Rockchip on Toothbrush's Blog

My Smart Assistant 3.0: Building a Fully Automated "Time Screening Room"

Sat, 12 Jul 2025 17:10:27 +0800

Welcome back to my “lazy” photography system series. In the previous chapters, we witnessed how an ordinary webcam was endowed with the ability to “think,” and how I built an elegant and convenient “control room” for it. However, the story doesn’t end there. Our clever photographer has been smartly “slacking off” every day, accurately capturing a vast number of valuable moments. These photos, like scattered pearls, contain the secrets of time, quietly resting deep within the server’s hard drive.

It’s time to string these pearls together, letting them shine brilliantly through the flow of light and shadow. Today, I will show you the final chapter of this project, which I personally find the most exciting — building a fully automated “Time Screening Room.” At the core of this screening room is a powerful and reliable shell script. It is the “behind-the-scenes director” of my entire system, entrusted with a sacred mission: before each quiet dawn arrives, automatically merging tens of thousands of photos from the previous day into wonderful time-lapse videos.

Project code: https://github.com/AndroidOL/camera-capture/tree/main/merge

From “Material Pileup” to “Visual Poems”: The Final “Billion” Step

Imagine what life would be like without this automated “director.” It would mean that every day, I’d have to act like a “miner” in the digital age — manually logging into a remote server to dig through tens of thousands of photos from the previous day, which might amount to dozens of gigabytes; then carefully downloading these “raw ores” to my local computer; next, launching complex video editing software, painstakingly dragging each photo onto the timeline, setting frame rates, encoders, bitrates… and finally clicking “render” after a long wait.

This entire process is not only a battle of patience and endurance with no creativity but also prone to errors. If one day I forget or get too busy at work and miss it, the “visual diary” for that day will have an irreplaceable gap. That’s not what I want.

What I need is a fully unattended, extremely stable and reliable, and smart enough automated process. It should be a tireless artist that quietly starts at a fixed time every day, for example, in the dead of night, and like a skilled film editor, carefully arranges the scattered materials of the previous day into a complete film, then elegantly places it on the digital shelf of the “family film library,” waiting for my review.

Thus, this labor of love — the automation script daily_photo_to_video.sh — was born. It’s far from a simple ffmpeg command wrapper; it is an industrial-grade robust “smart production line.”

The Director’s Extraordinary Skills: Revealing the Core Design of the Automation Script

The design philosophy of this script is “pursue ultimate stability and efficiency.” I armed it to the teeth so that it can calmly handle all foreseeable and even unforeseeable complex situations.

1. Ultimate Efficiency: Squeezing Every Drop of Hardware Potential

Highway — Ramdisk (In-Memory Disk)
Reading tens of thousands of scattered small images on a traditional hard drive while writing a huge video file immediately exposes disk I/O bottlenecks, making the whole process painfully slow. My solution is radical: directly carve out a high-speed virtual disk (Ramdisk) in the server’s precious memory.

When the script starts, it first synchronizes all photos to be processed from the photo server to this memory disk via high-speed network. Then, all intensive read-write operations of video synthesis (encoding) are performed entirely on this virtual disk. Memory read/write speeds are tens or even hundreds of times faster than ordinary hard drives, which is like building a “star highway” that ignores physical laws and travels at light speed for data migration, greatly shortening processing time.

Professional Engine — Hardware Encoding (HEVC_RKMPP)
Video encoding is a typical CPU-intensive task. Relying solely on the CPU to “hard compute” is like asking a versatile but tired butler to do a highly specialized woodworking job — inefficient and causing the server’s “heartbeat” (CPU usage) to spike and the fans to roar. Fortunately, the server I use (based on Rockchip chips) has a dedicated video processing unit (VPU).

My script precisely calls ffmpeg and specifies the use of the hevc_rkmpp hardware encoder. This is equivalent to handing the professional work to a specially hired, highly skilled “film editor,” letting the VPU handle the heaviest encoding tasks, fully freeing the CPU. Meanwhile, I choose the efficient HEVC (H.265) encoding format, which maintains very high picture quality while compressing the video size to about half of traditional H.264, saving massive precious space in my “family film library.”

2. Foundation of Stability: Industrial-Grade Fault Tolerance and Self-Healing

A spaceship sailing alone in the wilderness values not only speed but also its ability to survive cosmic storms. Likewise, a script that needs to run unattended for years requires strong resilience. I built four layers of “security protocols” to handle various possible accidents.

Scenario	Script’s Smart Response Strategy
Unexpected script interruption	Uses `trap` cleanup mechanism to automatically clear temp files and unmount Ramdisk on interruption
Concurrent run conflicts	Uses `flock` file lock to avoid conflicts from multiple instances running concurrently
Insufficient storage space	Pre-checks available space in target path to avoid write failures midway
Target disk disconnection	Automatically switches to backup storage path, ensures task completion, and pauses old task catch-up

3. Smart Scheduling: Never Miss a Day’s Highlights

I want every day’s video in my “screening room” to be complete, like a coherent chronicle. Therefore, the smartest and proudest part of the script is its almost obsessive “completion” logic.

When running, the script:

Reviews logs from the past 7 days, checking for missing or failed videos;
If omissions are found, it prioritizes filling gaps starting from the earliest day;
Verifies that remote materials still exist to avoid pointless runs;
Only after confirming completeness does it process “yesterday’s” new materials.

This way, the system gains memory and responsibility, always maintaining a complete “visual diary.”

The Birth of a “Family Epic”

Now, this has become a ritual-filled daily routine. Every morning, when I brew coffee and open the “family film library,” a brand-new time-lapse video named after yesterday’s date quietly rests there.

It might capture the entire thunderstorm outside the window, or my figure working late into the night, or simply portray the traces of sunlight flowing through the home.

This automated script is like a silent but powerful heart, injecting endless vitality into my entire smart photography project. It transforms day-after-day trivial visual data into distilled visual poems full of time’s essence, turning cold data into warm memories, allowing me to truly “see” and “cherish” those silently passing moments.

The Joy of Creation

The joy of creation lies here — using code and logic to build a system that automatically creates, organizes, and presents beauty for us.

Next steps might be training an AI to automatically add the most fitting background music to these videos, or editing “weekly highlights.”

The journey to a “better life” through creation is endless, and that is its true charm.

My Smart Assistant: A Time-Lapse System That Knows When to Be Lazy

Sat, 12 Jul 2025 15:38:51 +0800

There’s a quiet poetry flowing through everyday life, often unnoticed. I became fascinated with time-lapse photography because it compresses long stretches of time into captivating moments, letting us witness slow transformations with the naked eye—it’s like magic unfolding through time.

So I dusted off an old USB webcam that had been asleep in a drawer for years, sparked by a gentle thought: I want to record life’s subtle breath—the kind we often overlook in our hurried pace. Like the succulent on the balcony slowly stretching toward sunlight; the drifting clouds outside, sculpted into shapes by an invisible hand; or how the evening sun casts long, warm shadows across the living room floor inch by inch…

Project Code: https://github.com/AndroidOL/camera-capture

From “Diligent Fool” to “Smart Slacker”: A Shift in Philosophy

At first, my approach was simple and direct, like many others: write a script to turn the webcam into a tireless “model worker,” snapping a photo every minute on the dot. The moment I finished and successfully ran the code, I felt a wave of achievement. But that joy quickly turned into mixed disappointment the next morning.

There they were—1,440 neatly organized photos, a testament to my program’s “diligence.” Yet scrolling through them felt lifeless and frozen. Especially during the long hours of the night—say, between 3:00 and 3:01 AM—everything looked identical: the dust on the coffee table hadn’t budged, and even the air seemed still. I realized I hadn’t built a tool for recording “change,” but a factory for producing repetition.

Once I calmly calculated this “digital legacy,” the scale of the problem became clear:

Alarming Storage Consumption

Daily output: 43,200 photos
Monthly output: ~1,296,000 photos
Annual output: Over 473 million photos

Assuming 0.2MB per 1920×1080 photo, that’s nearly 8GB per day—reasonable for a desktop setup, but excessive for embedded devices like the ESP32 or RK boards.

Destructive Post-Production Workload

Footage screening: To produce a few minutes of highlights, you’d have to sift through hundreds of thousands of nearly identical “waste shots.”
Time cost: This isn’t artistic creation—it’s soul-draining manual labor.

To better illustrate the contrast, I created this table:

Feature	“Diligent Fool” Mode	“Smart Slacker” Mode
Operating Principle	Uniform, scheduled capturing	Intelligent, on-demand capture
Photos per hour	Constant 1,800	Dozens to hundreds (based on changes)
Storage efficiency	Very low, linear explosion	Very high, only saves useful data
Post-workload	Huge, painful screening	Minimal, most shots are valuable
Final output quality	Filled with static “waste time”	Tight pacing, focused on change

Clearly, the latter is what I needed—a wise system that knows when to act and when to chill.

Giving the Camera a Brain: Laziness as Productivity

So I set out to upgrade my system with real intelligence. This time, I didn’t teach it to work harder—I taught it a more elegant skill: how to slack off smartly.

Its core principle is like a seasoned intelligence officer who only reports when something actually matters—no boring “all clear” updates, just action when there’s movement.

To empower this discernment, I designed a three-step “smart pipeline”:

Visual Purification: Seeing the Essence, Not the Surface

Morning cool light and evening warm light may color the scene differently, even if nothing has moved. So before comparison, the system converts both current and archived frames to grayscale “sketch mode.”

Quantifying Change: Giving Weight to Differences

The system overlays two grayscale images pixel by pixel, generating a “change map.” It calculates the percentage of light areas to assess the degree of difference.

The Art of Decision-Making: A Smart Laziness Threshold

I set a threshold—say 0.5%. If the detected change is below this, the frame is skipped. Only significant changes are recorded, saving resources and capturing only the meaningful.

A Thoughtful Housekeeper: A System That Runs Itself

To ensure 24/7 stability, I implemented multiple layers of resilience:

A Custom “Biological Clock”

Sampling frequencies tuned to household rhythms:

Morning glow (6–8 AM): every 5 seconds
Daytime activity (8 AM–6 PM): every 2 seconds
Evening mood (6–10 PM): every 5 seconds
Midnight patrol (10 PM–6 AM): every 10 seconds

Strategic Storage Management

When storage usage hits 85%, the system auto-deletes the oldest photos to maintain space.

Survival Instincts

In case of power loss or camera disconnects, the system includes auto-reconnect and retry logic to self-heal and stay online.

Unexpected Rewards: Becoming a Quiet Observer of Life

The results far exceeded my expectations. It allowed me to observe:

A Miniature Home Sundial

By tracking sunlight shifts on the floor, it became a massive clock of shadow and light.

The Silent Dance of Plants

In time-lapse, my succulent turned toward the sun like it was engaging in a dialogue with life.

The Moving Landscape Outside

It captured drifting clouds, droplets melting from the roof after snow, everything like a living poem.

The Joy of Creation Lies in the Process

From the initial spark to a mature system, this 0-to-1 journey gave me indescribable satisfaction.

The beauty of technology isn’t in complexity, but in how we use it to understand and embrace life.

If you also feel passion for creation, start with a tiny idea. The real barrier is never the tools—it’s whether you’re willing to tinker, to explore.

Begin your journey now. Go capture your own, one-of-a-kind story in time.