Input CCTV Video -> Detection using CNN -> Push Notification to Mobile

Paper #2 - Crime Scene Prediction by Detecting Threatening Objects Using Convolutional Neural Network

From the scene the objects are extracted, CNN is used to identify. Detecting weapons is a bigger challenge than identifying / classifying it.

Paper #3 - Automatic Handgun Detection Alarm in Videos Using Deep Learning

• Reformulated problem to reducing number of False Positives

Detection Models

• Sliding Window Approach
• Large number of candidate windows
• Runs Classifier on all windows
• HOG based model
• Good at 0.07 frames per second / pedestrian

Region Proposals

• Region based CNN
• Approach selects accurate candidate regions
• CNN to extract features, SVM to classify them
• Good at 7 frames per Second
• Alarm Activation Time per Interval (AATI)

Paper #4 - Automated Detection of Firearms and Knives in a CCTV Image (Michał Grega*, Andrzej Matiolanski, Piotr Guzik and Mikołaj Leszczuk)

Challenges

• Blurryness
• Low Resolution

Knife Detection Algorithm

• Input Image -> Sliding Window -> Feature Extraction -> SVM Classification -> Decision

Firearm Detection Algorithm
OpenCV Techniques used

• Background Detection
• Canny Edge Detection
• PCA

Neural Network operates on the Image Obtained after PCA (Reduced Dimensions)
OpenCV Post Evaluation

• Dialation
• Erosion
• Difference between two images

Experimentation - Tested this haar cascade based implementation, Works fine for Gun Detection
Results

Other Interesting Projects
Real Time Implementation of Gunshot Detection System
tensorflow-gun-detection
Deep Neural Net Approach To Identify Guns

Happy Learning!!!