In 2015, we had an idea to create a surveillance system that could understand what was happening in a video - not just detect motion in the frame but also recognize objects and analyze their behavior. It seemed that the most interesting and challenging part would be the detectors and neural network analytics. Everything else should have already been written and implemented in standard components and solutions. When we started the project, we thought the same.
At first glance, it seemed simple: take standard, ready-made components, assemble the system, add analytics, and get a product. Everything seemed to work, but there was a catch.
We discovered that video analytics accounted for only 15–20% of the effort! The main challenge wasn’t in the analytics but in the core functionality of the surveillance system—frame-by-frame video stream processing, which had to be universal, stable, and with minimal hardware load. This functionality, which seemed "standard" and "solved," turned out to be the real issue.
Implementing basic functionality requires large libraries and their numerous wrappers for different development environments. But we needed more than just a demo version—we needed something that would work long-term, stably, and without memory leaks. Cameras must operate 24/7 without failures, regardless of stream quality. However, when a data stream is corrupted or contains errors, problems arise. The reasons for this can vary, from poor-quality software on the camera to peculiarities in H.264 and H.265 codecs. Many camera manufacturers customize their software by adding their own parameters.
It turned out that many open-source solutions, which developers use as a foundation, have deep-seated memory leaks and issues. A developer realizes they can't handle the project alone, burns out on bugs, and releases the source code to attract new enthusiasts. These enthusiasts fix some issues, new "optimistic developers" join, and the cycle continues. The result is Frankenstein-like building blocks that commercial companies later try to assemble into a stable surveillance system.
We saw the same symptoms in other systems as in Web Camera Pro. They promise extensive functionality, but the real issue is stability. Cameras work intermittently, sometimes recording, sometimes not. Motion detection stops functioning, or other features break down. The system runs for a while and then crashes. The root cause is often not just poor software but the very Frankenstein-like blocks at the foundation of the system.
In an ideal mono-brand scenario with an expensive camera, everything may be fine-tuned. But as soon as an "incorrect" camera or a problematic stream appears, system stability issues arise. Solving this requires deep intervention at the core—rewriting the foundational video stream reading and processing functionality, monitoring memory leaks, and optimizing code.
For example, Dahua and Hikvision develop software only for their cameras, making it easier for them. But universal software must work with any device. We re-evaluated every "building block" of our system and completely rewrote the program with a new team. In essence, we discarded everything old and created SmartVision from scratch.
Current Capabilities of SmartVision:
Download SmartVision here:
https://smartvision.dev/cctv-software-download.htm
What’s Next?
We believe surveillance software should not just record video continuously but actively inform users about what is happening and where.
The mobile app should function like a live journal for each camera:
Camera 1 | 12:00 – A man (Person Name 1) with a dog entered Entrance No.1.
Camera 2 | 12:05 – A man (Person Name 1) with a dog entered the 23rd-floor hall.
Camera 3 | 12:06 – Toyota Car, license plate X777YY11, entered the parking.
Camera 4 | 12:07 – Room 1. Lights turned on, and a loud noise was detected.
Each event should be searchable, filterable, and linked to video playback, with automatic transcription of conversations in rooms. This functionality is currently in development.
We are looking forward to receiving your feedback and criticism on the new version. Thank you!
At first glance, it seemed simple: take standard, ready-made components, assemble the system, add analytics, and get a product. Everything seemed to work, but there was a catch.
We discovered that video analytics accounted for only 15–20% of the effort! The main challenge wasn’t in the analytics but in the core functionality of the surveillance system—frame-by-frame video stream processing, which had to be universal, stable, and with minimal hardware load. This functionality, which seemed "standard" and "solved," turned out to be the real issue.
Implementing basic functionality requires large libraries and their numerous wrappers for different development environments. But we needed more than just a demo version—we needed something that would work long-term, stably, and without memory leaks. Cameras must operate 24/7 without failures, regardless of stream quality. However, when a data stream is corrupted or contains errors, problems arise. The reasons for this can vary, from poor-quality software on the camera to peculiarities in H.264 and H.265 codecs. Many camera manufacturers customize their software by adding their own parameters.
It turned out that many open-source solutions, which developers use as a foundation, have deep-seated memory leaks and issues. A developer realizes they can't handle the project alone, burns out on bugs, and releases the source code to attract new enthusiasts. These enthusiasts fix some issues, new "optimistic developers" join, and the cycle continues. The result is Frankenstein-like building blocks that commercial companies later try to assemble into a stable surveillance system.
We saw the same symptoms in other systems as in Web Camera Pro. They promise extensive functionality, but the real issue is stability. Cameras work intermittently, sometimes recording, sometimes not. Motion detection stops functioning, or other features break down. The system runs for a while and then crashes. The root cause is often not just poor software but the very Frankenstein-like blocks at the foundation of the system.
In an ideal mono-brand scenario with an expensive camera, everything may be fine-tuned. But as soon as an "incorrect" camera or a problematic stream appears, system stability issues arise. Solving this requires deep intervention at the core—rewriting the foundational video stream reading and processing functionality, monitoring memory leaks, and optimizing code.
For example, Dahua and Hikvision develop software only for their cameras, making it easier for them. But universal software must work with any device. We re-evaluated every "building block" of our system and completely rewrote the program with a new team. In essence, we discarded everything old and created SmartVision from scratch.
Current Capabilities of SmartVision:
- Monitoring – Real-time video viewing with minimal PC load.
- Continuous Recording – Video recording in an open MP4 format with efficient compression to save disk space.
- Time-lapse Recording – Frame-by-frame recording helps preserve key events without excessive disk usage.
- Motion Detector Recording – Smart detection saves only relevant footage, optimizing storage use.
- Intelligent Video Analytics – Face and license plate recognition, object, fire, and smoke detection, AI-powered video analytics.
- System Stability – Automatic recovery after stream failures or connection loss.
- Cloud Integration – Enables remote access without complex configurations and allows push notifications for events.
Download SmartVision here:
https://smartvision.dev/cctv-software-download.htm
What’s Next?
We believe surveillance software should not just record video continuously but actively inform users about what is happening and where.
The mobile app should function like a live journal for each camera:
Camera 1 | 12:00 – A man (Person Name 1) with a dog entered Entrance No.1.
Camera 2 | 12:05 – A man (Person Name 1) with a dog entered the 23rd-floor hall.
Camera 3 | 12:06 – Toyota Car, license plate X777YY11, entered the parking.
Camera 4 | 12:07 – Room 1. Lights turned on, and a loud noise was detected.
Each event should be searchable, filterable, and linked to video playback, with automatic transcription of conversations in rooms. This functionality is currently in development.
We are looking forward to receiving your feedback and criticism on the new version. Thank you!